A method and device for monitoring and warning of spontaneous combustion risk of a new energy vehicle

Abstract

The application discloses a new energy vehicle self-ignition risk monitoring and early warning method and device, and belongs to the technical field of charging safety monitoring. The method obtains the visible spectrum image, infrared thermal image temperature data and charging pile communication data of the target vehicle in the charging scene; only vehicle existence detection is performed, and the battery monitoring area is adaptively determined according to the vehicle chassis battery position knowledge base, and the final monitoring position is located in the area by sliding window hot spot tracking. The method adaptively adjusts the weights of the temperature rise value and the temperature rise rate in combination with the protection layer attenuation coefficient, converts the multi-modal features into basic credibility distribution, and performs hierarchical determination after fusion by Dempster-Shafer evidence theory, and actively cuts off the charging power supply in the premonitory stage of thermal runaway. The application accesses the charging pile communication data to distinguish normal fast charging heat and abnormal temperature rise, which is helpful to inhibit false alarms in the fast charging scene; the feature gas sensor is an optional configuration, and the system can automatically operate in degraded mode when the configuration is not performed.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle charging safety monitoring technology, and more specifically, to a method and device for monitoring and early warning of spontaneous combustion risk of new energy vehicles based on multimodal perception information fusion and charging pile communication data. Background Technology

[0002] Lithium-ion power batteries may enter a state of thermal runaway under conditions of overcharging, short circuit, or aging. From the onset of abnormal temperature rise inside the cell to the eventual open flame, the process can sequentially involve stages such as the membrane melting and exacerbation of internal short circuits, electrolyte decomposition and gas generation, opening of the pressure relief valve, smoke emission, and a rapid rise in surface temperature. The time window from gas generation and smoke emission to open flame is often only tens of seconds or even shorter, and the monitoring system should provide early warnings during the abnormal gas generation and temperature rise stages.

[0003] Existing fire monitoring solutions for electric vehicle charging processes suffer from several common problems. Firstly, the monitoring area is inaccurately located. Battery packs are typically laid flat under the chassis and encased in a metal shell; however, the location and shell thickness vary significantly across different vehicle models. Heat conduction from the shell is delayed, with the surface temperature rising much slower than the core temperature of the internal battery cells. If a uniform temperature threshold is used for different models, thick-shell models are prone to missed detections, while thin-shell models may experience false alarms due to normal heat generation during fast charging.

[0004] Secondly, there is a lack of communication data support from charging piles. During DC fast charging, the current can reach over 200 amps, and a normal temperature rise of several degrees Celsius on the surface of the battery pack casing is a physical necessity. External thermal imagers and gas sensors can only detect physical quantities on the casing surface and cannot obtain charging parameters such as the vehicle-side current to determine the cause of the heat generation. This makes it easy for normal fast charging temperature rise to be misjudged as a precursor to thermal runaway.

[0005] Secondly, the single-sensor, fixed-threshold approach is unreliable. Environmental interferences such as changes in sunlight, nearby heat sources, dust, and water mist can all lead to false alarms, and fixed thresholds cannot fully utilize early warning features such as the rate of temperature rise. In addition, most solutions stop after only outputting an alarm signal, failing to form a closed-loop response of "monitoring-early warning-active power cut-off-fire suppression linkage." Waiting until thermal runaway develops into open flames before intervening significantly increases the difficulty and cost of handling the situation. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and device for monitoring and warning of spontaneous combustion risk of new energy vehicles, which is caused by the lack of charging pile communication data access, failure to distinguish between battery location and casing thermal delay characteristics, fixed threshold judgment by a single sensor, lack of active processing closed loop, and lack of system self-test function.

[0007] To address the aforementioned technical problems, this invention provides a method for monitoring and warning of spontaneous combustion risks in new energy vehicles. First, multi-source sensing data of the target vehicle is acquired in a charging scenario, including visible spectral images, infrared thermal imaging temperature data, and charging pile communication data. The visible spectral images are acquired by a visible light camera deployed above the charging position; the infrared thermal imaging temperature data is acquired by an infrared thermal imaging camera; and the charging pile communication data is obtained through the data transmission interface in the charging pile-vehicle charging communication protocol. This charging pile communication data includes at least one of the following: charging current value, charging voltage value, battery state of charge, and fault code information from the charging pile or the vehicle. The multi-source sensing data also includes characteristic gas concentration data, which is acquired by a gas sensor deployed near the charging position. The characteristic gas includes at least one of carbon monoxide, electrolyte volatile organic compounds, and hydrogen.

[0008] Secondly, vehicle presence detection is performed on the visible spectrum image; this detection only determines whether a vehicle target exists in the image, without identifying the brand, model, or year. A single binary classification target detection model processes the input image, outputting the confidence score and bounding box coordinates of candidate targets. After detecting a vehicle target, a vehicle chassis battery location knowledge base is pre-built to adaptively determine the vehicle chassis battery monitoring area template. The knowledge base pre-stores the normalized coordinate rectangle range of various vehicle chassis battery areas relative to the vehicle detection bounding box, as well as the attenuation coefficient of the protective layer reflecting the thermal conductivity characteristics of the battery pack casing. After obtaining the template, hotspot tracking is performed within the template area using a sliding window: a fixed-size sliding window traverses the template area, calculates the average temperature of each window, and selects the window with the highest average temperature as the final battery monitoring area. For unknown target vehicles with a confidence score below the threshold or no matching record in the knowledge base, an unknown target backtracking strategy is adopted, scanning several candidate battery areas under the vehicle body and selecting the one with the highest temperature as the monitoring area.

[0009] Then, based on the determined battery monitoring area, the temperature rise value and rate of temperature rise relative to the ambient temperature are extracted. The rate of temperature rise is obtained by calculating the slope of the average temperature of the battery monitoring area over the most recent N consecutive frames using a sliding window linear regression. Simultaneously, the charging pile communication data is parsed to identify fault code states and charging parameters. Combining the following multi-dimensional feature information, basic confidence assignments are generated for each sensing modality, based on the hypothesis set consisting of normal state N, thermal runaway precursor P, open flame F, and uncertain state θ: In thermal imaging mode, the weights of temperature rise value and temperature rise rate are adaptively adjusted based on the attenuation coefficient of the protective layer. When the attenuation coefficient of the protective layer is large, the response of the shell surface temperature to the internal cell temperature change is lagging. In this case, the weight of the temperature rise rate is increased and the weight of the absolute temperature rise value is decreased, so that the judgment depends more on the temperature change trend rather than the instantaneous absolute temperature value.

[0010] The gas mode concentration increments are calculated using the frozen baseline method. During the initial preheating phase, gas concentration samples are collected to establish a baseline. After preheating, the baseline is frozen, and subsequent analyses are based on increments relative to this frozen baseline. When the concentration increment of any of the following—carbon monoxide, electrolyte vapor, or hydrogen—exceeds a preset threshold, a confidence assignment is generated primarily pointing to precursors of thermal runaway. When none of the gas concentration increments exceed the threshold, the gas mode confidence quality is entirely assigned to the uncertain state hypothesis. If no gas sensors are configured, the gas mode confidence quality is by default entirely assigned to the uncertain state hypothesis.

[0011] The visible light mode is used to detect smoke and flame in visible spectrum images. Background subtraction is used to identify abnormal motion regions relative to the background, and color features and inter-frame flicker features are used to distinguish smoke and flame regions. During the detection, regions that have been identified as vehicle targets by vehicle presence detection are excluded to avoid misjudging normal vehicles as combustion products.

[0012] The charging pile communication mode analyzes communication data. When the charging pile reports one of the fault codes such as overvoltage, overcurrent, insulation, overtemperature, or cell abnormality, a strong confidence assignment pointing to the precursor of thermal runaway is generated; when it reports that it is in a normal high-current charging state and there is no fault code, the confidence quality is mainly allocated to the uncertain state hypothesis and a small amount is allocated to the normal state hypothesis; when the communication link is interrupted or data cannot be obtained, the confidence quality of this mode is entirely allocated to the uncertain state hypothesis.

[0013] When allocating basic credibility for each modality, if no abnormal evidence is detected for a certain modality, its credibility quality is entirely allocated to the uncertain state hypothesis, rather than to the normal state hypothesis.

[0014] The charging pile communication data is treated as an independent sensing mode alongside the thermal imaging mode and the visible light mode. Basic confidence assignment is directly generated at this mode level. By using charging current, voltage, state of charge and fault code information, the temperature rise detected by the thermal imaging mode is "explained as a cause". When the communication data indicates that it is under normal high-current charging and there are no fault codes, this mode assigns most of the confidence to the uncertain state hypothesis, thereby suppressing the misjudgment of normal fast charging heat as a precursor to thermal runaway.

[0015] After the basic confidence levels for each modality are generated, the Dempster-Shafer evidence theory is used to combine them to obtain the confidence values ​​of each hypothesis after fusion. During fusion, the total conflict coefficient K between each source of evidence is first calculated; when K is below a preset threshold, orthogonal summation rule fusion is used; when K reaches or exceeds the preset threshold, the weighted voting backup rule is switched to, that is, each modality independently selects the hypothesis with the highest confidence level as the voting result, and the hypothesis with the highest weighted vote is used as the fusion result. A hysteresis interval is set near the threshold to avoid frequent switching of fusion rules at the boundary.

[0016] After integration, risk assessment is performed based on credibility, classifying vehicle charging safety status into four levels: Normal, Attention, Warning, and Alarm. An open flame hypothesis reaching or exceeding the first threshold is classified as an Alarm, triggering automatic fire extinguishing and fire suppression linkage. A precursor hypothesis reaching or exceeding the second threshold, or the sum of the precursor and open flame hypotheses reaching or exceeding the third threshold, is classified as a Warning, actively cutting off charging power. A precursor hypothesis reaching or exceeding the fourth threshold, or the sum of the precursor and open flame hypotheses reaching or exceeding the fifth threshold, is classified as an Attention level. All other levels are classified as Normal. A hysteresis mechanism is introduced: a preset number of consecutive frame upgrade conditions must be met before a level is upgraded, and a preset number of consecutive frame downgrade conditions must be met before a level is downgraded.

[0017] Furthermore, the above method may also include a system health self-check step. Continuous self-check during system operation: For each monitored charging location, the charging status data reported by the charging pile is periodically queried and cross-checked with the vehicle presence detection results from the camera at that charging location. When the charging pile continuously reports that it is charging, but the corresponding camera fails to detect a vehicle in multiple consecutive frames, the camera at that charging location is deemed abnormal and an alarm is triggered. The system can adaptively adjust the threshold for the number of consecutive undetected frames used to determine abnormality based on ambient light intensity, appropriately relaxing it in low light or at night.

[0018] The above method can be summarized in the following steps: S1: Acquire multi-source perception data of the target vehicle in the charging scenario; S2: Perform vehicle presence detection on the visible spectrum image. If the vehicle is present, output the target confidence score and bounding box coordinates. S3: Determine if a vehicle target exists. If yes, proceed to S4; otherwise, return to S1. S4: Query the normalized coordinate rectangle and the protective layer attenuation coefficient based on the knowledge base; S5: Determine if the knowledge base is successfully matched. If yes, proceed to S6; otherwise, proceed to S7. S6: Determine the battery monitoring area template according to the knowledge base template; S7: Employ an unknown target back-off strategy, scan the candidate areas under the vehicle body and select the area with the highest temperature as the monitoring area; S8: Perform sliding window hotspot tracking within the monitoring area template to determine the final battery monitoring area; S9: Extract multimodal features, including temperature rise value and rate of temperature rise in thermal imaging mode, smoke or flame features in visible light mode, fault codes and charging parameter analysis in charging pile mode, and concentration increment in gas mode. S10: Generate basic credibility assignments for each perceptual modality based on the hypothesis set; S11: Calculate the total conflict coefficient K among all sources of evidence; S12: Determine whether K is less than the preset threshold. If yes, proceed to S13a; otherwise, proceed to S13b. S13a: Employs standard Dempster-Shafer orthogonal and rule-based fusion; S13b: Switch to weighted voting fallback rule fusion; S14: Based on the credibility after fusion, risk is classified and determined. At the early warning level, the charging power is actively cut off, and at the alarm level, automatic fire extinguishing and fire-fighting linkage are further activated.

[0019] This invention also provides a spontaneous combustion risk monitoring and early warning device for new energy vehicles. The device includes: an image acquisition module, a thermal imaging acquisition module, a communication acquisition module, a vehicle detection and battery positioning module, a multimodal analysis module, an evidence fusion module, and a hierarchical judgment and linkage control module. The vehicle detection and battery positioning module is configured to: receive the visible spectrum image, call a preset binary classification target detection model to perform vehicle presence detection, and output target confidence and bounding box coordinates; determine a battery monitoring area template based on a pre-built vehicle chassis battery location knowledge base; and perform sliding window hotspot tracking within the battery monitoring area template, using the window with the highest average temperature as the battery monitoring area. All the above modules are deployed and run on edge computing nodes and communicate with various sensing and control devices through data interfaces.

[0020] Compared with existing technologies, this invention has at least the following advantages: by combining vehicle presence detection with a chassis battery location knowledge base and hot spot tracking, adaptive positioning of the battery monitoring area is achieved; by accessing charging pile communication data, it can distinguish between normal fast charging heat generation and abnormal temperature rise, reducing the false alarm rate in fast charging scenarios; by introducing a protective layer attenuation coefficient to adjust the weight allocation of thermal imaging modes, it alleviates the missed alarms caused by thermal hysteresis of the metal casing; by employing multimodal evidence theory fusion, it improves the system stability under single-channel failure or environmental interference; and by actively cutting off the charging power supply in the early stage of thermal runaway, it forms a closed loop of monitoring, early warning, and handling. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system architecture provided for an embodiment of the present invention.

[0022] Figure 2 This is a flowchart of a method provided in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The following embodiments are used to explain the present invention, but not to limit the scope of protection of the present invention.

[0024] Example 1: Overall Process and Hot Spot Tracking The system is deployed at the charging station. A visible light camera and an infrared thermal imaging camera are installed above the charging station to collect visible spectral images and infrared thermal temperature data, respectively. A communication module connects to the charging pile and acquires current, voltage, state of charge, and fault code data via the communication protocol interface between the charging pile and the vehicle. A characteristic gas sensor is optional and can be deployed near the charging station to collect data on carbon monoxide, electrolyte vapor, and hydrogen concentrations.

[0025] After detecting a vehicle target, the system queries the vehicle chassis battery location knowledge base based on the target bounding box coordinates. Each record in the knowledge base includes a vehicle type identifier, a normalized coordinate rectangle of the battery area, and a protective layer attenuation coefficient. If a matching record is found, the battery area is mapped to the current vehicle bounding box; if no match is found or the confidence level is low, it is marked as an unknown target, and the candidate area under the vehicle body is scanned, selecting the one with the highest temperature.

[0026] After determining the template, sliding window thermal spot tracking is performed. The sliding window size is approximately one-third of the template size, and the step size is half of the sliding window size. Each window is traversed within the template range, and the average temperature is calculated. The window with the highest average temperature is selected as the battery monitoring area. On common edge computing hardware, this tracking is performed on infrared thermal images with a resolution of 320×240, and the single-frame processing time meets the system's operational requirements.

[0027] Example 2: Distinguishing between normal fast charging heat generation and abnormal temperature rise using charging pile communication data During DC fast charging, the high current inevitably causes a normal temperature rise in the battery pack casing. Without access to the charging pile's communication data, relying solely on external temperature monitoring presents a dilemma regarding thresholds: a threshold that is too low may misjudge normal fast charging heat as abnormal, while a threshold that is too high may miss some abnormal events involving slow temperature increases.

[0028] In this embodiment, charging pile communication data is accessed as an independent sensing mode. If fault codes such as overvoltage, overcurrent, insulation, overtemperature, or individual cell abnormalities are present in the communication data, this mode generates a confidence allocation pointing to thermal runaway precursors. If the communication data shows normal high-current charging without fault codes, this mode allocates most of its confidence quality to the uncertain state hypothesis and only a small amount to the normal state hypothesis, thereby providing a "causal explanation" for the temperature rise detected by the thermal imaging mode and suppressing false alarms in fast charging scenarios. When the communication link is interrupted, all confidence quality of this mode is allocated to the uncertain state hypothesis.

[0029] Example 3: Adjustment of the attenuation coefficient and weight of the protective layer The attenuation coefficient of the protective layer reflects the degree of delay in temperature conduction from the battery pack's metal casing to the internal battery cells. This coefficient is determined through thermal conduction experiments: the battery pack of the target vehicle is heated at a constant rate, and the response time difference between the core temperature of the cells and the surface temperature of the casing is recorded. The attenuation coefficient is then calculated based on this time difference. The thicker the casing and the greater the thermal resistance, the greater the attenuation coefficient.

[0030] When assigning basic reliability for thermal imaging mode generation, the weights of temperature rise value and temperature rise rate are adaptively allocated based on this coefficient. When the attenuation coefficient is large, the response of the shell surface temperature to changes in internal temperature has a significant lag. In this case, the weight of the temperature rise rate is increased and the weight of the absolute temperature rise value is decreased, making the judgment more dependent on the temperature change trend rather than the instantaneous absolute value. When the attenuation coefficient is close to 1, the two are treated with equal weight.

[0031] Example 4: Integration of Evidence Theories and Conflict Resolution The fusion engine receives the basic confidence assignments for each modality and combines them using pairwise orthogonal sum rules. Before fusion, the total conflict coefficient between each source of evidence is calculated. When the conflict coefficient is below a preset value, the standard fusion rule is used; when the conflict coefficient reaches or exceeds the preset value, it switches to a weighted voting backup rule—each modality independently selects its highest-confidence hypothesis for voting, and the hypothesis with the highest weighted vote is used as the fusion result. A hysteresis interval is set near the threshold to avoid frequent rule switching.

[0032] Example 5: Graded Risk Assessment and Active Power-Off Control After fusion, the hypotheses are graded based on their credibility values. When the credibility of the open flame hypothesis reaches the first threshold, it is classified as an alarm level, triggering automatic fire extinguishing and fire suppression linkage. When the credibility of the precursor hypothesis reaches the second threshold, or when the combined credibility of the precursor and open flame hypotheses reaches the third threshold, it is classified as a warning level, triggering active power cut-off and audible and visual alarms. When the credibility of the precursor hypothesis reaches the fourth threshold, it is classified as a level of concern. All others are classified as normal. The specific thresholds for each level are determined during engineering implementation based on labeled data and on-site calibration.

[0033] The judgment results are processed by the hysteresis module. Upgrading is only performed if the upgrade conditions are met for several consecutive frames, and downgrading is only performed if the downgrading conditions are met for several consecutive frames. Upgrading requires fewer consecutive frames to ensure a fast response to real events, while downgrading requires more consecutive frames to avoid prematurely exiting the warning state.

[0034] When the warning level is reached, a power-off command is immediately sent to the charging station, simultaneously triggering the audible and visual alarms and alarm notifications. When the alarm level is reached, the automatic fire extinguishing device is activated and the fire protection system is activated in conjunction with the alarm system.

[0035] The system also features a health self-check function. It periodically queries the reported status of each charging station and cross-compares it with the vehicle presence detection results from the corresponding camera. If a charging station continuously reports that it is charging, but the corresponding camera does not detect a vehicle in multiple consecutive frames, the system determines that the sensing device at that charging station is malfunctioning and issues an alarm.

[0036] The above description is merely a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring and early warning of spontaneous combustion risk in new energy vehicles, characterized in that, Includes the following steps: Acquire multi-source perception data of the target vehicle in a charging scenario. The multi-source perception data includes visible spectrum images, infrared thermal imaging temperature data, and charging pile communication data. The charging pile communication data includes at least one of charging current value, charging voltage value, battery state of charge, and fault code information. Vehicle presence detection is performed on the visible spectrum image, which only determines whether a vehicle target exists in the image; based on a pre-built vehicle chassis battery location knowledge base, the vehicle chassis battery monitoring area template is adaptively determined, and hot spot tracking is performed in the template area using a sliding window method, with the window with the highest average temperature in the sliding window being used as the final battery monitoring area; Based on the temperature rise value and temperature rise rate of the battery monitoring area relative to the ambient temperature, combined with the smoke or flame features in the visible spectrum image and the analysis results of the charging pile communication data, basic confidence assignments for each sensing mode are generated for the hypothesis set consisting of normal state, thermal runaway precursor, open flame and uncertain state. The Dempster-Shafer evidence theory is used to combine the basic credibility assignments of each perceptual modality to obtain the credibility values ​​of each hypothesis after fusion. The risk is assessed based on the fused confidence level. When the assessment result corresponds to the risk level of thermal runaway precursor, the charging power of the vehicle is actively cut off.

2. The method according to claim 1, characterized in that, The vehicle chassis battery location knowledge base pre-stores the normalized coordinate rectangle range of various vehicle chassis battery areas relative to the vehicle detection enclosure, as well as the protective layer attenuation coefficient reflecting the thermal conductivity characteristics of the battery pack casing.

3. The method according to claim 2, characterized in that, When the detected vehicle target has no matching record in the knowledge base, or the detection confidence is lower than the preset threshold, an unknown target backoff strategy is adopted to determine the battery monitoring area, that is, to scan multiple candidate areas under the vehicle body and select the one with the highest temperature as the monitoring area.

4. The method according to claim 2, characterized in that, When generating the basic confidence level of the thermal imaging mode, the weights of the temperature rise value and the temperature rise rate are adaptively adjusted based on the attenuation coefficient of the protective layer. The larger the attenuation coefficient of the protective layer, the higher the weight of the temperature rise rate and the lower the weight of the absolute temperature rise.

5. The method according to claim 1, characterized in that, When generating the basic credibility allocation for the charging pile communication mode: if a fault code reported by the charging pile is parsed, a credibility allocation pointing to the precursor of thermal runaway is generated; if the charging pile reports that it is in a normal high-current charging state and there is no fault code, the credibility quality is mainly allocated to the uncertain state hypothesis, and a small amount of quality is allocated to the normal state hypothesis; if the charging pile communication link is interrupted or data cannot be obtained, all credibility quality is allocated to the uncertain state hypothesis.

6. The method according to claim 1, characterized in that, The multi-source sensing data also includes characteristic gas concentration data, wherein the characteristic gas includes at least one of carbon monoxide, electrolyte volatile organic compounds, and hydrogen; the basic confidence assignment of the gas modes is generated based on the concentration increment relative to the frozen baseline; when no gas sensor is configured, the full confidence quality of the gas modes is assigned by default to the uncertain state assumption.

7. The method according to claim 1, characterized in that, When no corresponding anomalous evidence is detected, the basic credibility quality of each perception modality is entirely assigned to the uncertain state hypothesis.

8. The method according to claim 1, characterized in that, In the process of fusing the Dempster-Shafer evidence theory, the conflict coefficient between the basic credibility assignments of each perceptual modality is calculated. When the conflict coefficient is lower than a preset threshold, standard orthogonal rule fusion is used. When the conflict coefficient reaches or exceeds the preset threshold, the system switches to weighted voting backup rule for fusion.

9. The method according to claim 1, characterized in that, The risk classification includes four levels: normal level, attention level, early warning level, and alarm level. The early warning level corresponds to the precursor of thermal runaway, triggering active power cut-off, audible and visual alarms, and alarm notifications. The alarm level corresponds to open flame, further triggering automatic fire extinguishing and fire-fighting linkage. The risk classification introduces a hysteresis mechanism, performing an upgrade only after a preset number of frames have been continuously met, and performing a downgrade only after a preset number of frames have been continuously met.

10. A spontaneous combustion risk monitoring and early warning device for new energy vehicles, characterized in that, include: Image acquisition module, used to acquire visible spectrum image data; The thermal imaging acquisition module is used to acquire infrared thermal image temperature data; The communication acquisition module is used to acquire communication data from the charging pile. The vehicle detection and battery positioning module is configured to: receive the visible spectrum image, call a preset binary target detection model to perform vehicle presence detection, and output target confidence and bounding box coordinates; determine a battery monitoring area template based on a pre-built vehicle chassis battery location knowledge base; and perform sliding window hotspot tracking within the battery monitoring area template, using the window with the highest average temperature as the battery monitoring area. The multimodal analysis module is used to generate basic credibility assignments for each sensing modality; The evidence fusion module is used to perform Dempster-Shafer evidence theory fusion; The risk classification and linkage control module is used to perform risk classification and active power-off control. The above modules are deployed and run on edge computing nodes, and communicate with various sensing and control devices through data interfaces.

Images