Fireproof door remote intelligent monitoring and alarm method and system based on internet of things

Abstract

The application relates to the technical field of Internet of Things monitoring, and discloses a fire door remote intelligent monitoring and alarm method and system based on the Internet of Things. The method comprises the following steps: obtaining image, temperature and door opening angle data through wireless networking and preprocessing to obtain a monitoring data set; performing Gaussian smoothing, standardization calibration and spatial coverage verification to obtain a multi-dimensional feature set; extracting image contrast features, temperature fluctuation values and door opening angle change rates to construct a comprehensive environment state vector; performing difference calculation with a reference state interval to generate an alarm priority score, and performing consistency verification when in a high-risk interval; generating an adjustment parameter according to a previous time domain data trend analysis, correcting the verification result to obtain a fire door operation state; and finally performing continuous judgment to generate an abnormal state identifier and send alarm information. The application can realize accurate identification and reliable alarm of the abnormal state of the fire door under complex environment and wireless networking transmission conditions.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) monitoring technology, and in particular to a method and system for remote intelligent monitoring and alarm of fire doors based on wireless networking. Background Technology

[0002] Fire doors are core fire-resistant partitions in fire-fighting scenarios such as high-rise buildings, commercial complexes, and evacuation routes. Their operational status is crucial for preventing the spread of fire and smoke. Real-time monitoring and automated supervision of fire door status is a core requirement of current building fire safety management.

[0003] In one existing technology, image acquisition, temperature detection, or angle sensing sensors are installed on fire doors to collect real-time operating data of the equipment. The collected data is then compared with a preset alarm threshold. When the data exceeds the threshold, an alarm signal is sent to the monitoring center via wireless networking to achieve remote monitoring of the status of individual fire doors.

[0004] However, existing technologies only make independent judgments based on a single monitoring parameter or simple threshold, failing to consider the interference of complex environments such as changes in lighting, airflow disturbances, and frequent personnel passage on sensor data. They also fail to account for the impact of asynchronous sampling of multi-source data and transmission delay fluctuations on feature extraction, thus failing to effectively filter out noise superposition in environmental data, leading to distorted feature recognition. In summary, existing technologies result in low accuracy and real-time performance in fire door operation status identification, and are prone to false alarms and missed alarms in complex environments. Summary of the Invention

[0005] This invention provides a remote intelligent monitoring and alarm method and system for fire doors based on the Internet of Things, in order to solve the problems of low accuracy and real-time performance in identifying the operating status of fire doors, and the tendency for false alarms and missed alarms to occur in complex environments.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a remote intelligent monitoring and alarm method for fire doors based on the Internet of Things, comprising:

[0007] Image data, temperature data, and door opening angle data are acquired through wireless networking and preprocessed to obtain a monitoring dataset.

[0008] Gaussian smoothing and normalization calibration are performed on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence, and a calibration angle sequence. Spatial coverage verification is performed on the smoothed image sequence, the calibration temperature sequence, and the calibration angle sequence to obtain a multidimensional feature set.

[0009] Image contrast features, temperature fluctuation values, and door opening angle change rate are extracted from the multidimensional feature set to construct a comprehensive environmental state vector.

[0010] The difference between the comprehensive environmental state vector and the preset baseline state interval is calculated to generate an alarm priority score. If the alarm priority score is in the preset high-risk interval, the comprehensive environmental state vector is compared for consistency to obtain a verification consistency result.

[0011] Based on the consistency verification results, extract the preceding time-domain data, perform trend analysis on the preceding time-domain data, and generate adjustment parameters;

[0012] The verification consistency result is corrected using the adjustment parameters, and the real-time monitoring data and the preceding time domain data are weighted, fused, and delayed to obtain the operating status of the fire door.

[0013] The system continuously assesses the operating status of the fire door, generates an abnormal status identifier, and sends an alarm message.

[0014] Secondly, the present invention provides a remote intelligent monitoring and alarm system for fire doors based on the Internet of Things, comprising:

[0015] The data processing module is used to acquire image data, temperature data, and door opening angle data through wireless networking, and to preprocess them to obtain a monitoring dataset;

[0016] The feature processing module is used to perform Gaussian smoothing and normalization calibration on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence and a calibration angle sequence, and to perform spatial coverage verification on the smoothed image sequence, the calibration temperature sequence and the calibration angle sequence to obtain a multi-dimensional feature set;

[0017] The vector construction module is used to extract image contrast features, temperature fluctuation values ​​and door opening angle change rate from the multi-dimensional feature set to construct a comprehensive environmental state vector.

[0018] The anomaly detection module is used to calculate the difference between the comprehensive environmental state vector and the preset baseline state interval to generate an alarm priority score. If the alarm priority score is in the preset high-risk interval, the comprehensive environmental state vector is compared for consistency to obtain a verification consistency result.

[0019] An interference identification module is used to extract preceding time-domain data based on the verification consistency result, perform trend analysis on the preceding time-domain data, and generate adjustment parameters.

[0020] The status assessment module is used to correct the verification consistency result using the adjustment parameters, and to perform weighted fusion and delay compensation on the real-time monitoring data and the preceding time domain data to obtain the operating status of the fire door.

[0021] An abnormality alarm module is used to continuously judge the operating status of the fire door, generate an abnormality status identifier, and send alarm information. Compared with the prior art, the present invention has the following advantages:

[0022] (1) This invention collects image data, temperature data and door opening angle data through wireless networking, and combines Gaussian smoothing, standardized calibration and spatial coverage verification to further extract image contrast features, temperature fluctuation values ​​and door opening angle change rate to form a comprehensive environmental state vector. It can transform single image, temperature or angle information into multi-dimensional associated state representation, thereby effectively solving the problem that existing technologies rely only on a single monitoring parameter or simple threshold for judgment, which is prone to feature distortion and misjudgment in complex scenarios such as light changes and airflow disturbances, and realizes more accurate identification of the operating status of fire doors.

[0023] (2) This invention generates an alarm priority score by calculating the difference between the comprehensive environmental state vector and the preset benchmark state interval. When the alarm priority score is in the preset high-risk interval, it combines the network transmission log, the source end attached check code and the receiver end check code to perform a consistency comparison of data integrity. It can verify the risks of packet loss, timeout, duplicate transmission and data tampering in the transmission link while assessing the degree of anomaly. This effectively solves the problem that the existing technology does not consider the delay fluctuation and integrity mismatch of multi-source data transmission, which leads to false alarms and missed alarms, and realizes the reliable output of the abnormal state judgment result.

[0024] (3) This invention extracts preceding time-domain data by verifying consistency results, constructs a continuous trend model to generate adjustment parameters, and uses the adjustment parameters to perform weighted fusion and delay compensation on real-time monitoring data and preceding time-domain data. Combined with the duration of abnormality, it generates an abnormal state identifier and sends alarm information. Under the condition of coexistence of dynamic environmental interference and response time delay, it can distinguish between instantaneous disturbances and continuous abnormalities, thereby effectively solving the problem that existing technologies are difficult to identify continuous abnormal states in a timely manner and have insufficient reliability of remote response, and realizing timely early warning and stable reporting of abnormal states of fire doors. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a remote intelligent monitoring and alarm method for fire doors based on the Internet of Things, provided in the first embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of a remote intelligent monitoring and alarm system for fire doors based on the Internet of Things, provided in the second embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, 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.

[0028] Reference Figure 1 The first embodiment of the present invention provides a remote intelligent monitoring and alarm method for fire doors based on the Internet of Things, including the following steps:

[0029] S11 acquires image data, temperature data, and door opening angle data through wireless networking, and performs preprocessing to obtain a monitoring dataset;

[0030] S12, Gaussian smoothing and normalization calibration are performed on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence and a calibration angle sequence. Spatial coverage verification is performed on the smoothed image sequence, the calibration temperature sequence and the calibration angle sequence to obtain a multidimensional feature set.

[0031] S13, extract image contrast features, temperature fluctuation values ​​and door opening angle change rate from the multi-dimensional feature set to construct a comprehensive environmental state vector;

[0032] S14, calculate the difference between the comprehensive environmental state vector and the preset baseline state interval to generate an alarm priority score. If the alarm priority score is in the preset high-risk interval, perform a consistency comparison on the comprehensive environmental state vector to obtain a consistency verification result.

[0033] S15, extract preceding time-domain data based on the verification consistency result, perform trend analysis on the preceding time-domain data, and generate adjustment parameters;

[0034] S16, the verification consistency result is corrected using the adjustment parameters, and the real-time monitoring data and the preceding time domain data are weighted and fused and delayed to obtain the operating status of the fire door;

[0035] S17. Based on the operating status of the fire door, continuously judge, generate an abnormal status identifier and send alarm information.

[0036] In step S11, image data, temperature data, and door opening angle data are acquired via wireless networking and preprocessed to obtain a monitoring dataset, including:

[0037] The system receives image data from the image sensor, temperature data from the temperature sensor, and door opening angle data from the angle sensor via a wireless communication network.

[0038] The image data, temperature data, and door opening angle data are each assigned a collection timestamp, and then rearranged in chronological order.

[0039] The rearranged image data, temperature data, and door opening angle data are subjected to abnormal null value removal and data format conversion to obtain preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data.

[0040] The preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data are time-aligned based on the acquisition timestamp and then structured and encapsulated to obtain the monitoring dataset.

[0041] Specifically, image sensors, temperature sensors, angle sensors, and gateway nodes are installed in the fire door monitoring area, and communication connections are established. After receiving data from each sensor, the gateway node parses and stores the image data, temperature data, and door opening angle data according to the sensor type. The image data consists of the image frame data output by the image sensor at the corresponding sampling time, the temperature data consists of the temperature sample value output by the temperature sensor at the corresponding sampling time, and the door opening angle data consists of the door opening angle sample value output by the angle sensor at the corresponding sampling time.

[0042] The data acquisition timestamps are recorded using a unified time reference, which is the standard time reference written to each sensor by the gateway node during the installation and commissioning phase. Each sensor synchronously records the corresponding sampling time when outputting image data, temperature data, and door opening angle data, and writes the sampling time into the corresponding data record, forming the acquisition timestamp. The unified time reference is established by the gateway node outputting a standard time. After receiving the standard time, each sensor corrects its local timing value and records the sampling time according to the corrected local timing value during subsequent sampling. After the acquisition timestamps are assigned, the image data, temperature data, and door opening angle data are rearranged in ascending order of acquisition timestamp.

[0043] It is worth noting that the process first checks whether the data payload in each data record is empty. Records with empty data payloads are directly removed. Then, outlier checks are performed on the non-empty data.

[0044] For temperature data, during the installation and commissioning phase, temperature samples are continuously collected in the closed, open, and personnel passage states of the fire doors. The temperature samples are sorted according to their size, and the 1st percentile is taken as the lower boundary of the field calibration, and the 99th percentile is taken as the upper boundary of the field calibration. The lower and upper boundaries of the field calibration are used as the effective measurement range of the temperature data. Temperature data outside the effective measurement range are judged as abnormal temperature data.

[0045] For door opening angle data, during the installation and commissioning phase, the angle values ​​corresponding to the fully closed and fully open positions of the door are recorded. The 1% and 99th percentile values ​​of the repeated measurement results are used as the lower and upper boundaries of the angle, respectively. The lower and upper boundaries of the angle constitute the effective measurement range of the door opening angle data. Door opening angle data outside the effective measurement range are judged as abnormal angle data.

[0046] For image data, check the image size, image decoding status, and image integrity. Image data whose image size is inconsistent with the set resolution, whose image decoding fails, or whose image content is incomplete is judged as abnormal image data.

[0047] After removing outliers, the retained image data is uniformly converted into a pixel matrix of a fixed size, the retained temperature data is uniformly converted into degrees Celsius values, and the retained door opening angle data is uniformly converted into angle values, thus obtaining preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data, respectively.

[0048] Specifically, during the installation and commissioning phase, the transmission delays of multiple sets of image data, temperature data, and door opening angle data were statistically analyzed. The 95th percentile of each of the three types of data transmission delays was calculated. Then, the residual time error after unifying the time base was read. The sum of the maximum value of the 95th percentile of the three types of data transmission delays and the residual time error was taken as the time alignment tolerance. After determining the time alignment tolerance, preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data with acquisition timestamp differences not exceeding the time alignment tolerance were grouped into the same time slice. The data within the same time slice was then structured and encapsulated to generate a monitoring record. The monitoring dataset consists of multiple monitoring records arranged chronologically. Each monitoring record includes at least an acquisition timestamp, preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data.

[0049] In step S12, Gaussian smoothing and normalization calibration are performed on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence, and a calibration angle sequence. Spatial coverage verification is then performed on the smoothed image sequence, the calibration temperature sequence, and the calibration angle sequence to obtain a multidimensional feature set, including:

[0050] The preprocessed image data, the preprocessed temperature data, and the preprocessed door opening angle data are separated from the monitoring dataset, and the preprocessed image data is Gaussian smoothed to obtain a smoothed image sequence.

[0051] The preprocessed temperature data and the preprocessed door opening angle data are separated from the monitoring dataset, and the preprocessed temperature data and the preprocessed door opening angle data are normalized respectively to obtain the calibration temperature sequence and the calibration angle sequence.

[0052] Read the sensor distribution coordinates, perform distribution statistics according to a preset spatial grid to form a sensor density distribution map, and perform matrix processing on the sensor density distribution map based on a preset density threshold to obtain a spatial coverage matrix.

[0053] The multidimensional feature set is obtained by removing invalid data at the edges of the smoothed image sequence, the calibration temperature sequence, and the calibration angle sequence using the spatial coverage matrix.

[0054] Specifically, each monitoring record in the monitoring dataset is read, and preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data are extracted according to field type. These are then arranged into image time-series queues, temperature time-series queues, and angle time-series queues according to the acquisition timestamp from smallest to largest. Subsequently, Gaussian smoothing is performed on each frame of preprocessed image data in the image time-series queue. For each pixel in the current frame of preprocessed image data, an odd-length convolution window is selected centered on that pixel, and the Euclidean distance between each pixel in the convolution window and the center pixel is calculated. The square of the Euclidean distance is then divided by twice the square of the Gaussian standard deviation, and the inverse of the result is used to calculate the natural exponent value, thus obtaining the corresponding original Gaussian weight.

[0055] After calculating all the original Gaussian weights within the convolution window, the total weights are summed. Then, each original Gaussian weight is divided by the total weight to obtain the normalized Gaussian weights, making the sum of all normalized Gaussian weights within the convolution window equal to one. Subsequently, each pixel value within the convolution window is multiplied by its corresponding normalized Gaussian weight and summed to obtain the smoothed pixel value of the center pixel. After processing all pixels in the current frame, the smoothed image of the current frame is obtained. The above processing is performed sequentially on all preprocessed image data in the image temporal queue to obtain a smoothed image sequence.

[0056] It should be noted that the side length of the convolution window and the Gaussian standard deviation are predetermined during the installation and debugging phase. Historical image samples of the fire door in closed, open, and personnel passage states are collected. The gray-level change area at the edge of the door is extracted, and the pixel span corresponding to the transition of gray-level from the background stable value to the door stable value is counted along the direction perpendicular to the edge of the door. The pixel span is determined as the edge transition width. Then, all edge transition widths are sorted according to their size, and the median value is taken as the edge transition width.

[0057] Subsequently, half of the edge transition width is determined as the radius of the candidate convolution window, and the smallest odd number not less than twice the radius of the candidate convolution window plus one is taken as the side length of the candidate convolution window; then, the Gaussian standard deviation corresponding to the side length of the candidate convolution window is determined according to the distance from the center of the convolution window to the edge of the window being equal to three Gaussian standard deviations.

[0058] Subsequently, Gaussian smoothing was applied to historical image samples using candidate convolution window side lengths and Gaussian standard deviations. The decrease in background gray-level variance and the preservation ratio of the door edge gradient were calculated before and after smoothing. The preservation ratio of the door edge gradient was the ratio of the peak gradient at the door edge after smoothing to the peak gradient at the door edge before smoothing. The door edge gradient preservation ratios corresponding to all candidate parameter groups were then sorted according to their magnitude, and the 25th percentile was used as the preset preservation threshold. Among the candidate parameter groups whose door edge gradient preservation ratio was not lower than the preset preservation threshold, the parameter group with the largest decrease in background gray-level variance was selected as the Gaussian smoothing parameter corresponding to the current installation point. During the runtime phase, the Gaussian smoothing parameter was consistently used to perform Gaussian smoothing on the preprocessed image data.

[0059] Specifically, the system reads each monitoring record from the monitoring dataset, extracts preprocessed temperature data and preprocessed door opening angle data according to field type, and retains the corresponding acquisition timestamp and sensor number for each data point. Then, the preprocessed temperature data is arranged in ascending order of acquisition timestamp to form a temperature time-series queue, and the preprocessed door opening angle data is arranged in ascending order to form an angle time-series queue. In cases where multiple temperature sensor data and multiple angle sensor data exist under the same acquisition timestamp, the corresponding values ​​are saved according to the sensor number for subsequent normalization processing.

[0060] It should be noted that the temperature normalization interval and angle normalization interval are determined separately, based on the historical temperature samples collected by each temperature sensor during the installation and commissioning phase. The historical temperature samples are sorted from largest to smallest, and the 1st percentile value is taken as the lower temperature boundary, and the 99th percentile value is taken as the upper temperature boundary. The angle normalization interval is determined based on the historical door opening angle samples collected multiple times in both fully closed and fully open states. The historical door opening angle samples are sorted from largest to smallest, and the 1st percentile value is taken as the lower angle boundary, and the 99th percentile value is taken as the upper angle boundary.

[0061] During normalization, for any temperature sample value, minmax normalization is performed. When the temperature sample value is less than or equal to the lower temperature limit, the normalization result is recorded as zero; when the temperature sample value is greater than or equal to the upper temperature limit, the normalization result is recorded as one. For any door opening angle sample value, normalization is performed in the same way as for temperature. After completing the normalization processing for all temperature sample values ​​and all door opening angle sample values, all normalized temperature values ​​are arranged in order of acquisition timestamp to obtain the calibration temperature sequence; all normalized angle values ​​are arranged in order of acquisition timestamp to obtain the calibration angle sequence.

[0062] Specifically, the distribution coordinates of each sensor in the installation record are read. The sensor distribution coordinates are recorded using a unified plane coordinate system, with the lower left corner of the fire door frame as the origin, the bottom edge of the door frame as the horizontal coordinate direction, and the side edge of the door frame as the vertical coordinate direction. After installation, the installation positions of each temperature sensor and angle sensor are mapped to a coordinate point in the unified plane coordinate system and written into the installation record.

[0063] Subsequently, the distribution coordinates of the sensors are statistically analyzed. First, the distance between any sensor distribution coordinate and its nearest neighbor sensor distribution coordinate is calculated, and all nearest neighbor distances are sorted from largest to smallest. The median value is taken as the grid side length. Then, the monitoring area corresponding to the fire door is divided into grids according to the grid side length, and multiple grid units are combined to form a preset spatial grid. After that, the number of sensor distribution coordinates falling within the preset spatial grid is counted, and the number of coordinates corresponding to each grid unit is written into the statistical unit corresponding to the grid position, thus forming a sensor density distribution map.

[0064] It is worth noting that, firstly, the number of all sensors with a value greater than zero in the sensor density distribution map is extracted and sorted from largest to smallest. The result after rounding up the 25th percentile value is taken as the preset density threshold. Then, each grid cell in the sensor density distribution map is compared with the preset density threshold one by one. When the number of sensors in a certain grid cell is greater than or equal to the preset density threshold, the grid cell is recorded as an effective coverage cell and a 1 is written in the corresponding matrix position. When the number of sensors in a certain grid cell is less than the preset density threshold, the grid cell is recorded as an invalid coverage cell and a zero is written in the corresponding matrix position. After completing the above processing for all grid cells, a spatial coverage matrix composed of zeros and ones is obtained.

[0065] Specifically, for smoothed image sequences, the correspondence between image pixel positions and preset spatial grids is first established. Then, image pixels falling into grid regions where the matrix unit value is zero are identified as edge invalid image data and removed from the corresponding smoothed image, retaining only image pixels falling into grid regions where the matrix unit value is one. For calibration temperature and calibration angle sequences, the sensor distribution coordinates corresponding to each temperature and angle sensor are read, and the grid unit where each sensor distribution coordinate is located is determined. When the value of the corresponding grid unit in the spatial coverage matrix is ​​zero, the calibration temperature data or calibration angle data of that sensor under the corresponding acquisition timestamp is identified as edge invalid data and removed. When the value of the corresponding grid unit in the spatial coverage matrix is ​​one, the corresponding calibration temperature data or calibration angle data is retained.

[0066] For each acquisition timestamp, smooth image data, calibration temperature data, and calibration angle data after removing invalid edge data are extracted. The various types of data retained under the same acquisition timestamp are encapsulated into a feature record. Each feature record includes at least the acquisition timestamp, as well as the smooth image data, calibration temperature data, and calibration angle data after removing invalid edge data. The feature records corresponding to all acquisition timestamps constitute a multidimensional feature set.

[0067] In step S13, image contrast features, temperature fluctuation values, and door opening angle change rate are extracted from the multidimensional feature set to construct a comprehensive environmental state vector, including:

[0068] The pixel grayscale values ​​of each frame of the image in the multidimensional feature set are statistically analyzed, and the image contrast features are determined based on the standard deviation of the pixel grayscale values.

[0069] The temperature difference between adjacent sampling times is calculated using the calibration temperature sequence in the multidimensional feature set to obtain the temperature fluctuation value.

[0070] Calculate the angle difference between adjacent sampling times in the calibration angle sequence of the multidimensional feature set, and determine the door opening angle change rate based on the angle difference and the interval between adjacent sampling times;

[0071] The image contrast features, temperature fluctuation values, and door opening angle change rate are concatenated and mapped to a preset state vector space to obtain the comprehensive environmental state vector.

[0072] Specifically, smoothed image data corresponding to each acquisition timestamp is read one by one from the multidimensional feature set, and the smoothed image data corresponding to each acquisition timestamp is determined as the image frame at the corresponding time. Then, all valid pixel positions in the current image frame are traversed, and the pixel grayscale values ​​corresponding to each valid pixel position are read to form the pixel grayscale value set of the current image frame. Then, the average grayscale value of all pixel grayscale values ​​in the current image frame is calculated based on the pixel grayscale value set, and the dispersion of each pixel grayscale value relative to the average grayscale value is calculated. The standard deviation corresponding to the dispersion is determined as the image contrast feature of the current image frame. The above processing is performed sequentially on the smoothed image data corresponding to each acquisition timestamp in the multidimensional feature set to obtain the image contrast feature corresponding to each acquisition timestamp.

[0073] It should be noted that the calibration temperature data corresponding to each acquisition timestamp is read one by one from the multidimensional feature set, and a calibration temperature sequence is formed in ascending order of acquisition timestamps. In the case of multiple calibration temperature data points for the same acquisition timestamp, all calibration temperature data for that timestamp are first extracted, then the arithmetic mean of all the data is calculated, and this arithmetic mean is determined as the sequence temperature value corresponding to that acquisition timestamp. After determining the sequence temperature values ​​corresponding to all acquisition timestamps, the sequence temperature value corresponding to the next sampling time is subtracted from the sequence temperature value corresponding to the previous sampling time according to the chronological order of adjacent sampling times to obtain the temperature difference between adjacent sampling times.

[0074] It is worth noting that the above difference calculation is performed sequentially on each group of adjacent sampling times in the calibration temperature sequence, and the absolute value of the temperature difference is determined as the temperature fluctuation value of the corresponding adjacent sampling time. This temperature fluctuation value is assigned to the next sampling time, and all temperature fluctuation values ​​are arranged in ascending order of the collection timestamp to obtain the temperature fluctuation value corresponding to each sampling time.

[0075] Specifically, the calibration angle data corresponding to each acquisition timestamp is read one by one from the multidimensional feature set, and a calibration angle sequence is formed according to the acquisition timestamps in ascending order. After the calibration angle sequence is constructed, the calibration angle data corresponding to the previous sampling time is subtracted from the calibration angle data corresponding to the next sampling time in the order of adjacent sampling times to obtain the angle difference between adjacent sampling times; then the acquisition timestamps corresponding to adjacent sampling times are read, and the time interval between the next sampling time and the previous sampling time is calculated; finally, the angle difference is divided by the time interval to obtain the door opening angle change rate corresponding to the adjacent sampling time, and this door opening angle change rate is attributed to the next sampling time.

[0076] It is worth noting that the image contrast features, temperature fluctuation values, and door opening angle change rate are unified to the same time base, with the temperature fluctuation value and door opening angle change rate using the subsequent sampling time as the corresponding timestamp. The three are matched according to the acquisition timestamp, and the image contrast features, temperature fluctuation values, and door opening angle change rate under the same acquisition timestamp are combined in a column-directed manner according to a preset order to obtain the initial feature vector corresponding to the current acquisition timestamp. The preset order is image contrast features, temperature fluctuation values, and door opening angle change rate. The above processing is performed on all acquisition timestamps in sequence to obtain an initial feature vector set that corresponds one-to-one with each acquisition timestamp.

[0077] It should be noted that during the installation and commissioning phase, historical sample data of the fire door in closed, open, and personnel passage states are collected, and the image contrast features, temperature fluctuation values, and door opening angle change rates corresponding to each historical sample are calculated according to the aforementioned steps. Subsequently, each historical sample is combined in column direction according to the order of image contrast features, temperature fluctuation values, and door opening angle change rates to obtain a set of historical feature vectors. Then, all values ​​of each dimension of the historical feature vector set are extracted and sorted in descending order. The 1st percentile value is taken as the lower boundary of the corresponding dimension, and the 99th percentile value is taken as the upper boundary of the corresponding dimension to obtain the preset state vector space.

[0078] It is worth noting that, based on the state vector space established during the installation and debugging phase, the initial feature vector corresponding to the current acquisition timestamp is normalized and mapped. The image contrast feature, temperature fluctuation value, and door opening angle change rate in the initial feature vector are read respectively and compared with the lower and upper boundaries of the corresponding dimensions in the state vector space. When any feature value of the image contrast feature, temperature fluctuation value, and door opening angle change rate is less than or equal to the corresponding lower boundary, the mapping value for that dimension is recorded as zero; when any feature value of the image contrast feature, temperature fluctuation value, and door opening angle change rate is greater than or equal to the corresponding upper boundary, the mapping value for that dimension is recorded as one; when any feature value of the image contrast feature, temperature fluctuation value, and door opening angle change rate is between the corresponding lower and upper boundaries, the corresponding lower boundary is subtracted from the feature value of that dimension, and then divided by the difference between the corresponding upper and lower boundaries to obtain the mapping value for that dimension.

[0079] After mapping all dimensions, the mapped values ​​of each dimension are combined in column direction according to the order of image contrast features, temperature fluctuation values ​​and door opening angle change rate to obtain the comprehensive environmental state vector corresponding to the current acquisition timestamp; after performing the above processing on all acquisition timestamps in sequence, the comprehensive environmental state vector is obtained.

[0080] In step S14, the difference between the integrated environmental state vector and a preset baseline state interval is calculated to generate an alarm priority score. If the alarm priority score is in a preset high-risk interval, the integrated environmental state vector is compared for consistency to obtain a consistency verification result, including:

[0081] Extract the center point vector and boundary threshold in the preset baseline state interval, calculate the dimension-by-dimensional difference between the comprehensive environmental state vector and the center point vector, and normalize the difference in each dimension to obtain the feature deviation value set.

[0082] The abnormality level is determined by comparing the set of feature deviation values ​​with the boundary threshold, and the abnormality level is processed according to the preset abnormality level quantification table to generate an alarm priority score.

[0083] If the alarm priority score is in a preset high-risk range, the network transmission log for the corresponding time period is read and the integrity of the feature deviation value set is verified based on the network transmission log to obtain the transmission status result.

[0084] Extract the source-side attached checksum corresponding to the transmission status result, generate the receiver-side checksum based on the transmission status result, and perform a consistency comparison between the source-side attached checksum and the receiver-side checksum to obtain a consistency result.

[0085] Specifically, during the installation and commissioning phase, historical sample data of fire doors in normal closed, normal open, and normal personnel passage states are collected. Following the previous steps, image contrast features, temperature fluctuation values, and door opening angle change rates are calculated for each historical sample. Then, the image contrast features, temperature fluctuation values, and door opening angle change rates corresponding to each historical sample are combined into a historical state vector set according to the collection timestamp. All historical values ​​of the image contrast features, temperature fluctuation values, and door opening angle change rates are extracted, sorted from largest to smallest, and the 1st percentile value is taken as the lower boundary of the corresponding feature, and the 99th percentile value is taken as the upper boundary of the corresponding feature. The lower and upper boundaries of the image contrast features, the lower and upper boundaries of the temperature fluctuation values, and the lower and upper boundaries of the door opening angle change rates are combined to obtain a preset baseline state interval.

[0086] It should be noted that the lower and upper boundaries corresponding to the image contrast feature, temperature fluctuation value, and door opening angle change rate in the preset baseline state interval are read. The median values ​​of the lower and upper boundaries of the image contrast feature, the lower and upper boundaries of the temperature fluctuation value, and the lower and upper boundaries of the door opening angle change rate are calculated respectively. The above three median values ​​are then combined in the order of the image contrast feature, temperature fluctuation value, and door opening angle change rate to obtain the center point vector.

[0087] It is worth noting that the lower and upper boundaries corresponding to the image contrast feature, temperature fluctuation value, and door opening angle change rate in the preset baseline state interval are read. Half of the difference between the upper and lower boundaries of the image contrast feature, half of the difference between the upper and lower boundaries of the temperature fluctuation value, and half of the difference between the upper and lower boundaries of the door opening angle change rate are calculated respectively. The widths of the above three half intervals are then combined sequentially according to the order of the image contrast feature, temperature fluctuation value, and door opening angle change rate to obtain the boundary threshold.

[0088] Specifically, the image contrast feature components, temperature fluctuation value components, and door opening angle change rate components in the comprehensive environmental state vector are read, and the median value of the corresponding position in the center point vector is subtracted from each of them to obtain the image contrast feature difference, temperature fluctuation value difference, and door opening angle change rate difference.

[0089] The image contrast feature component, temperature fluctuation value component, and door opening angle change rate component in the comprehensive environmental state vector are read respectively, and the median value of the corresponding position in the center point vector is subtracted respectively to obtain the image contrast feature difference, temperature fluctuation value difference, and door opening angle change rate difference.

[0090] Read the image contrast feature threshold, temperature fluctuation value threshold, and door opening angle change rate threshold from the boundary threshold respectively. Divide the absolute value of the difference in image contrast features by the image contrast feature threshold to obtain the image contrast feature deviation value; divide the absolute value of the difference in temperature fluctuation values ​​by the temperature fluctuation value threshold to obtain the temperature fluctuation value deviation value; divide the absolute value of the difference in door opening angle change rate by the door opening angle change rate threshold to obtain the door opening angle change rate deviation value.

[0091] The deviation values ​​of image contrast features, temperature fluctuation values, and door opening angle change rate are combined sequentially according to the order of image contrast features, temperature fluctuation values, and door opening angle change rate to obtain a set of feature deviation values.

[0092] It is worth noting that the image contrast feature deviation value, temperature fluctuation value deviation value, and door opening angle change rate deviation value are read from the feature deviation value set, and the corresponding image contrast feature threshold, temperature fluctuation value threshold, and door opening angle change rate threshold are read from the boundary threshold. Then, the image contrast feature deviation value is compared with the image contrast feature threshold, the temperature fluctuation value deviation value is compared with the temperature fluctuation value threshold, and the door opening angle change rate deviation value is compared with the door opening angle change rate threshold. The excess ratio of each feature deviation value relative to the corresponding threshold is calculated.

[0093] Specifically, the degree of anomaly is determined based on the proportion of each feature's deviation relative to its corresponding threshold. When a feature's deviation is less than or equal to its corresponding threshold, the proportion of that feature's deviation is recorded as zero. When a feature's deviation is greater than its corresponding threshold, the difference between the feature's deviation and the threshold is divided by the threshold to obtain the proportion of that feature's deviation. After calculating the proportions for image contrast, temperature fluctuation, and door opening angle change rate, the maximum value among the three proportions is taken as the degree of anomaly corresponding to the current acquisition timestamp.

[0094] During the installation and commissioning phase, historical samples were labeled with their status based on actual on-site observation records. Samples corresponding to periods when fire doors remained closed or normally open without any abnormal alarms were labeled as normal operation samples; samples showing only personnel passage, short-term airflow disturbances, or changes in lighting, with the door automatically returning to its original state during the observation period, were labeled as slightly disturbed samples; samples exhibiting at least one of the following: continuous door opening, continuous temperature increase, or continuous abnormal changes in image contrast, with the abnormal state persisting during the observation period, were labeled as obviously abnormal samples; and samples exhibiting both continuous door opening and continuous temperature increase, or an abnormally high rate of temperature increase, with high-risk alarm records triggered on-site, were labeled as high-risk abnormal samples.

[0095] Based on the annotation results, the recovery time of the gate from the perturbation state to the pre-perturbation state was extracted from all slightly perturbed samples, and sorted according to the magnitude. The 95th percentile value was taken as the preset recovery time. The duration of the abnormal state was continuously maintained in all obviously abnormal samples, and sorted according to the magnitude. The 5th percentile value was taken as the preset abnormal duration. The temperature rise per unit time was extracted from all high-risk abnormal samples, and sorted according to the magnitude. The 5th percentile value was taken as the preset high-risk heating rate.

[0096] Specifically, the abnormality values ​​corresponding to normal operation samples, slightly disturbed samples, obviously abnormal samples, and high-risk abnormal samples are extracted and sorted according to their magnitude. The 99th percentile of the abnormality value of normal operation samples is taken as the upper limit of the first-level interval, the 99th percentile of the abnormality value of slightly disturbed samples is taken as the upper limit of the second-level interval, and the 99th percentile of the abnormality value of obviously abnormal samples is taken as the upper limit of the third-level interval. The abnormality value above the upper limit of the third-level interval is determined as the fourth-level interval. Thus, the first-level interval, second-level interval, third-level interval, and fourth-level interval corresponding to the abnormality value are obtained.

[0097] The first-level interval, second-level interval, third-level interval, and fourth-level interval are respectively mapped to the first-level alarm priority score, second-level alarm priority score, third-level alarm priority score, and fourth-level alarm priority score to form an anomaly degree quantification table.

[0098] It should be noted that the anomaly severity quantification table is read, and the anomaly severity corresponding to the current collection timestamp is compared with each level of interval in the anomaly severity quantification table one by one; when the anomaly severity falls into the first level interval, a first-level alarm priority score is generated; when the anomaly severity falls into the second level interval, a second-level alarm priority score is generated; when the anomaly severity falls into the third level interval, a third-level alarm priority score is generated; when the anomaly severity falls into the fourth level interval, a fourth-level alarm priority score is generated; thus, the alarm priority score is obtained.

[0099] The alarm priority score is compared with the preset high-risk interval in the anomaly quantification table. When the alarm priority score falls into the preset high-risk interval, the data record corresponding to the current comprehensive environmental state vector is determined to enter the integrity verification process. The preset high-risk interval is the interval corresponding to the fourth-level alarm priority.

[0100] It should be noted that the corresponding time period is determined based on the acquisition timestamp corresponding to the current comprehensive environmental state vector. Specifically, the current acquisition timestamp corresponding to the generation of the feature deviation value set is read, and the acquisition timestamp corresponding to the previous sampling time is also read; then, the time interval between the acquisition timestamp corresponding to the previous sampling time and the current acquisition timestamp is determined as the corresponding time period. In the case where the current acquisition timestamp does not have a previous sampling time, the time interval corresponding to one sampling period backward from the current acquisition timestamp is determined as the corresponding time period, thus obtaining the corresponding time period.

[0101] Specifically, the network transmission log includes at least the sending node number, receiving node number, data type identifier, sending time, receiving time, sequence number, fragment sequence number, and payload length. Log records whose data type identifier matches the data type corresponding to the current feature deviation value set and whose sending time falls within the corresponding time period are selected from the network transmission log. The selected log records are then identified as the network transmission logs corresponding to the current verification object.

[0102] It is worth noting that, firstly, the set of sending sequence numbers recorded by the sending node and the set of receiving sequence numbers recorded by the receiving node within the corresponding time period are counted. When a sequence number exists in the sending sequence number set but not in the receiving sequence number set, the corresponding data is determined to be lost data. Then, the time difference between the sending time and the receiving time corresponding to each sequence number is calculated, and the time difference is compared with the transmission timeout threshold. When the time difference corresponding to a certain sequence number is greater than the transmission timeout threshold, the corresponding data is determined to be timed out data. Subsequently, the sequence numbers that appear repeatedly in the receiving sequence number set are counted. When the same sequence number appears more than once in the receiving sequence number set, the corresponding data is determined to be duplicate transmission data.

[0103] Specifically, during the installation and debugging phase, historical network transmission logs are collected, and the time difference between the sending and receiving times corresponding to each sequence number is calculated. Then, all time differences are sorted according to their magnitude, and the 99th percentile value is taken as the transmission timeout threshold to obtain the transmission timeout threshold.

[0104] When there is no packet loss, no timeout, and no duplicate transmission within the corresponding time period, the transmission status result is determined as a complete transmission; when there is at least one of the following situations within the corresponding time period: packet loss, timeout, or duplicate transmission, the transmission status result is determined as an incomplete transmission, and the transmission status result is obtained.

[0105] It should be noted that, firstly, based on the collection timestamp, node number, and sequence number corresponding to the transmission status result, the original data record on the sending side corresponding to the current feature deviation value set is located; then, the check code field written by the sending node when sending data is extracted from the original data record on the sending side, and the check code field is determined as the check code attached to the source end.

[0106] When the transmission status result is complete transmission, all received data corresponding to the current acquisition timestamp, node number, and sequence number in the data record of the receiving side is read, and all received data is reassembled in ascending order of sequence number and fragment number to obtain the effective payload data of the receiving end. The same check calculation method as the source end attached check code generation method is used to perform check calculation on the effective payload data of the receiving end byte by byte, and the calculation result is determined as the receiving end check code. The source end attached check code and the receiving end check code are compared bit by bit. When all bits of the two are the same, the check consistency result is determined as check consistency; otherwise, it is determined as check inconsistency.

[0107] When the transmission status result is an incomplete transmission, there is no need to calculate the check code; the check consistency result is directly determined to be an inconsistency.

[0108] In step S15, prior time-domain data is extracted based on the verification consistency result, trend analysis is performed on the prior time-domain data, and adjustment parameters are generated, including:

[0109] The time of an anomaly occurrence is determined based on the verification consistency result, the time of an anomaly occurrence is set as the endpoint, and the backtracking period corresponding to the time of an anomaly occurrence is determined.

[0110] Image data, temperature data, and door opening angle data are extracted from the backtracking period. Based on the image data, temperature data, and door opening angle data, the image contrast features, temperature fluctuation values, and door opening angle change rates corresponding to each sampling time are calculated to obtain the preceding time domain data.

[0111] The preceding time-domain data at adjacent sampling times are subjected to difference calculations to obtain continuous feature values;

[0112] A continuous trend model is constructed using the continuous feature values, and the changing trend of the preceding time domain data is determined based on the continuous trend model.

[0113] Based on the changing trend and the preset fluctuation threshold, the degree of trend disturbance is determined, and the adjustment parameters are generated based on the degree of trend disturbance.

[0114] It should be noted that when the verification consistency result is consistent and the transmission status result is complete transmission, the current collection timestamp is determined as the time of the anomaly. When the verification consistency result is inconsistent or the transmission status result is incomplete transmission, the historical records are searched in reverse order from the current time according to the collection timestamp, and the most recent collection timestamp in the historical records with a verification consistency result of consistent, a transmission status result of complete transmission, and an alarm priority score in the preset high-risk range is extracted and determined as the time of the anomaly. When there is no collection timestamp in the historical records that meets the above conditions, the current collection timestamp is determined as the time of the anomaly, thus obtaining the time of the anomaly.

[0115] First, extract the obvious and high-risk abnormal samples marked during the installation and debugging phase, identify them as historical abnormal samples, and obtain the time of occurrence of each historical abnormal sample. Then, retrieve the comprehensive environmental state vector corresponding to each historical sampling time one by one from the time of occurrence of each abnormal sample, and determine the most recent collection timestamp where the comprehensive environmental state vector falls back into the preset baseline state interval as the backtracking start time of the corresponding historical abnormal sample. Subsequently, calculate the time length between the time of occurrence of each historical abnormal sample and the corresponding backtracking start time to obtain multiple sets of historical backtracking durations. Then, sort all the historical backtracking durations in descending order, and take the 95th percentile value as the backtracking duration to obtain the backtracking duration.

[0116] Specifically, the time length corresponding to the time period corresponding to the time period before the time of the anomaly is traced back is used to obtain the time period of the traceback. Then, the time interval between the time period of the time period of the traceback and the time of the anomaly is determined as the time period of the traceback. When the time period of the traceback is earlier than the earliest collection timestamp that can be read from the current monitoring data, the earliest collection timestamp is determined as the time period of the traceback, and the time interval between the earliest collection timestamp and the time of the anomaly is determined as the time period of the traceback. The historical monitoring records corresponding to the time period of the traceback are read, and the image data, temperature data, and door opening angle data within the time period of the traceback are extracted in ascending order of collection timestamp. Then, the image data is formed into a historical image sequence in the order of collection timestamp, the temperature data is formed into a historical temperature sequence in the order of collection timestamp, and the door opening angle data is formed into a historical angle sequence in the order of collection timestamp.

[0117] For each frame in the historical image sequence, following the aforementioned method of statistically analyzing pixel grayscale values ​​and determining image contrast features based on the standard deviation of pixel grayscale values, the image contrast features corresponding to each sampling time are calculated frame by frame to obtain a historical image contrast feature sequence. Following the aforementioned method of calculating the temperature difference between adjacent sampling times, the temperature data of adjacent sampling times in the historical temperature sequence are sequentially subjected to difference calculation, and the absolute value of the temperature difference between each adjacent sampling time is determined as the temperature fluctuation value at the corresponding sampling time, thus obtaining a historical temperature fluctuation value sequence.

[0118] Similarly, following the aforementioned method of calculating the angle difference between adjacent sampling times and determining the door opening angle change rate based on the interval between adjacent sampling times, the door opening angle data at adjacent sampling times in the historical angle sequence are calculated sequentially to obtain the historical door opening angle change rate sequence. Likewise, the historical image contrast feature sequence, historical temperature fluctuation value sequence, and historical door opening angle change rate sequence are matched according to the acquisition timestamp, and the image contrast feature, temperature fluctuation value, and door opening angle change rate corresponding to the same acquisition timestamp are combined into a single time-series feature record; all time-series feature records within the retrospective period constitute the preceding time-domain data.

[0119] Read each temporal feature record from the preceding time-domain data and arrange them in ascending order of the acquisition timestamp. Each temporal feature record includes at least the image contrast feature, temperature fluctuation value, and door opening angle change rate at the corresponding sampling time. Then, according to the chronological order of adjacent sampling times, pair the temporal feature record corresponding to the previous sampling time with the temporal feature record corresponding to the next sampling time to obtain adjacent sampling time pairs.

[0120] Specifically, the image contrast feature corresponding to the previous sampling time is subtracted from the image contrast feature corresponding to the subsequent sampling time to obtain the image contrast difference value; the temperature fluctuation value corresponding to the previous sampling time is subtracted from the temperature fluctuation value corresponding to the subsequent sampling time to obtain the temperature fluctuation difference value; the door opening angle change rate corresponding to the previous sampling time is subtracted from the door opening angle change rate corresponding to the subsequent sampling time to obtain the door opening angle change rate difference value, thus obtaining the difference result. The image contrast difference value, temperature fluctuation difference value, and door opening angle change rate difference value corresponding to the same adjacent sampling time are sequentially combined according to the order of image contrast feature, temperature fluctuation value, and door opening angle change rate to obtain the continuity feature value of the corresponding adjacent sampling time; after performing the above processing sequentially on all adjacent sampling times in the preceding time domain data, the continuity feature value is obtained.

[0121] It is worth noting that the continuous trend model consists of an image contrast trend sub-model, a temperature fluctuation trend sub-model, and a door opening angle change rate trend sub-model. Among them, the image contrast trend sub-model is used to characterize the relationship between the image contrast difference value and the sampling time, the temperature fluctuation trend sub-model is used to characterize the relationship between the temperature fluctuation difference value and the sampling time, and the door opening angle change rate trend sub-model is used to characterize the relationship between the door opening angle change rate difference value and the sampling time.

[0122] It should be noted that the sampling times corresponding to the preceding time-domain data are numbered in ascending order to obtain a time sequence. This time sequence is used as the input variable for each trend sub-model, and the image contrast difference sequence, temperature fluctuation difference sequence, and door opening angle change rate difference sequence are used as the output variables for their respective trend sub-models. For any trend sub-model, a first-order linear model is used as the model architecture. The first-order linear model includes a slope parameter and an intercept parameter. The slope parameter represents the direction and rate of change of the corresponding continuous feature value over time, while the intercept parameter represents the baseline level of the corresponding continuous feature value at the starting point of time.

[0123] Specifically, the slope and intercept parameters of each trend sub-model are determined using the least squares fitting method. For any trend sub-model, a one-to-one correspondence is first established between all time indices and all observed values ​​corresponding to that trend sub-model. Then, the fitting objective is set to minimize the sum of squared errors between all observed values ​​and the model output value. Subsequently, the slope and intercept parameters corresponding to the minimum sum of squared errors are solved to obtain the fitting result of that trend sub-model. After performing the above fitting process on the image contrast trend sub-model, temperature fluctuation trend sub-model, and door opening angle change rate trend sub-model, a continuous trend model is obtained.

[0124] It should be noted that for any sampling time, the model output value is first calculated based on the slope parameter, intercept parameter, and time sequence number of the corresponding trend sub-model. Then, the model output value is subtracted from the actual observed value of the corresponding continuous feature value to obtain the fitting residual corresponding to that sampling time. Subsequently, the fitting residuals corresponding to all sampling times are squared and summed to obtain the residual sum of squares corresponding to the trend sub-model. During the installation and debugging phase, historical normal samples, historical disturbance samples, and historical abnormal samples are collected, and the continuous feature values ​​corresponding to each historical sample are calculated according to the previous steps. Then, first-order linear fitting is performed on the image contrast difference value sequence, temperature fluctuation difference value sequence, and door opening angle change rate difference value sequence corresponding to each historical sample to obtain the slope parameter and residual sum of squares corresponding to each trend sub-model. Subsequently, the residual sums of squares corresponding to all historical samples are sorted according to their magnitude, and the 90th percentile value is taken as the residual threshold. Then, the absolute values ​​of the slope parameters corresponding to all historical samples are sorted according to their magnitude, and the 10th percentile value is taken as the zero slope judgment threshold to obtain the trend judgment parameters.

[0125] It should be noted that for any trend sub-model, when its residual sum of squares is greater than the residual threshold, the trend corresponding to the trend sub-model is determined to be a fluctuating trend; when its residual sum of squares is less than or equal to the residual threshold and the slope parameter is greater than the zero slope determination threshold, the trend corresponding to the trend sub-model is determined to be an upward trend; when its residual sum of squares is less than or equal to the residual threshold and the slope parameter is less than the negative zero slope determination threshold, the trend corresponding to the trend sub-model is determined to be a downward trend; when its residual sum of squares is less than or equal to the residual threshold and the slope parameter is between the negative zero slope determination threshold and the zero slope determination threshold, the trend corresponding to the trend sub-model is determined to be a stationary trend.

[0126] The trends corresponding to the image contrast trend model, temperature fluctuation trend model, and door opening angle change rate trend model are sequentially combined according to the order of image contrast features, temperature fluctuation values, and door opening angle change rate to obtain the trend of the preceding time-domain data. The trends corresponding to the image contrast trend model, temperature fluctuation trend model, and door opening angle change rate trend model in the preceding time-domain data are read respectively, and the slope parameters and residual sums of squares corresponding to these sub-models are read respectively. Subsequently, the absolute value of the slope parameter is determined as the trend amplitude of the corresponding trend sub-model, and the residual sum of squares is determined as the fluctuation residual value of the corresponding trend sub-model.

[0127] It is worth noting that during the installation and debugging phase, historical normal samples, historical disturbance samples, and historical abnormal samples were collected, and the change trend, slope parameter, and residual sum of squares corresponding to each historical sample were calculated according to the previous steps. Then, the trend amplitude sets corresponding to the image contrast trend sub-model, the temperature fluctuation trend sub-model, and the door opening angle change rate trend sub-model were extracted respectively, and sorted in descending order. The 90th percentile value of each was taken as the image contrast fluctuation threshold, temperature fluctuation threshold, and door opening angle change rate fluctuation threshold. Subsequently, the image contrast fluctuation threshold, temperature fluctuation threshold, and door opening angle change rate fluctuation threshold were combined to obtain the preset fluctuation threshold.

[0128] Specifically, the trend amplitude corresponding to the image contrast trend sub-model is compared with the image contrast fluctuation threshold, the trend amplitude corresponding to the temperature fluctuation trend sub-model is compared with the temperature fluctuation threshold, and the trend amplitude corresponding to the door opening angle change rate trend sub-model is compared with the door opening angle change rate fluctuation threshold. When the trend amplitude corresponding to a certain trend sub-model is less than or equal to the corresponding fluctuation threshold, the perturbation degree corresponding to that trend sub-model is recorded as zero. When the trend amplitude corresponding to a certain trend sub-model is greater than the corresponding fluctuation threshold, the ratio of the trend amplitude corresponding to that trend sub-model to the corresponding fluctuation threshold is determined as the perturbation degree corresponding to that trend sub-model. The maximum value among the image contrast perturbation degree, temperature fluctuation perturbation degree, and door opening angle change rate perturbation degree is taken as the trend perturbation degree corresponding to the current backtracking period.

[0129] The trend disturbance degree corresponding to each historical sample is extracted, and the optimal correction ratio determined by iterating through the correction ratios during the installation and debugging phase is read. The optimal correction ratio is the proportion that minimizes the error between the corrected fire door operating state and the actual state of the historical sample. Subsequently, the trend disturbance degree corresponding to all historical samples is divided into multiple disturbance intervals according to their magnitude, and the arithmetic mean of all optimal correction ratios within each disturbance interval is calculated. A mapping relationship is established between each disturbance interval and its corresponding arithmetic mean to obtain an adjustment parameter mapping table.

[0130] Read the adjustment parameter mapping table and compare the current trend disturbance level with each disturbance interval in the adjustment parameter mapping table one by one; when the trend disturbance level falls into a certain disturbance interval, read the correction ratio value corresponding to the disturbance interval and determine the correction ratio value as the adjustment parameter.

[0131] In step S16, the verification consistency result is corrected using the adjustment parameters, and the real-time monitoring data and preceding time-domain data are weighted, fused, and delay-compensated to obtain the fire door's operating status, including:

[0132] The alarm priority score in the verification consistency result is corrected according to the adjustment parameters to obtain the corrected status value. The first weight coefficient corresponding to the real-time monitoring data is determined according to the corrected status value, and the second weight coefficient corresponding to the preceding time domain data is determined.

[0133] The first weighting coefficient is used to weight the image contrast features, temperature fluctuation values ​​and door opening angle change rate in the real-time monitoring data. The second weighting coefficient is used to weight the image contrast features, temperature fluctuation values ​​and door opening angle change rate in the preceding time domain data. The weighted real-time monitoring data and the weighted preceding time domain data are then fused to obtain a fused monitoring sequence.

[0134] Extract the response time delay corresponding to the fused monitoring sequence, and generate a delay compensation value based on the response time delay.

[0135] The timing of the fusion monitoring sequence is reconstructed based on the delay compensation value, and the operating status of the fire door is determined based on the amplitude change of the reconstructed fusion monitoring sequence.

[0136] It is worth noting that when the verification consistency result is consistent, the consistency coefficient is set to one, and the correction ratio value corresponding to the adjustment parameter is read. The correction ratio value is a value between zero and one. The alarm priority score is multiplied by the consistency coefficient, and then multiplied by one minus the correction ratio value to obtain the correction status value. When the verification consistency result is inconsistent, the consistency coefficient is set to zero, and the correction status value is set to zero.

[0137] Specifically, first, all alarm priority scores in the anomaly severity quantification table are read, and the highest alarm priority score is taken as the weight normalization base value. Then, the corrected state value is divided by the weight normalization base value to obtain the initial weight coefficient. When the initial weight coefficient is less than zero, the first weight coefficient is determined to be zero. When the initial weight coefficient is greater than one, the first weight coefficient is determined to be one. Otherwise, the initial weight coefficient is determined to be the first weight coefficient.

[0138] It is worth noting that the second weighting coefficient is obtained by subtracting the first weighting coefficient from one. This makes the sum of the first and second weighting coefficients equal to one. The image contrast features, temperature fluctuation values, and door opening angle change rates in the real-time monitoring data are multiplied by the first weighting coefficient to obtain the weighted real-time monitoring data. Then, the image contrast features, temperature fluctuation values, and door opening angle change rates at each sampling time in the preceding time-domain data are multiplied by the second weighting coefficient to obtain the weighted preceding time-domain data.

[0139] It should be noted that the current acquisition timestamp corresponding to the real-time monitoring data is read, and the data corresponding to the target sampling time adjacent to the current acquisition timestamp is extracted from the weighted preceding time domain data. When there is a sampling time with the same current acquisition timestamp in the weighted preceding time domain data, the data corresponding to that sampling time is directly determined as the target preceding data. When there is no sampling time with the same current acquisition timestamp in the weighted preceding time domain data, the data corresponding to the two sampling times located before and after the current acquisition timestamp with the closest time interval are extracted, and the arithmetic mean of the two is determined as the target preceding data, thus obtaining the target preceding data.

[0140] Specifically, the weighted real-time monitoring data and the target preceding data are fused. Specifically, the image contrast features in the weighted real-time monitoring data are added to the image contrast features in the target preceding data to obtain fused image contrast features; the temperature fluctuation values ​​in the weighted real-time monitoring data are added to the temperature fluctuation values ​​in the target preceding data to obtain fused temperature fluctuation values; and the door opening angle change rate in the weighted real-time monitoring data is added to the door opening angle change rate in the target preceding data to obtain fused door opening angle change rate, thus obtaining fused feature data. The fused image contrast features, fused temperature fluctuation values, and fused door opening angle change rate are then combined sequentially according to their order of arrangement, and the fused feature data corresponding to each acquisition timestamp are arranged chronologically to obtain the fused monitoring sequence.

[0141] It should be noted that for any fusion feature record, the time difference between the current acquisition timestamp and the corresponding sampling time of the target preceding data is first calculated, and then the time difference is multiplied by the second weighting coefficient to obtain the response time delay corresponding to the fusion feature record. After performing the above processing on all fusion feature records in the fusion monitoring sequence, the response time delay sequence is obtained.

[0142] Sort all response time delays in the response time delay sequence from largest to smallest, and take the median value as the delay compensation value corresponding to the current fusion monitoring sequence; when the number of records in the fusion monitoring sequence is less than three, directly take the response time delay corresponding to the current time as the delay compensation value to obtain the delay compensation value. For each fusion feature record in the fusion monitoring sequence, the delay compensation value is subtracted from the current acquisition timestamp to obtain the reconstruction timestamp. If the reconstruction timestamp is earlier than the earliest acquisition timestamp of the monitoring data, the earliest acquisition timestamp is used as the reconstruction timestamp. The fusion feature records are then rearranged in ascending order of reconstruction timestamps to form the reconstructed fusion monitoring sequence. When the interval between the reconstruction timestamps of two adjacent fusion feature records is greater than 1.5 times the original sampling period, it is determined as a sampling gap. A reconstruction sampling time is inserted at the gap position according to the original sampling period, and the interpolation results of image contrast features, temperature fluctuation values, and door opening angle change rate at the reconstruction sampling time are calculated using linear interpolation. For cases with long gap intervals, the linear interpolation results are used as approximate estimates for subsequent state determination. Subsequently, the interpolation results are written into the corresponding reconstruction sampling time to obtain a temporally continuous reconstructed fusion monitoring sequence.

[0143] It should be noted that during the installation and commissioning phase, samples of static stable operation, opening change, closing change, and abnormal disturbance of fire doors were collected, and the corresponding reconstructed and fused monitoring sequences were obtained according to the previous steps. Then, adjacent amplitude changes of image contrast features, adjacent amplitude changes of temperature fluctuation values, and adjacent amplitude changes of door opening angle change rate were extracted from each type of sample and sorted according to the size pattern.

[0144] The following thresholds are used to determine the operating status: The 95th percentile of the absolute value of adjacent amplitude changes in the door opening angle change rate in the static stable operation sample is used as the static stable angle threshold. The 5th percentile of adjacent amplitude changes in the door opening angle change rate in the opening change sample is used as the opening change threshold. The 5th percentile of adjacent amplitude changes in the door opening angle change rate in the closing change sample is used as the closing change threshold. The 5th percentile of adjacent amplitude changes in the temperature fluctuation value in the abnormal disturbance sample is used as the temperature disturbance threshold. The 5th percentile of adjacent amplitude changes in the image contrast feature in the abnormal disturbance sample is used as the image disturbance threshold. During installation and commissioning, the static stable angle threshold should be less than the opening change threshold and less than the closing change threshold to avoid status determination conflicts. The 5th percentile of adjacent amplitude changes in the temperature fluctuation value in the abnormal disturbance sample is used as the temperature disturbance threshold. The 5th percentile of adjacent amplitude changes in the image contrast feature in the abnormal disturbance sample is used as the image disturbance threshold.

[0145] The differences in image contrast features, temperature fluctuation values, and door opening angle change rates at adjacent sampling times in the reconstructed fused monitoring sequence are calculated separately. The absolute values ​​of these differences are taken as the adjacent amplitude changes in image contrast features, temperature fluctuation values, and door opening angle change rates, respectively. A judgment is then made: when the adjacent amplitude changes in door opening angle change rates are less than or equal to a static steady-state angle threshold, the adjacent amplitude changes in temperature fluctuation values ​​are less than a temperature disturbance threshold, and the adjacent amplitude changes in image contrast features are less than an image disturbance threshold, the fire door's operating state is determined to be a static steady-state operating state.

[0146] When the adjacent amplitude changes of the door opening angle change rate are greater than or equal to the opening change threshold, and the door opening angle change rate corresponding to the current sampling time in the reconstructed fusion monitoring sequence is positive, the fire door's operating state is determined to be in the opening change state; when the adjacent amplitude changes of the door opening angle change rate are greater than or equal to the closing change threshold, and the door opening angle change rate corresponding to the current sampling time in the reconstructed fusion monitoring sequence is negative, the fire door's operating state is determined to be in the closing change state.

[0147] When the adjacent amplitude changes of temperature fluctuation value are greater than or equal to the temperature disturbance threshold, or the adjacent amplitude changes of image contrast feature are greater than or equal to the image disturbance threshold, the operating status of the fire door is determined to be an abnormal disturbance state, and the operating status of the fire door is obtained.

[0148] In step S17, the operating status of the fire door is continuously assessed, an abnormal status identifier is generated, and an alarm message is sent, including:

[0149] Extract the operating status of the fire door corresponding to multiple consecutive sampling times, and map the operating status of the fire door to a timestamp sequence;

[0150] The time intervals between adjacent anomaly assessment results in the timestamp sequence are accumulated to obtain the anomaly duration. The anomaly duration is compared with a preset duration threshold. When the anomaly duration is greater than the preset duration threshold, the anomaly status identifier is generated.

[0151] An alarm message is generated based on the abnormal status identifier, the time of the abnormality, and the operating status of the fire door, and the alarm message is sent to the monitoring terminal according to a preset transmission protocol.

[0152] It is worth noting that during the installation and commissioning phase, historical normal samples are extracted from the labeled normal operation samples and slightly disturbed samples, and historical abnormal samples are extracted from the labeled obvious abnormal samples and high-risk abnormal samples. The number of sampling points that continuously maintain abnormal operating states in each historical sample is counted, and all the continuously maintained sampling points are sorted from largest to smallest, with the 95th percentile value used as the continuous sampling point threshold. Subsequently, during the operation phase, historical operating state records are retrieved one by one from the current acquisition timestamp backward, where the original sampling period is the uniform time interval for sensor data acquisition. When the acquisition time interval between two adjacent operating state records is not greater than twice the original sampling period, the two operating state records are determined as continuous sampling records. When the number of continuous sampling records reaches the continuous sampling point threshold, the operating state of the fire door corresponding to the continuous sampling record is extracted to obtain the operating state of the fire door corresponding to multiple consecutive sampling times.

[0153] The operating status of fire doors corresponding to multiple consecutive sampling times is mapped to a timestamp sequence. This is achieved by reading the operating status of fire doors and their corresponding collection timestamps one by one, arranging them in ascending order of collection timestamps; then establishing a one-to-one correspondence between each collection timestamp and the corresponding fire door operating status, forming a timestamp sequence composed of collection timestamps and fire door operating statuses.

[0154] Specifically, first, the timestamps corresponding to two adjacent operating status records are read, and the time difference between the timestamps of the later and earlier operating status records is calculated. When the operating status of the fire door corresponding to two adjacent operating status records is any of the abnormal states—abnormal disturbance, open / closed change, or closed change—the time difference is included in the cumulative duration. When there is a non-abnormal state in two adjacent operating status records, the current accumulation process is terminated. After performing the above accumulation on all consecutive adjacent abnormal assessment results, the duration of the abnormality is obtained.

[0155] The duration of each instance where the fire door's operational status was briefly deemed abnormal due to environmental disturbances was extracted from all historical normal samples. These instances were then sorted by duration, and the 99th percentile was used as the duration threshold. This threshold corresponds to the upper limit of the normal fluctuation duration. When the abnormal duration is less than or equal to the duration threshold, the abnormal state corresponding to the current abnormal assessment result is defined as an instantaneous abnormal state; when the abnormal duration exceeds the duration threshold, the abnormal state corresponding to the current abnormal assessment result is defined as a continuous abnormal state.

[0156] It should be noted that the process reads the time of occurrence of the anomaly, the duration of the anomaly, and the current operating status of the fire door corresponding to the current continuous anomaly assessment result. These three parameters are then combined according to a preset field order to obtain an anomaly status identifier. The time of occurrence of the anomaly, the duration of the anomaly, and the operating status of the fire door are extracted from the anomaly status identifier. The fire door operating status is used as the anomaly category information, and combined with the time of occurrence of the anomaly according to a preset field order to obtain alarm information. This alarm information is then encapsulated as an MQTT message and published to the monitoring terminal through the gateway node.

[0157] In summary, this invention discloses a remote intelligent monitoring and alarm method for fire doors based on the Internet of Things, which solves the technical problem that existing technologies are unable to accurately identify and reliably alarm abnormal states of fire doors under complex environmental disturbances and wireless network transmission conditions, and realizes stable monitoring and abnormal early warning of fire door operation status.

[0158] Reference Figure 2 The second embodiment of the present invention provides a remote intelligent monitoring and alarm system for fire doors based on the Internet of Things, comprising:

[0159] Secondly, the present invention provides a remote intelligent monitoring and alarm system for fire doors based on the Internet of Things, comprising:

[0160] The data processing module is used to acquire image data, temperature data, and door opening angle data through wireless networking, and to preprocess them to obtain a monitoring dataset;

[0161] The feature processing module is used to perform Gaussian smoothing and normalization calibration on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence and a calibration angle sequence, and to perform spatial coverage verification on the smoothed image sequence, the calibration temperature sequence and the calibration angle sequence to obtain a multi-dimensional feature set;

[0162] The vector construction module is used to extract image contrast features, temperature fluctuation values ​​and door opening angle change rate from the multi-dimensional feature set to construct a comprehensive environmental state vector.

[0163] The anomaly detection module is used to calculate the difference between the comprehensive environmental state vector and the preset baseline state interval to generate an alarm priority score. If the alarm priority score is in the preset high-risk interval, the comprehensive environmental state vector is compared for consistency to obtain a verification consistency result.

[0164] An interference identification module is used to extract preceding time-domain data based on the verification consistency result, perform trend analysis on the preceding time-domain data, and generate adjustment parameters.

[0165] The status assessment module is used to correct the verification consistency result using the adjustment parameters, and to perform weighted fusion and delay compensation on the real-time monitoring data and the preceding time domain data to obtain the operating status of the fire door.

[0166] The abnormal alarm module is used to continuously judge the operating status of the fire door, generate an abnormal status identifier, and send alarm information.

[0167] It should be noted that the IoT-based remote intelligent monitoring and alarm system for fire doors provided in this embodiment of the invention is used to execute all the process steps of the IoT-based remote intelligent monitoring and alarm method for fire doors in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0168] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0169] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A remote intelligent monitoring and alarm method for fire doors based on the Internet of Things, characterized in that it includes: Image data, temperature data, and door opening angle data are acquired through wireless networking and preprocessed to obtain a monitoring dataset. Gaussian smoothing and normalization calibration are performed on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence, and a calibration angle sequence. Spatial coverage verification is performed on the smoothed image sequence, the calibration temperature sequence, and the calibration angle sequence to obtain a multidimensional feature set. Image contrast features, temperature fluctuation values, and door opening angle change rate are extracted from the multidimensional feature set to construct a comprehensive environmental state vector. The difference between the comprehensive environmental state vector and the preset baseline state interval is calculated to generate an alarm priority score. If the alarm priority score is in the preset high-risk interval, the comprehensive environmental state vector is compared for consistency to obtain a verification consistency result. Based on the consistency verification results, extract the preceding time-domain data, perform trend analysis on the preceding time-domain data, and generate adjustment parameters; The verification consistency result is corrected using the adjustment parameters, and the real-time monitoring data and the preceding time domain data are weighted, fused, and delayed to obtain the operating status of the fire door. The system continuously assesses the operating status of the fire door, generates an abnormal status identifier, and sends an alarm message.

2. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things as described in claim 1, characterized in that, The process involves acquiring image data, temperature data, and door opening angle data via wireless networking, and preprocessing this data to obtain a monitoring dataset, including: The system receives image data from the image sensor, temperature data from the temperature sensor, and door opening angle data from the angle sensor via a wireless communication network. The image data, temperature data, and door opening angle data are each assigned a collection timestamp, and then rearranged in chronological order. The rearranged image data, temperature data, and door opening angle data are subjected to abnormal null value removal and data format conversion to obtain preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data. The preprocessed image data, preprocessed temperature data, and preprocessed door opening angle data are time-aligned based on the acquisition timestamp and then structured and encapsulated to obtain the monitoring dataset.

3. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things as described in claim 2, characterized in that, The data in the monitoring dataset are subjected to Gaussian smoothing and normalization calibration to obtain a smoothed image sequence, a calibration temperature sequence, and a calibration angle sequence. Spatial coverage verification is then performed on the smoothed image sequence, the calibration temperature sequence, and the calibration angle sequence to obtain a multi-dimensional feature set, including: The preprocessed image data, the preprocessed temperature data, and the preprocessed door opening angle data are separated from the monitoring dataset, and the preprocessed image data is Gaussian smoothed to obtain a smoothed image sequence. The preprocessed temperature data and the preprocessed door opening angle data are separated from the monitoring dataset, and the preprocessed temperature data and the preprocessed door opening angle data are normalized respectively to obtain the calibration temperature sequence and the calibration angle sequence. Read the sensor distribution coordinates, perform distribution statistics according to a preset spatial grid to form a sensor density distribution map, and perform matrix processing on the sensor density distribution map based on a preset density threshold to obtain a spatial coverage matrix. The multidimensional feature set is obtained by removing invalid data at the edges of the smoothed image sequence, the calibration temperature sequence, and the calibration angle sequence using the spatial coverage matrix.

4. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things as described in claim 1, characterized in that, The step of extracting image contrast features, temperature fluctuation values, and door opening angle change rate from the multi-dimensional feature set to construct a comprehensive environmental state vector includes: The pixel grayscale values ​​of each frame of the image in the multidimensional feature set are statistically analyzed, and the image contrast features are determined based on the standard deviation of the pixel grayscale values. The temperature difference between adjacent sampling times is calculated using the calibration temperature sequence in the multidimensional feature set to obtain the temperature fluctuation value. Calculate the angle difference between adjacent sampling times in the calibration angle sequence of the multidimensional feature set, and determine the door opening angle change rate based on the angle difference and the interval between adjacent sampling times; The image contrast features, temperature fluctuation values, and door opening angle change rate are concatenated and mapped to a preset state vector space to obtain the comprehensive environmental state vector.

5. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things according to claim 1, characterized in that, The step involves calculating the difference between the integrated environmental state vector and a preset baseline state interval to generate an alarm priority score. If the alarm priority score falls within a preset high-risk interval, a consistency comparison is performed on the integrated environmental state vector to obtain a consistency verification result, including: Extract the center point vector and boundary threshold in the preset baseline state interval, calculate the dimension-by-dimensional difference between the comprehensive environmental state vector and the center point vector, and normalize the difference in each dimension to obtain the feature deviation value set. The abnormality level is determined by comparing the set of feature deviation values ​​with the boundary threshold, and the abnormality level is processed according to the preset abnormality level quantification table to generate an alarm priority score. If the alarm priority score is in a preset high-risk range, the network transmission log for the corresponding time period is read and the integrity of the feature deviation value set is verified based on the network transmission log to obtain the transmission status result. Extract the source-side attached checksum corresponding to the transmission status result, generate the receiver-side checksum based on the transmission status result, and perform a consistency comparison between the source-side attached checksum and the receiver-side checksum to obtain a consistency result.

6. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things according to claim 1, characterized in that, Based on the consistency verification results, prior time-domain data is extracted, trend analysis is performed on the prior time-domain data, and adjustment parameters are generated, including: The time of an anomaly occurrence is determined based on the verification consistency result, the time of an anomaly occurrence is set as the endpoint, and the backtracking period corresponding to the time of an anomaly occurrence is determined. Image data, temperature data, and door opening angle data are extracted from the backtracking period. Based on the image data, temperature data, and door opening angle data, the image contrast features, temperature fluctuation values, and door opening angle change rates corresponding to each sampling time are calculated to obtain the preceding time domain data. The preceding time-domain data at adjacent sampling times are subjected to difference calculations to obtain continuous feature values; A continuous trend model is constructed using the continuous feature values, and the changing trend of the preceding time domain data is determined based on the continuous trend model. Based on the changing trend and the preset fluctuation threshold, the degree of trend disturbance is determined, and the adjustment parameters are generated based on the degree of trend disturbance.

7. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things according to claim 1, characterized in that, The verification consistency result is corrected using the adjustment parameters, and the real-time monitoring data and preceding time-domain data are weighted, fused, and delay-compensated to obtain the fire door's operating status, including: The alarm priority score in the verification consistency result is corrected according to the adjustment parameters to obtain the corrected status value. The first weight coefficient corresponding to the real-time monitoring data is determined according to the corrected status value, and the second weight coefficient corresponding to the preceding time domain data is determined. The first weighting coefficient is used to weight the image contrast features, temperature fluctuation values ​​and door opening angle change rate in the real-time monitoring data. The second weighting coefficient is used to weight the image contrast features, temperature fluctuation values ​​and door opening angle change rate in the preceding time domain data. The weighted real-time monitoring data and the weighted preceding time domain data are then fused to obtain a fused monitoring sequence. Extract the response time delay corresponding to the fused monitoring sequence, and generate a delay compensation value based on the response time delay. The timing of the fusion monitoring sequence is reconstructed based on the delay compensation value, and the operating status of the fire door is determined based on the amplitude change of the reconstructed fusion monitoring sequence.

8. The method for remote intelligent monitoring and alarm of fire doors based on the Internet of Things according to claim 6, characterized in that, The operating status of the fire door is continuously assessed, an abnormal status indicator is generated, and alarm information is sent, including: Extract the operating status of the fire door corresponding to multiple consecutive sampling times, and map the operating status of the fire door to a timestamp sequence; The time intervals between adjacent anomaly assessment results in the timestamp sequence are accumulated to obtain the anomaly duration. The anomaly duration is compared with a preset duration threshold. When the anomaly duration is greater than the preset duration threshold, the anomaly status identifier is generated. An alarm message is generated based on the abnormal status identifier, the time of the abnormality, and the operating status of the fire door, and the alarm message is sent to the monitoring terminal according to a preset transmission protocol.

9. A remote intelligent monitoring and alarm system for fire doors based on the Internet of Things, characterized in that, include: The data processing module is used to acquire image data, temperature data, and door opening angle data through wireless networking, and to preprocess them to obtain a monitoring dataset; The feature processing module is used to perform Gaussian smoothing and normalization calibration on the data in the monitoring dataset to obtain a smoothed image sequence, a calibration temperature sequence and a calibration angle sequence, and to perform spatial coverage verification on the smoothed image sequence, the calibration temperature sequence and the calibration angle sequence to obtain a multi-dimensional feature set; The vector construction module is used to extract image contrast features, temperature fluctuation values ​​and door opening angle change rate from the multi-dimensional feature set to construct a comprehensive environmental state vector. The anomaly detection module is used to calculate the difference between the comprehensive environmental state vector and the preset baseline state interval to generate an alarm priority score. If the alarm priority score is in the preset high-risk interval, the comprehensive environmental state vector is compared for consistency to obtain a verification consistency result. An interference identification module is used to extract preceding time-domain data based on the verification consistency result, perform trend analysis on the preceding time-domain data, and generate adjustment parameters. The status assessment module is used to correct the verification consistency result using the adjustment parameters, and to perform weighted fusion and delay compensation on the real-time monitoring data and the preceding time domain data to obtain the operating status of the fire door. The abnormal alarm module is used to continuously judge the operating status of the fire door, generate an abnormal status identifier, and send alarm information.

Images