Intelligent environment sensing system based on multi-mode sensor

Abstract

The invention discloses an intelligent environment sensing system based on a multi-modal sensor, and relates to the technical field of intelligent environment sensing, the reliability of an underground garage environment sensing system is remarkably improved through multi-dimensional innovation, dynamic modal decomposition and a TCN network are combined, and the reliability of the system is improved. Differentiated characteristics of hot vehicles of the fuel vehicles and thermal runaway of the electric vehicles are effectively distinguished, and the misjudgment rate is reduced; a data attenuation mechanism triggered by vibration and an overlapping region linkage rule are introduced, and equipment abnormity and environment sudden change interference are resisted; and meanwhile, a sensing decision verification closed loop is formed by a dynamic learning module, so that the adaptability of the system to a new scene is continuously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental intelligent perception, and in particular to an environmental intelligent perception system based on a multimodal sensor. Background Art

[0002] With the rapid increase in the proportion of electric vehicles, traditional ventilation control solutions based on CO concentration have frequently failed. The mainstream system has been upgraded to a multimodal perception architecture that integrates three sets of devices: CO sensors, particulate matter detectors, and infrared thermal imaging devices. Through data fusion, it addresses the differentiated risk characteristics of the two types of vehicles.

[0003] The current solution still has defects in identifying sudden situations. When a fuel vehicle is warming up, the CO concentration climbs at a rate of 200 ppm / min, while the increase in particulate matter is minimal. In the early stages of thermal runaway in electric vehicles, PM2.5 levels will rise exponentially, but the CO value will remain stable. The existing spatiotemporal feature fusion algorithm has difficulty capturing these millisecond-level dynamic differences and often confuses the two scenarios. When multiple vehicles experience anomalies at the same time, the sensor data stream will produce coupling interference, causing the output of the LSTM-based prediction model to oscillate.

[0004] Some solutions use a federated learning framework to improve model generalization and adapt to the structural characteristics of different garages through distributed training; or introduce spiking neural networks (SNNs) to process asynchronous pulse signals from sensors. However, these solutions are constrained by the computing power of edge devices. When running SNNs, the memory usage of the gateway soars, resulting in increased response delays. Therefore, an environmental intelligent perception system based on multimodal sensors is urgently needed to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an environmental intelligent perception system based on multimodal sensors to solve the problems of time synchronization deviation and coupling interference in the existing system during multimodal data fusion, which leads to delays in the initial identification of thermal runaway of electric vehicles and frequent false triggering. The latest improvement solution is limited by computing power and cost.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: The embodiment of the present invention provides an environment intelligent perception system based on a multimodal sensor, which includes: Sensor group, including CO concentration sensor, particulate matter detector, infrared thermal imaging module and vibration sensor; Data processing module, with built-in timestamp alignment unit, dynamic mode decomposition algorithm unit and risk prediction unit; a ventilation control module, receiving an output signal from the data processing module; Among them, the timestamp alignment unit performs millisecond-level synchronization marking on the multimodal sensor data stream, the dynamic mode decomposition algorithm unit separates the combustion characteristic mode and the conventional emission mode from the mixed signal, and the risk prediction unit triggers the graded response instruction according to the modal energy ratio.

[0008] As a preferred solution of the multimodal sensor-based intelligent environmental perception system described in the present invention, the vibration sensor is integrated into the equipment mounting bracket. When the axial acceleration detected by the vibration sensor exceeds the preset safety value, the sensor data credibility attenuation mechanism is triggered, which is specifically manifested as follows: Apply an attenuation weight coefficient to the multimodal data from the current moment; When the accumulated attenuation time exceeds the set threshold, historical normal operating condition data is automatically called to replace the abnormal period input.

[0009] As a preferred solution of the multimodal sensor-based intelligent environmental perception system described in the present invention, the infrared thermal imaging module and the particle detector have a 50%-70% overlap in their spatial coverage areas. Within this overlap: When the number of thermal anomaly points exceeds the set value, the sampling frequency of the particulate matter detector is forced to increase to 3 times the baseline value; the sampling frequency increase strategy follows: base frequency × (1 + 0.05 current number of thermal anomaly points). When the current number of thermal anomaly points is greater than 20, the highest sampling frequency is locked; When the abnormal thermal distribution conforms to the battery module arrangement characteristics, the instantaneous peak data of the CO sensor is ignored.

[0010] As a preferred solution of the multimodal sensor-based environmental intelligent perception system described in the present invention, when the vibration sensor inside the equipment mounting bracket detects abnormal acceleration, the sensor group automatically triggers an attenuation mechanism and processes the multimodal data according to time-varying weights until it is replaced with historical normal operating condition data after a timeout. The credibility attenuation mechanism includes: When the real-time axial acceleration satisfy: , in, represents the unit step function, Indicates time The real-time axial acceleration, Represents the preset safety threshold. This function is used to determine the triggering moment. Trigger time , in, Indicates taking the earliest moment that satisfies the conditions; Since the trigger moment, the attenuation weight is assigned to the multimodal data at each moment: , in, Indicates time The data credibility weight, represents the natural exponential function, Represents the decay rate, which is used to control the weight decay speed; The decay rate By half-life parameter Sure: , in, represents the natural logarithm operation, Indicates the time required for the weight to decay from 1 to 0.5; To determine whether to call historical replacement, define the cumulative triggering time: , in, Indicates from To the current moment The cumulative activation time, the integral operation represents the cumulative time, Represents the time variable in the integration process, used to accumulate the effective time since the trigger; when , then for subsequent data streams Directly call historical normal operating condition data: , in, Indicates the cumulative attenuation time threshold. Indicates the data output after replacement, Represents the multimodal data segment after mapping the historical working conditions to the current moment, It is the mapping moment, used to align the data length.

[0011] As a preferred solution of the multimodal sensor-based environment intelligent perception system of the present invention, the dynamic modal decomposition algorithm unit is configured as follows: When the absolute value of the CO concentration change rate is greater than the first threshold and the particulate matter number change rate is less than the second threshold, the fuel vehicle hot vehicle discrimination mode is activated; When the rate of change of the number of particulate matter exceeds the third threshold and the hysteresis time difference of the CO concentration change is greater than the set time, the electric vehicle thermal runaway warning mode is activated.

[0012] As a preferred solution of the multimodal sensor-based intelligent environmental perception system of the present invention, the risk prediction unit includes a temporal convolutional network module, the input layer of which receives the following features: Time domain differential of CO concentration and particulate matter number; Moving trajectory characteristics of the highest temperature point in the infrared thermal imaging area; The abnormal vibration spectrum of the equipment installation platform captured by the vibration sensor. The abnormal vibration spectrum characteristics include the alarm characteristics generated when the vibration sensor's energy proportion in the 5-15Hz frequency band exceeds 50% of the steady-state operating value; The output layer generates a risk level signal, which includes three status codes: normal, warning, and emergency.

[0013] As a preferred solution of the multimodal sensor-based environmental intelligent perception system described in the present invention, in which: in the dynamic mode decomposition algorithm unit, three thresholds are adaptively set based on historical normal operating condition statistics, and mode activation logic is constructed accordingly. The process includes: Assume that the first-order time-domain differential series of CO concentration and particulate matter quantity under historical normal operating conditions are and ,definition: , in, Indicates the first threshold, subscript 1 indicates the corresponding fuel vehicle hot vehicle discrimination mode, represents the statistical mean of the time derivative of CO concentration under historical normal operating conditions, where · represents the time derivative and the upper horizontal bar represents the statistical mean. represents the standard deviation of the first-order time-domain differential of CO concentration under historical normal operating conditions, where is a statistical operator, Indicates the safety factor, used to adjust the alarm trigger sensitivity; , in, represents the second threshold, and the subscript 2 represents the corresponding particle stability fluctuation judgment. It represents the statistical mean of the time derivative of the number of particulate matter under historical normal working conditions, It represents the standard deviation of the first-order time-domain differential of the number of particulate matter under historical normal working conditions. Represents the sensitivity coefficient, which is used to control the tolerance to small fluctuations; , in, Indicates the third threshold, subscript 3 indicates the corresponding electric vehicle thermal runaway warning judgment, Indicates the alarm coefficient, which is used to control the alarm intensity for large fluctuations; Define the unit step function: , in, represents the unit step function, Represents function input variables; Define the mode activation function: , in, Indicates the activation signal of the hot vehicle identification mode for fuel vehicles. Represents the first-order time-domain differential of the real-time CO concentration with respect to time, It represents the first-order time-domain differential of the real-time particle number with respect to time, and the symbol |.| represents the absolute value operation. , in, Indicates the activation signal of the electric vehicle thermal runaway warning mode. represents the time lag between the change in particle number and the response of CO concentration, where Δ represents the difference operation, Indicates the preset hysteresis threshold.

[0014] As a preferred solution of the multimodal sensor-based intelligent environmental perception system of the present invention, the ventilation control module implements a differentiated response strategy: When receiving an early warning signal, the system starts the low-speed ventilation mode and activates cross-verification of sensors in adjacent areas; When an emergency signal is received, full-power exhaust is forced to start and a linkage command is sent to the fire protection system.

[0015] As a preferred solution of the multimodal sensor-based intelligent environmental perception system of the present invention, the differentiated response strategy executed by the ventilation control module further includes: When the conflict status code is received three times in succession, the device starts the self-test procedure and switches to the backup sensor group.

[0016] As a preferred solution of the multimodal sensor-based intelligent environmental perception system described in the present invention, a dynamic learning module is further configured, and its working logic is as follows: After each ventilation control action was completed, a 20-second multimodal data segment before and after the event was extracted; When the actual fire alarm is inconsistent with the system warning result, a feature vector difference report is automatically generated; Adjust the mode cutoff order selection strategy in the dynamic mode decomposition algorithm based on the difference report.

[0017] The beneficial effects of the present invention are as follows: the present invention significantly improves the reliability of the underground garage environment perception system through multi-dimensional innovation, adopts the combination of dynamic mode decomposition and TCN network to effectively distinguish the differentiated characteristics of fuel vehicle hot running and electric vehicle thermal runaway, and reduce the misjudgment rate; introduces the vibration-triggered data attenuation mechanism and the overlapping area linkage rule to resist equipment anomalies and environmental mutation interference; at the same time, forms a perception decision verification closed loop with a dynamic learning module to continuously improve the system's adaptability to new scenarios; in addition, the parameterized design of half-life, statistical threshold generation and other schemes reduce the difficulty of deployment and debugging. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a schematic diagram of the framework of an environmental intelligent perception system based on multimodal sensors in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0023] Example 1, with reference to Figure 1 This embodiment provides an intelligent environment perception system based on a multimodal sensor, including: Sensor group, including CO concentration sensor, particulate matter detector, infrared thermal imaging module and vibration sensor; The vibration sensor is integrated into the equipment mounting bracket. When the axial acceleration detected by the sensor exceeds the preset safety value, the sensor data credibility attenuation mechanism is triggered. The specific performance is as follows: Apply an attenuation weight coefficient to the multimodal data from the current moment; When the accumulated decay time exceeds the set threshold, historical normal operating condition data will be automatically called to replace the abnormal period input; There is a 50%-70% overlap between the spatial coverage areas of the infrared thermal imaging module and the particle detector. Within this overlap area: When the number of thermal anomaly points exceeds the set value, the sampling frequency of the particulate matter detector is forced to increase to 3 times the baseline value; the sampling frequency increase strategy follows: base frequency × (1 + 0.05 current number of thermal anomaly points). When the current number of thermal anomaly points is greater than 20, the highest sampling frequency is locked; When the abnormal thermal distribution conforms to the battery module arrangement characteristics, the instantaneous peak data of the CO sensor is ignored; In the sensor group, when the vibration sensor inside the equipment mounting bracket detects abnormal acceleration, the attenuation mechanism is automatically triggered and multimodal data is processed according to time-varying weights until it is replaced with historical normal operating condition data after a timeout. The credibility attenuation mechanism includes: When the real-time axial acceleration satisfy: , in, represents the unit step function, Indicates time The real-time axial acceleration, Represents the preset safety threshold. This function is used to determine the triggering moment. Trigger time , in, Indicates taking the earliest moment that satisfies the conditions; Since the trigger moment, the attenuation weight is assigned to the multimodal data at each moment: , in, Indicates time The data credibility weight, represents the natural exponential function, Represents the decay rate, which is used to control the weight decay speed; The decay rate By half-life parameter Sure: , in, represents the natural logarithm operation, Indicates the time required for the weight to decay from 1 to 0.5; To determine whether to call historical replacement, define the cumulative triggering time: , in, Indicates from To the current moment The cumulative activation time, the integral operation represents the cumulative time, Represents the time variable in the integration process, used to accumulate the effective time since the trigger; when , then for subsequent data streams Directly call historical normal operating condition data: , in, Indicates the cumulative attenuation time threshold. Indicates the data output after replacement, Represents the multimodal data segment after mapping the historical working conditions to the current moment, is the mapping moment, used to align data length, using periodic mapping or sliding window alignment strategy; Specifically, this mechanism uses vibration anomaly as a trigger condition, uses activation function to accurately identify the first abnormal moment, and uses exponential decay weight to smoothly transition to avoid sudden changes from impacting downstream algorithms. The half-life parameterized design takes into account both sensitivity and robustness, and can be adjusted according to equipment characteristics during engineering deployment. ,The cumulative duration judgment ensures that short-term bumps will not cause complete data invalidation, and historical replacement is called only when the anomaly persists, thereby ensuring data availability while reducing the risk of misusing historical data; Data processing module, with built-in timestamp alignment unit, dynamic mode decomposition algorithm unit and risk prediction unit; The dynamic mode decomposition algorithm unit is configured as follows: When the absolute value of the CO concentration change rate is greater than the first threshold and the particulate matter number change rate is less than the second threshold, the fuel vehicle hot vehicle discrimination mode is activated; When the rate of change of the number of particulate matter exceeds the third threshold and the hysteresis time difference of the CO concentration change is greater than the set time, the electric vehicle thermal runaway warning mode is activated; The risk prediction unit contains a temporal convolutional network module, whose input layer receives the following features: Time domain differential of CO concentration and particulate matter number; Moving trajectory characteristics of the highest temperature point in the infrared thermal imaging area; The abnormal vibration spectrum of the equipment installation platform captured by the vibration sensor. The abnormal vibration spectrum characteristics include the alarm characteristics generated when the vibration sensor's energy proportion in the 5-15Hz frequency band exceeds 50% of the steady-state operating value; The output layer generates risk level signals, which include three status codes: normal, warning, and emergency. In the dynamic modal decomposition algorithm unit, three thresholds are adaptively set based on historical normal operating condition statistics, and the mode activation logic is constructed accordingly. The process includes: Assume that the first-order time-domain differential series of CO concentration and particulate matter quantity under historical normal operating conditions are and ,definition: , in, Indicates the first threshold, subscript 1 indicates the corresponding fuel vehicle hot vehicle discrimination mode, represents the statistical mean of the time derivative (first-order time domain differential) of CO concentration under historical normal operating conditions, where · represents the time derivative and the upper horizontal bar represents the statistical mean. represents the standard deviation of the first-order time-domain differential of CO concentration under historical normal operating conditions, where is a statistical operator, Indicates the safety factor, used to adjust the alarm trigger sensitivity; , in, represents the second threshold, and the subscript 2 represents the corresponding particle stability fluctuation judgment. It represents the statistical mean of the time derivative (first-order time domain differential) of the number of particulate matter under historical normal working conditions. It represents the standard deviation of the first-order time-domain differential of the number of particulate matter under historical normal working conditions. Represents the sensitivity coefficient, which is used to control the tolerance to small fluctuations; , in, Indicates the third threshold, subscript 3 indicates the corresponding electric vehicle thermal runaway warning judgment, Indicates the alarm coefficient, which is used to control the alarm intensity for large fluctuations; Define the unit step function: , in, represents the unit step function, Represents function input variables; Define the mode activation function: , in, Indicates the activation signal of the hot vehicle identification mode for fuel vehicles. Represents the first-order time-domain differential of the real-time CO concentration with respect to time, It represents the first-order time-domain differential of the real-time particle number with respect to time, and the symbol |.| represents the absolute value operation. , in, Indicates the activation signal of the electric vehicle thermal runaway warning mode. represents the time lag between the change in particle number and the response of CO concentration, where Δ represents the difference operation, Indicates the preset hysteresis threshold; Specifically, the threshold is automatically generated by historical statistics, avoiding manual parameter adjustment based on experience. In the judgment of hot engine of fuel vehicle, the first threshold Combined safety factor Ensure timely response when CO concentration rises sharply, and the second threshold Combined sensitivity coefficient It suppresses misjudgment caused by small fluctuations in particulate matter. For electric vehicle thermal runaway warning, the third threshold Sensitive detection of sudden increase of particulate matter, incremental hysteresis With preset time delay Jointly construct dual criteria to effectively filter conventional emission delays and unit step functions Provides clear boundaries for logical judgment; A ventilation control module receives an output signal from the data processing module; The ventilation control module implements differentiated response strategies: When receiving an early warning signal, the system starts the low-speed ventilation mode and activates cross-verification of sensors in adjacent areas; When receiving an emergency signal, the exhaust system will be forced to start at full power and a linkage command will be sent to the fire protection system; When the conflict status code is received three times in succession, the device starts the self-test program and switches to the backup sensor group; Among them, the timestamp alignment unit synchronizes the multimodal sensor data streams at the millisecond level, the dynamic mode decomposition algorithm unit separates the combustion characteristic mode and the conventional emission mode from the mixed signal, and the risk prediction unit triggers the graded response instructions according to the modal energy ratio; The system is also equipped with a dynamic learning module, whose working logic is as follows: After each ventilation control action was completed, a 20-second multimodal data segment before and after the event was extracted; When the actual fire alarm is inconsistent with the system warning result, a feature vector difference report is automatically generated; Adjust the mode cutoff order selection strategy in the dynamic mode decomposition algorithm based on the difference report.

[0024] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent environmental perception system based on multimodal sensors, characterized in that: include, Sensor group, including CO concentration sensor, particulate matter detector, infrared thermal imaging module and vibration sensor; Data processing module, with built-in timestamp alignment unit, dynamic mode decomposition algorithm unit and risk prediction unit; a ventilation control module, receiving an output signal from the data processing module; Among them, the timestamp alignment unit performs millisecond-level synchronization marking on the multimodal sensor data stream, the dynamic mode decomposition algorithm unit separates the combustion characteristic mode and the conventional emission mode from the mixed signal, and the risk prediction unit triggers the graded response instruction according to the modal energy ratio.

2. The multimodal sensor-based intelligent environmental perception system according to claim 1, characterized in that: The vibration sensor is integrated into the equipment mounting bracket. When it detects that the axial acceleration exceeds the preset safety value, the sensor data credibility attenuation mechanism is triggered. Specifically, Apply an attenuation weight coefficient to the multimodal data from the current moment; When the accumulated attenuation time exceeds the set threshold, historical normal operating condition data is automatically called to replace the abnormal period input.

3. The multimodal sensor-based intelligent environmental perception system according to claim 2, characterized in that: There is a 50%-70% overlap between the spatial coverage areas of the infrared thermal imaging module and the particle detector. Within this overlap area: When the number of thermal anomaly points exceeds the set value, the sampling frequency of the particulate matter detector is forced to increase to 3 times the baseline value; the sampling frequency increase strategy follows: base frequency × (1 + 0.05 current number of thermal anomaly points). When the current number of thermal anomaly points is greater than 20, the highest sampling frequency is locked; When the abnormal thermal distribution conforms to the battery module arrangement characteristics, the instantaneous peak data of the CO sensor is ignored.

4. The multimodal sensor-based intelligent environmental perception system according to claim 3, characterized in that: In the sensor group, when the vibration sensor inside the equipment mounting bracket detects abnormal acceleration, the attenuation mechanism is automatically triggered and the multimodal data is processed according to the time-varying weight until it is replaced with historical normal operating condition data after a timeout. The credibility attenuation mechanism includes: When the real-time axial acceleration satisfy: ， in, represents the unit step function, Indicates time The real-time axial acceleration, Represents the preset safety threshold. This function is used to determine the triggering moment. Trigger time ， in, Indicates taking the earliest moment that satisfies the conditions; Since the trigger moment, the attenuation weight is assigned to the multimodal data at each moment: ， in, Indicates time The data credibility weight, represents the natural exponential function, Represents the decay rate, which is used to control the weight decay speed; The decay rate By half-life parameter Sure: ， in, represents the natural logarithm operation, Indicates the time required for the weight to decay from 1 to 0.5; To determine whether to call historical replacement, define the cumulative triggering time: ， in, Indicates from To the current moment The cumulative activation time, the integral operation represents the cumulative time, Represents the time variable in the integration process, used to accumulate the effective time since the trigger; when , then for subsequent data streams Directly call historical normal operating condition data: ， in, Indicates the cumulative attenuation time threshold. Indicates the data output after replacement, Represents the multimodal data segment after mapping the historical working conditions to the current moment, It is the mapping moment, used to align the data length.

5. The multimodal sensor-based intelligent environmental perception system according to claim 1, characterized in that: The dynamic mode decomposition algorithm unit is configured as follows: When the absolute value of the CO concentration change rate is greater than the first threshold and the particulate matter number change rate is less than the second threshold, the fuel vehicle hot vehicle discrimination mode is activated; When the rate of change of the number of particulate matter exceeds the third threshold and the hysteresis time difference of the CO concentration change is greater than the set time, the electric vehicle thermal runaway warning mode is activated.

6. The multimodal sensor-based intelligent environmental perception system according to claim 5, characterized in that: The risk prediction unit includes a temporal convolutional network module, whose input layer receives the following features: Time domain differential of CO concentration and particulate matter number; Moving trajectory characteristics of the highest temperature point in the infrared thermal imaging area; The abnormal vibration spectrum of the equipment installation platform captured by the vibration sensor. The abnormal vibration spectrum characteristics include the alarm characteristics generated when the vibration sensor's energy proportion in the 5-15Hz frequency band exceeds 50% of the steady-state operating value; The output layer generates a risk level signal, which includes three status codes: normal, warning, and emergency.

7. The multimodal sensor-based intelligent environmental perception system according to claim 6, characterized in that: In the dynamic modal decomposition algorithm unit, three thresholds are adaptively set based on historical normal operating condition statistics, and mode activation logic is constructed accordingly. The process includes: Assume that the first-order time-domain differential series of CO concentration and particulate matter quantity under historical normal operating conditions are and ,definition: ， in, Indicates the first threshold, subscript 1 indicates the corresponding fuel vehicle hot vehicle discrimination mode, represents the statistical mean of the time derivative of CO concentration under historical normal operating conditions, where · represents the time derivative and the upper horizontal bar represents the statistical mean. represents the standard deviation of the first-order time-domain differential of CO concentration under historical normal operating conditions, where is a statistical operator, Indicates the safety factor, used to adjust the alarm trigger sensitivity; ， in, represents the second threshold, and the subscript 2 represents the corresponding particle stability fluctuation judgment. It represents the statistical mean of the time derivative of the number of particulate matter under historical normal working conditions, It represents the standard deviation of the first-order time-domain differential of the number of particulate matter under historical normal working conditions. Represents the sensitivity coefficient, which is used to control the tolerance to small fluctuations; ， in, Indicates the third threshold, subscript 3 indicates the corresponding electric vehicle thermal runaway warning judgment, Indicates the alarm coefficient, which is used to control the alarm intensity for large fluctuations; Define the unit step function: ， in, represents the unit step function, Represents function input variables; Define the mode activation function: ， in, Indicates the activation signal of the hot vehicle identification mode for fuel vehicles. Represents the first-order time-domain differential of the real-time CO concentration with respect to time, It represents the first-order time-domain differential of the real-time particle number with respect to time, and the symbol | | represents absolute value operation, ， in, Indicates the activation signal of the electric vehicle thermal runaway warning mode. represents the time lag between the change in particle number and the response of CO concentration, where Δ represents the difference operation, Indicates the preset hysteresis threshold.

8. The multimodal sensor-based intelligent environmental perception system according to claim 1, characterized in that: The ventilation control module implements a differentiated response strategy: When receiving an early warning signal, the system starts the low-speed ventilation mode and activates cross-verification of sensors in adjacent areas; When an emergency signal is received, full-power exhaust is forced to start and a linkage command is sent to the fire protection system.

9. The multimodal sensor-based intelligent environmental perception system according to claim 8, characterized in that: The differentiated response strategy implemented by the ventilation control module further includes: When the conflict status code is received three times in succession, the device starts the self-test procedure and switches to the backup sensor group.

10. The multimodal sensor-based intelligent environmental perception system according to claim 1, characterized in that: A dynamic learning module is also configured, and its working logic is as follows: After each ventilation control action was completed, a 20-second multimodal data segment before and after the event was extracted; When the actual fire alarm is inconsistent with the system warning result, a feature vector difference report is automatically generated; Adjust the mode cutoff order selection strategy in the dynamic mode decomposition algorithm based on the difference report.

Images