Elevator environment abnormity monitoring method and system

Abstract

The invention discloses an elevator environment abnormity monitoring method and system, and the method comprises the steps: collecting environment data in a lift car in real time through a multi-mode sensor array, and generating a dynamic environment characteristic spectrum; based on the dynamic environment characteristic spectrum, outputting an environment health index with high confidence; generating a partition abnormal probability thermodynamic diagram according to the environmental health index and the passenger density detection data; based on the partition anomaly probability thermodynamic diagram, the real-time environment parameters, the partition anomaly probability thermodynamic diagram and risk early warning prompts are displayed on a car display screen in an overlapped mode; and according to the duration time and the spatial diffusion trend of the partition abnormal probability thermodynamic diagram, an environment deterioration early warning signal is generated through a space-time prediction model, and when the level of the environment deterioration early warning signal reaches a preset signal level, a gradient alarm mechanism is triggered, and a building management system is linked to adjust an elevator scheduling strategy. According to the embodiment of the invention, the accuracy and response efficiency of elevator environment safety management can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of elevator monitoring, and particularly relates to a method and system for monitoring abnormal elevator environments. Background Art

[0002] With the popularization of high-rise buildings and the increasing frequency of elevator use, the environmental safety issues in elevator cars have received growing attention. Traditional elevator environment monitoring mainly relies on single temperature or gas sensors, suffering from problems such as insufficient monitoring dimensions and isolated data analysis, making it difficult to comprehensively evaluate the comprehensive environmental quality inside the car. In the prior art, although the monitoring system based on the Internet of Things can achieve multi-parameter collection, there are still obvious defects in data processing: on the one hand, the spatio-temporal correlation of multi-source sensor data is not fully utilized, resulting in limited accuracy of anomaly detection; on the other hand, the environmental anomaly warning mechanism lacks dynamic adaptability and cannot intelligently adjust the monitoring strategy according to the real-time risk level. In addition, the current systems mostly adopt fixed-threshold alarm methods, which are difficult to accurately predict the trend of environmental deterioration and have low linkage response efficiency with the building management system. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for monitoring abnormal elevator environments to solve the deficiencies in the prior art, and to improve the accuracy and response efficiency of elevator environmental safety management through multi-modal data fusion, dynamic risk assessment, and predictive warning.

[0004] An embodiment of the present application provides a method for monitoring abnormal elevator environments, the method comprising: Real-time collecting environmental data inside the car through a multi-modal sensor array, and using a spatio-temporal convolutional network to fuse the spatial distribution features and time series fluctuation features of the sensor data to generate a dynamic environmental feature map, wherein the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint data; Based on the dynamic environmental feature map, reconstructing and denoising an environmental state model through a generative adversarial network, wherein the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the environmental anomaly probability in combination with the historical normal environment database and outputs a high-confidence environmental health index; According to the environmental health index and the passenger density detection data, constructing a multi-modal anomaly scoring model, and using a graph neural network to associate the environmental parameter correlations between car areas to generate a heat map of partition anomaly probabilities, wherein the heat map of partition anomaly probabilities marks the positions of high-risk areas and the anomaly levels; Based on the heat map of partition anomaly probabilities, dynamically allocating monitoring resources through an adaptive attention mechanism, preferentially enhancing the sensor sampling frequency in high-risk areas, and superimposing and displaying real-time environmental parameters, the heat map of partition anomaly probabilities, and risk warning prompts on the car display screen; Based on the duration and spatial diffusion trend of the partition anomaly probability heat map, an environmental deterioration warning signal is generated through a spatio-temporal prediction model. When the level of the environmental deterioration warning signal reaches the preset signal level, the gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, forming a closed-loop monitoring response.

[0005] Optionally, the environmental data inside the car is collected in real time through a multi-modal sensor array, and the spatial distribution characteristics and time series fluctuation characteristics of the sensor data are fused by a spatio-temporal convolutional network to generate a dynamic environmental feature map. Among them, the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint data, including: According to the original environmental data stream collected by the multi-modal sensor array, a sliding time window is used to segment the temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint signals to generate time-aligned multi-modal data slices; Perform spatial position encoding on the multi-modal data slices, divide the car into grid regions and label the topological relationship of the sensor nodes to construct a spatio-temporal correlation tensor; Input the spatio-temporal correlation tensor into a three-dimensional convolutional-long short-term memory hybrid network, extract local region features through a spatial convolution kernel, and capture the fluctuation rules across time windows in combination with the time-axis LSTM to output a set of feature vectors that fuse spatio-temporal features; Based on the set of feature vectors, perform regional feature aggregation, and use the graph pooling algorithm to generate a dynamic environmental feature map covering the entire car. Among them, the graph node encodes the regional environmental state, and the edge weight represents the correlation of environmental parameters between regions.

[0006] Optionally, based on the dynamic environmental feature map, a denoising environmental state model is reconstructed through a generative adversarial network. Among them, the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the environmental anomaly probability in combination with the historical normal environmental database and outputs a high-confidence environmental health index, including: Input the dynamic environmental feature map into the generator network, and reconstruct the denoised environmental state map through a multi-scale convolutional layer with residual connections. The generator outputs the estimated values of environmental parameters after noise suppression; Input the estimated values of environmental parameters and the standard state map in the historical normal environmental database into the discriminator network, calculate the distribution difference degree between the two through spectral normalization processing, and output the environmental anomaly probability score; In the adversarial training stage, the generator reversely optimizes the parameters according to the anomaly probability feedback by the discriminator, uses the Wasserstein distance to constrain the authenticity of the generated data, and iteratively improves the denoising ability; Fuse the parameter deviation degrees of each node in the denoised environmental state map, calculate the comprehensive influence weights of temperature, humidity, CO2 concentration, and volatile organic compounds through the entropy weight method, and generate a multi-dimensional environmental health index; Normalize and calibrate the multi-dimensional environmental health index, and generate a high-confidence environmental health index in the range of 0-1 by combining the anomaly probability output by the discriminator.

[0007] Optionally, construct a multi-modal anomaly scoring model based on the environmental health index and passenger density detection data, use a graph neural network to associate the correlation of environmental parameters between car areas, and generate a heat map of zonal anomaly probabilities. Among them, the heat map of zonal anomaly probabilities marks the positions of high-risk areas and anomaly levels, including: Perform spatio-temporal alignment on the environmental health index and crowd density detection data based on cameras, and dynamically allocate environmental health weights and passenger density weights for different areas through an attention mechanism to generate a fusion weight matrix; Construct a graph neural network model, map the car areas to graph nodes, use the correlation of environmental parameters as edge attributes, and model the anomaly propagation path between areas through graph attention layers; Update the graph node features according to the fusion weight matrix, use a graph convolutional network to iterate the anomaly propagation signal, calculate the anomaly accumulation value of each node, and generate an initial anomaly score distribution; Perform spatial interpolation processing on the initial anomaly score, use a heat diffusion algorithm to simulate the propagation trend of anomalies in the car, and output a heat map of zonal anomaly probabilities that marks the positions of high-risk areas and anomaly levels.

[0008] Optionally, based on the heat map of zonal anomaly probabilities, dynamically allocate monitoring resources through an adaptive attention mechanism, preferentially enhance the sensor sampling frequency in high-risk areas, and superimpose and display real-time environmental parameters, the heat map of zonal anomaly probabilities, and risk warning prompts on the car display screen, including: Divide the monitoring priorities according to the anomaly levels in the heat map of zonal anomaly probabilities, and dynamically calculate the resource allocation coefficient for high-risk areas using an attention weight function. The value of the resource allocation coefficient has an exponential relationship with the anomaly level; Adjust the sampling strategy of the sensor array, start a super-resolution sampling mode for high-risk areas, and increase the sampling frequency of temperature / humidity sensors to 3 times the reference value; Embed a heat map overlay module in the rendering engine of the car display screen, and perform pixel-level fusion of real-time environmental parameters, the heat map of zonal anomaly probabilities, and the camera video stream through a transparency gradient algorithm; Construct a risk warning dynamic annotation system. When the anomaly level of a certain area exceeds the preset level, automatically superimpose a flashing warning box and voice prompt at the corresponding position.

[0009] Optionally, according to the duration and spatial diffusion trend of the partition anomaly probability heat map, an environmental deterioration warning signal is generated through a spatio-temporal prediction model. When the level of the environmental deterioration warning signal reaches a preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, forming a closed-loop monitoring response, including: Extract the spatio-temporal change characteristics of the partition anomaly probability heat map, model the duration, diffusion speed and adjacent area correlation strength of the anomaly area through a spatio-temporal graph convolutional network, and output a spatio-temporal feature vector; Input the spatio-temporal feature vector into a pre-trained LSTM-Transformer hybrid prediction model to predict the environmental deterioration trend within a preset number of minutes in the future and generate a probabilistic warning signal; Design a three-level gradient alarm mechanism. When the probabilistic warning signal reaches level one, an audible and visual warning in the car is triggered. When it reaches level two, the ventilation system is linked to enhance ventilation. When it reaches level three, an elevator emergency dispatching request is sent to the building management system; Establish a closed-loop verification mechanism, compare the actual environmental change data with the prediction results, dynamically update the parameters of the hybrid prediction model through reinforcement learning, continuously optimize the warning accuracy, and form a closed-loop monitoring response.

[0010] Another embodiment of the present application provides an elevator environment anomaly monitoring system, and the system includes: A fusion module for real-time collecting environmental data in the car through a multi-modal sensor array, and using a spatio-temporal convolutional network to fuse the spatial distribution characteristics and time series fluctuation characteristics of the sensor data to generate a dynamic environmental feature map. Among them, the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds and voiceprint data; A reconstruction module for reconstructing and denoising the environmental state model based on the dynamic environmental feature map through a generative adversarial network. The generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the environmental anomaly probability in combination with the historical normal environment database and outputs a high-confidence environmental health index; An association module for constructing a multi-modal anomaly scoring model according to the environmental health index and the passenger density detection data, using a graph neural network to associate the environmental parameter correlations between car areas, and generating a partition anomaly probability heat map. The partition anomaly probability heat map calibrates the positions of high-risk areas and the anomaly levels; A display module for dynamically allocating monitoring resources based on the partition anomaly probability heat map through an adaptive attention mechanism, preferentially enhancing the sensor sampling frequency in high-risk areas, and superimposing and displaying real-time environmental parameters, the partition anomaly probability heat map and risk warning prompts on the car display screen; An early warning module, configured to generate an environmental deterioration early warning signal through a spatio-temporal prediction model according to the duration and spatial diffusion trend of the partition anomaly probability heat map. When the level of the environmental deterioration early warning signal reaches a preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, forming a closed-loop monitoring response.

[0011] Another embodiment of the present application provides a storage medium in which a computer program is stored. Wherein, the computer program is set to execute the method described in any one of the above when running.

[0012] Another embodiment of the present application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is set to run the computer program to execute the method described in any one of the above.

[0013] Compared with the prior art, an elevator environment anomaly monitoring method provided by the present invention collects environmental data in the car in real time through a multi-modal sensor array to generate a dynamic environmental feature map; based on the dynamic environmental feature map, outputs an environmental health index with high confidence; generates a partition anomaly probability heat map according to the environmental health index and passenger density detection data; based on the partition anomaly probability heat map, superimposes and displays real-time environmental parameters, partition anomaly probability heat map and risk warning prompts on the car display screen; according to the duration and spatial diffusion trend of the partition anomaly probability heat map, generates an environmental deterioration early warning signal through a spatio-temporal prediction model. When the level of the environmental deterioration early warning signal reaches a preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, thereby improving the accuracy and response efficiency of elevator environment safety management through multi-modal data fusion, dynamic risk assessment and predictive early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a hardware structure block diagram of a computer terminal for an elevator environment anomaly monitoring method provided by an embodiment of the present invention; Figure 2 It is a flow chart of an elevator environment anomaly monitoring method provided by an embodiment of the present invention; Figure 3 It is a structural diagram of an elevator environment anomaly monitoring system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention, and cannot be construed as a limitation of the present invention.

[0016] An embodiment of the present invention first provides an elevator environment anomaly monitoring method, which can be applied to an electronic device, such as a computer terminal, specifically, such as an ordinary computer, etc.

[0017] The following takes the operation on a computer terminal as an example for a detailed description thereof. Figure 1 The following is a hardware structure block diagram of a computer terminal for an elevator environment anomaly monitoring method provided by an embodiment of the present invention. As Figure 1 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory.

[0018] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can be caused to execute any elevator environment anomaly monitoring method.

[0019] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0020] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can be caused to execute any elevator environment anomaly monitoring method.

[0021] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0022] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0023] Refer to Figure 2 , an embodiment of the present invention provides an elevator environment anomaly monitoring method, which may include the following steps: S201, Real-time collect the environmental data inside the elevator car through a multi-modal sensor array, and use a spatio-temporal convolutional network to fuse the spatial distribution characteristics and time series fluctuation characteristics of the sensor data to generate a dynamic environmental feature map. Among them, the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint data; Specifically, according to the original environmental data stream collected by the multi-modal sensor array, a sliding time window can be used to segment the temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint signals to generate time-aligned multi-modal data slices; A multi-modal sensor array (model: EnvSense Pro) is installed inside the elevator car, which contains 16 nodes and is evenly distributed on the ceiling and side walls. The sensor collects temperature (unit: degree Celsius), humidity (percentage RH), CO2 concentration (unit: ppm), volatile organic compound VOC (unit: μg / m 3 ) and voiceprint (sampling rate 16kHz) per second. The original data stream needs to solve the time asynchrony problem: Sliding time window design: The window length is fixed at 5 seconds (covering 500 data points), and the step size is 0.5 seconds (that is, a new window is generated every 0.5 seconds).

[0024] Alignment rule: Take the CO2 concentration collection moment as the reference timestamp (because its response is the slowest), and the rest of the data is aligned through linear interpolation. For example, if a temperature sensor is delayed by 0.1 second, its equivalent value at the reference moment is calculated through interpolation.

[0025] Multi-modal slice generation: Each time window outputs a structured data packet (Data Packet), including: Temperature: The instantaneous values of 16 nodes (such as [23.5, 24.1,..., 22.8]); Voiceprint: The 5-second audio waveform is compressed into a 20-dimensional feature vector through Mel-Frequency Cepstral Coefficients (MFCC).

[0026] Example slice: In the slice with timestamp T = 12.5 seconds, the CO2 concentration at the car door = 850 ppm (the normal threshold is 1000 ppm), and the VOC at the rear wall = 120 μg / m 3 (exceeding the threshold of 80 μg / m 3 ).

[0027] Perform spatial position encoding on the multi-modal data slices, divide the elevator car into grid areas and mark the topological relationship of the sensor nodes, and construct a spatio-temporal correlation tensor; Spatial position encoding converts the physical position into a machine-readable topological structure: Carriage grid zoning: Divide the bottom surface of the carriage into a 5×3 grid (each grid is 0.4m×0.6m), and divide it into 2 layers vertically (height 1.2m / 2.4m) to form 30 three-dimensional units (5×3×2).

[0028] Sensor mapping: Each sensor is marked with the grid number it belongs to (for example, node 7 is located in the upper layer of grid B3).

[0029] Topological relationship construction: The adjacency matrix marks the spatial relationship between grids: Horizontal adjacency: Weight 1.0 (such as grids A1 and A2); Vertical adjacency: Weight 0.8 (such as the upper and lower layers of A1); Diagonal adjacency: Weight 0.5 (such as A1 and B2); Dynamic association: If the environmental parameters of two grids change synchronously for 3 consecutive seconds (such as temperature difference <0.5℃), the weight is additionally increased by 0.2.

[0030] Generation of spatio-temporal correlation tensor: Output a four-dimensional tensor (dimension: 30 grids × 5 seconds × 5 modalities × 3 topological weights): Modality dimension: Temperature, humidity, CO2, VOC, voiceprint characteristics; Topological dimension: Weights of horizontal / vertical / diagonal relationships.

[0031] Example: The temperature value of the upper layer of grid C2 is 24.3℃ at T = 12 seconds, and its topological weight with the adjacent grid B2 = 1.0.

[0032] Input the spatio-temporal correlation tensor into a three-dimensional convolutional-long short-term memory hybrid network, extract local region features through a spatial convolution kernel, combine the time-axis LSTM to capture the fluctuation law across time windows, and output a set of feature vectors that fuse spatio-temporal features; The three-dimensional convolutional-LSTM hybrid network (3D CNN-LSTM) is processed in two stages: Spatial feature extraction (3D CNN): The convolution kernel size is 3×3×2 (length × width × time layer), and the stride is 1×1×1, covering adjacent grids and time sequence segments.

[0033] Feature mapping example: Focus on the local high-temperature area (such as temperature difference >2℃ within a 3×3 grid), and output a feature map to mark abnormal hot spots (such as the response value of the B2 area of the grid is 0.85).

[0034] Use the leaky rectified linear unit (Leaky ReLU) activation function: Multiply the negative input by a coefficient of 0.01 to avoid gradient disappearance.

[0035] Temporal Pattern Capture (LSTM): Long Short-Term Memory units process the time dimension, with a memory capacity of 100 states per unit.

[0036] Modeling of Fluctuation Patterns: Input: Convolutional features of 5 consecutive time windows (2 windows per second, covering 2.5 seconds of history); Output: Identification of periodic changes (e.g., CO2 concentration rising by 50 ppm every 10 seconds) or mutation events (e.g., VOC soaring by 40 μg / m within 0.2 seconds). 3 )

[0037] Feature Vector Set Generation: Each grid outputs a 128-dimensional feature vector, including: Spatial features (64 dimensions): Aggregated values of local area parameters; Temporal features (64 dimensions): Encoding of fluctuation trends (e.g., "rising steadily", "violent oscillation").

[0038] Example: In the vector of grid A1, the spatial dimension includes a temperature gradient of 0.7 (high gradient), and the temporal dimension includes a CO2 acceleration flag of 1 (continuous growth).

[0039] Based on the feature vector set, regional feature aggregation is performed, and a dynamic environmental feature map covering the entire car domain is generated using the graph pooling algorithm. Among them, the map nodes encode the regional environmental states, and the edge weights represent the correlation of environmental parameters between regions.

[0040] The Graph Pooling algorithm aggregates scattered features into a global map: Graph Structure Construction: Nodes: 30 grid cells, with attributes being 128-dimensional feature vectors.

[0041] Edges: Topological weights (adjacency matrix values in step 2), reflecting the influence intensity between regions.

[0042] Feature Aggregation Mechanism: Node Feature Update: Calculate the weighted influence of adjacent nodes through the Graph Attention Network (GAT): The weight of grid A1 affected by B1 is 0.6, and the weight affected by C1 is 0.3.

[0043] Update formula: The new feature of A1 = original feature × 0.7 + feature of B1 × 0.6 × 0.3 + feature of C1 × 0.3 × 0.3.

[0044] Hierarchical Pooling: Cluster 30 nodes into 6 super nodes (e.g., front / rear / left / right / upper / lower), and use the Node Decimation algorithm to retain key features.

[0045] Dynamic environment feature map output: Node data: Summary of the environmental status of each supernode (such as "high temperature and high humidity in the rear").

[0046] Edge weight: Correlation coefficient of parameters between regions (0 - 1). For example, the temperature correlation between the front and rear is 0.92 (strong correlation), and the VOC correlation between the left and right sides is 0.35 (weak correlation).

[0047] Visualization map: Supernodes highlighted in red represent abnormal regions (such as the abnormal value of the upper - layer temperature node > 0.8), and the thickness of the connection lines represents the strength of the correlation.

[0048] This method synchronously collects environmental parameters such as temperature, humidity, gas concentration, and sound feature data through multi - type sensors distributed at different positions in the car. The spatio - temporal convolutional network is used to deeply process these data, which can not only capture the spatial correlation between sensor nodes but also analyze the law of each parameter changing over time. Finally, a feature map reflecting the overall environmental status of the car is formed. This fusion processing overcomes the limitations of single - sensor or single - moment data, realizes the full - dimensional perception and fusion of environmental data, and provides high - precision input features for subsequent anomaly detection. The joint analysis of spatio - temporal features can discover potential environmental anomaly patterns. For example, local high temperature accompanied by a specific voiceprint may indicate equipment failure, creating conditions for early warning.

[0049] S202, based on the dynamic environment feature map, reconstruct and denoise the environmental status model through a generative adversarial network. Among them, the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the environmental anomaly probability by combining with the historical normal environment database and outputs a high - confidence environmental health index; Specifically, the dynamic environment feature map can be input into the generator network, and the denoised environmental status map is reconstructed through the multi - scale convolutional layer with residual connections. The generator outputs the estimated values of environmental parameters after noise suppression; The generator network adopts a "deep convolutional architecture" (with 12 - layer depth) specifically designed to eliminate sensor noise. Its core is to process the input data through a "multi - scale convolutional layer with residual connections": Multi - scale convolution design: Small - scale convolution kernel (3×3 grid coverage area): Capture local mutations (such as an instantaneous temperature spike in a 0.4m×0.6m grid).

[0050] Medium - scale convolution kernel (5×5 grid area): Detect abnormal regional associations (such as synchronous humidity increase in 5 adjacent grids).

[0051] Large - scale convolution kernel (covering 1 / 2 of the car area): Identify global trends (such as overall CO2 accumulation in the rear of the car).

[0052] Residual connection mechanism: The output of each layer is superimposed on the original input (residual connection) to retain the background of the real data. For example, when the small-scale convolution misjudges the temperature noise of a certain grid (such as a 0.5°C fluctuation caused by air conditioner wind) as abnormal, the residual connection can reduce its weight to less than 0.2.

[0053] Denoising and reconstruction process: Input the dynamic environment feature map (30 supernode features), and after being refined layer by layer through the convolutional layer: Suppress instantaneous interference: Eliminate the sudden high-frequency noise (such as the mechanical sound of elevator operation) mis-triggered by the voiceprint sensor.

[0054] Correct drift error: Calibrate the long-term reading deviation (such as continuously being 3%RH higher) of the humidity sensor caused by condensation.

[0055] The output is the denoised environment state map, and each node contains optimized parameter values (such as the temperature of grid B2 is corrected from 24.8°C to 24.1°C).

[0056] Input the estimated environmental parameter values and the standard state map in the historical normal environment database into the discriminator network, calculate the distribution difference degree between the two through spectral normalization processing, and output the environmental anomaly probability score; The discriminator network acts as a "quality inspector" to compare the output of the generator with the historical normal data: Historical database construction: Store 10,000 hours of normal operation data (temperature 18 - 26°C, humidity 40 - 60%RH, CO2 < 1000 ppm, VOC < 80 μg / m 3 )

[0057] Standard state map: Classify and model according to seasons / time periods (such as the standard CO2 = 900 ± 50 ppm during the early morning rush hour in summer).

[0058] Spectral Normalization: Core function: Limit the singular value of the discriminator weight matrix to prevent the divergence of adversarial training.

[0059] Implementation process: Calculate the spectral norm of the weight matrix W, σ(W) = the largest singular value.

[0060] Normalization operation: Replace W with W / σ(W) to ensure that the discriminator satisfies Lipschitz Continuity.

[0061] Calculation of distribution difference degree: Use Jensen-Shannon Divergence (JSD) to quantify similarity: Input: Estimated environmental parameter values output by the generator (such as the temperature of grid C3 is 25.2°C) vs the historical mean value at the same location in the database (24.5°C).

[0062] Output: Difference degree score (0 - 1), for example, 0.7 indicates a high degree of anomaly.

[0063] Anomaly probability score: The difference degree score is mapped to a probability value through the Sigmoid function (such as JSD = 0.7 → anomaly probability 0.85).

[0064] During the adversarial training stage, the generator reversely optimizes the parameters according to the anomaly probability feedback by the discriminator, uses the Wasserstein distance to constrain the authenticity of the generated data, and iteratively improves the denoising ability; Adversarial training is continuously optimized through the game between the generator and the discriminator: Reverse optimization mechanism: The generator receives the anomaly probability output by the discriminator (such as 0.85), and updates the weights of the convolutional layer through the Backpropagation algorithm: If a certain convolutional kernel causes the anomaly probability of multiple grids to be > 0.8, its weight is reduced by 30%.

[0065] If a certain residual connection effectively suppresses false alarms (anomaly probability < 0.2), its weight is increased by 15%.

[0066] Wasserstein distance constraint: Core function: Replace the cross-entropy loss of the traditional GAN to solve the problem of unstable training.

[0067] Implementation principle: Calculate the Earth Mover's Distance between the generated data distribution P_g and the real data distribution P_r.

[0068] Constraint objective: Minimize the Wasserstein distance (target value < 0.05) to ensure that the generated data approximates the real distribution.

[0069] Example: When the generator misjudges high-temperature noise as a real fire, the Wasserstein distance rises to 0.12, triggering weight penalty.

[0070] Iterative improvement process: Each round of training contains 100 generator-discriminator games.

[0071] Termination condition: The Wasserstein distance < 0.05 and the anomaly probability error < 5% for 10 consecutive rounds.

[0072] Training result: The correlation coefficient between the denoised data and the real data reaches 0.98.

[0073] Fuse the parameter deviation degrees of each node in the environmental state map after denoising, calculate the comprehensive influence weights of temperature, humidity, CO2 concentration, and volatile organic compounds through the entropy weight method, and generate a multi-dimensional environmental health index; Entropy Weight Method objectively quantifies the influence of different environmental parameters: Calculation of parameter deviation degree: Definition: The absolute deviation between the current value and the historical reference value (e.g., the temperature in grid A1 is 24.8°C vs. the reference of 23.5°C → deviation degree of 1.3°C).

[0074] Normalization processing: The deviation degree is divided by the historical maximum deviation (e.g., the maximum temperature deviation is 5°C → normalized value of 0.26).

[0075] Entropy weight method weight allocation: Calculation of information entropy: The entropy value of temperature E_temp = -Σ(p_i × ln p_i), where p_i is the proportion of the temperature deviation degree of each grid.

[0076] The smaller the entropy value (e.g., 0.2 vs. 0.6), the greater the parameter variability and the higher the weight.

[0077] Weight formula: The temperature weight W_temp = (1 - E_temp) / Σ(1 - E_k) (k covers all 4 parameters).

[0078] Example: If E_temp = 0.2, E_hum = 0.5, E_co2 = 0.3, E_voc = 0.4, then: W_temp = (1 - 0.2) / [(1 - 0.2) + (1 - 0.5) + (1 - 0.3) + (1 - 0.4)] = 0.8 / 2.6 ≈ 0.31.

[0079] Synthesis of multi-dimensional health index: Node health index = Σ(parameter deviation degree × weight).

[0080] Example: For a certain grid with a temperature deviation degree of 0.3, humidity of 0.1, CO2 of 0.4, and VOC of 0.2 → health index = 0.3×0.31 + 0.1×0.19 + 0.4×0.27 + 0.2×0.23 = 0.28.

[0081] Normalize and calibrate the multi-dimensional environmental health index, and generate a high-confidence environmental health index in the range of 0-1 by combining the anomaly probability output by the discriminator.

[0082] The final health index needs to integrate objective data and the subjective evaluation of the discriminator: ‌Normalization and calibration‌: Linear scaling: Project the multi-dimensional health index onto the range of 0-1 (for example, the original value of 0.28 corresponds to the calibrated value of 0.35).

[0083] Non-linear correction: Use a ‌piecewise S-shaped curve‌ (Sigmoid Curve) to enhance the discrimination of extreme values: Health index < 0.2 → Output close to 0 (safe); Health index > 0.8 → Output close to 1 (high risk).

[0084] ‌Anomaly probability fusion‌: The anomaly probability output by the discriminator (such as 0.85) is used as the confidence coefficient: ‌Weighted formula‌: Final health index = calibrated index × 0.7 + anomaly probability × 0.3.

[0085] Example: Calibrated value 0.35 × 0.7 + 0.85 × 0.3 = 0.245 + 0.255 = 0.50.

[0086] ‌High-confidence output‌: Threshold classification: 0-0.3 (green, safe); 0.3-0.6 (yellow, warning); 0.6-1.0 (red, alert).

[0087] ‌Application example‌: When the VOC in the rear of the elevator exceeds the standard, the final index = 0.73 (red alert), triggering the generation of the next-step partition heat map.

[0088] Using the adversarial learning mechanism of the generative adversarial network, the generator is responsible for eliminating sensor noise and outlier interference and reconstructing the real environmental state; the discriminator identifies the deviation degree of the current environment from the normal mode by comparing the historical normal data distribution. Through continuous adversarial optimization, the two finally output a quantified environmental health score, accurately reflecting the car environment state, effectively solving the misjudgment problem caused by sensor noise and occasional interference, and improving the reliability of anomaly detection. The health index provides a standardized basis for subsequent decisions, realizing the transformation from raw data to actionable indicators.

[0089] S203. Based on the environmental health index and the passenger density detection data, construct a multi-modal anomaly scoring model, use a graph neural network to associate the correlation of environmental parameters between car areas, and generate a heat map of partition anomaly probabilities, where the heat map of partition anomaly probabilities marks the positions of high-risk areas and the anomaly levels. Specifically, the environmental health index and the crowd density detection data based on cameras can be aligned in time and space, and the environmental health weights and passenger density weights of different areas can be dynamically allocated through an attention mechanism to generate a fusion weight matrix. Time and space alignment is the basis for fusing multi-source data. A wide-angle camera (resolution 1920×1080, frame rate 30fps) installed on the top of the car captures the passenger distribution in real time, while the environmental health index comes from the high-confidence results output in the previous step (updated every 0.5 seconds). The alignment process is divided into three steps: Timestamp synchronization: Each frame of the camera is marked with a millisecond-level timestamp (such as 13:05:27.845), and the deviation from the generation time of the environmental health index should be less than 50 milliseconds. If the environmental index is updated at T = 27.890 seconds, then the camera frame closest to it in the interval T = 27.845 - 27.945 is selected (such as frame T = 27.892).

[0090] Spatial grid mapping: Divide the camera image into a 5×3 grid (corresponding one-to-one with the physical grid in the previous step), and use the YOLOv5 object detection algorithm (You Only Look Once, a real-time object recognition model) to count the number of passengers in each grid. For example, 3 people are detected in grid B2, and the passenger density value = 3 (the highest density threshold is set to 5 people / grid).

[0091] Attention weight assignment: Dynamic weight function: The environmental health weight (W_e) is determined by the health index of this grid: when the index > 0.6, W_e = 0.8; when the index < 0.3, W_e = 0.2.

[0092] The passenger density weight (W_p) is non-linearly related to the number of people: 1 person → 0.3, 3 people → 0.7, 5 people → 1.0.

[0093] Fusion rule: The weights of high-risk areas (such as areas with dense passengers and abnormal environment) are superimposed. For example, the health index of grid C3 is 0.65 (high anomaly) + density 4 people → W_e = 0.8, W_p = 0.9 → fusion weight = √(0.8×0.9) = 0.85.

[0094] Finally, a weight matrix of 30 grids (5×3×2) is output, storing the environmental weight and passenger density weight of each grid.

[0095] Build a graph neural network model, map the car area to graph nodes, use the environmental parameter correlation as edge attributes, and model the abnormal propagation path between regions through the graph attention layer; The core of the Graph Neural Network (GNN) is to transform the car into a topological network: Graph structure initialization: Nodes: 30 grid cells, and node features include environmental health index, passenger density weight, and historical anomaly frequency.

[0096] Edges: Based on the topological weights (adjacent grid association strength) in the previous step, superimpose the dynamic parameter correlation: If the temperature difference between two grids is <1°C for 5 consecutive seconds, the correlation weight increases by 0.1.

[0097] If frequent passenger movement causes interaction between regions (such as from A1 to B1), increase the movement path weight by 0.2.

[0098] Operation mechanism of the Graph Attention Layer (GAT): Each node calculates the influence of its neighbors: Grid A1 pays attention to adjacent B1, A2, and the lower layer of A1: Environmental weight of B1 is 0.7 → Influence score = 0.7 × Correlation weight 0.9 = 0.63; Environmental weight of A2 is 0.5 → Score = 0.5 × 0.6 = 0.3.

[0099] Weighted aggregation: New feature of A1 = (Original feature × 0.4 + Feature of B1 × 0.63 + Feature of A2 × 0.3).

[0100] Abnormal propagation modeling: The high-temperature area (grid C3) transfers heat parameters to adjacent grids, and the propagation intensity is controlled by the edge weight (such as the weight from C3 to B3 is 0.8, and the propagation efficiency is 80%).

[0101] Dynamic parameter update: Recalculate the edge attributes every 5 seconds to adapt to the topological changes caused by passenger movement.

[0102] Update the graph node features according to the fusion weight matrix, use the graph convolutional network to iterate the abnormal propagation signal, calculate the abnormal accumulation value of each node, and generate the initial abnormal score distribution; The Graph Convolutional Network (GCN) captures abnormal propagation through multiple layers of iteration: Node feature update: Input the fusion weight matrix of Step 1 (environmental weight + density weight) and concatenate it with the original node features: Grid B2 features = [health index 0.65, density weight 0.9, historical anomaly 0.3] → After concatenation, it becomes [0.65, 0.9, 0.3, W_e = 0.8, W_p = 0.9].

[0103] Update through two layers of GCN: The first layer outputs a 128-dimensional vector (extracting local anomaly features); The second layer outputs a 64-dimensional vector (fusing global propagation signals).

[0104] Anomaly propagation iteration: Iteration rule: Each node absorbs the anomaly values of adjacent nodes and transfers them with attenuation according to the edge weights: The initial anomaly score of grid C3 is 0.8 and it is transferred to B3 (weight 0.8): The received value by B3 = 0.8 × 0.8 = 0.64, and the self-score of B3 is 0.6 → After update = 0.6 + 0.64 = 1.24 (needs to be normalized).

[0105] Cumulative mechanism: If the grid score increases by > 0.1 for three consecutive rounds of iteration, a positive feedback loop is triggered (the score rises rapidly).

[0106] Initial anomaly score generation: Final score = node's own health index × 0.6 + neighbor propagation contribution × 0.4.

[0107] Example distribution: The score in the rear high-temperature area (C2, C3) is 0.75 - 0.92; the score in the front low-risk area (A1, A2) is 0.15 - 0.30; Output the score matrix of 30 grids (5×3), and the numerical range is 0 - 1.

[0108] Perform spatial interpolation on the initial anomaly score, use the heat diffusion algorithm to simulate the propagation trend of anomalies in the car, and output the zoned anomaly probability heat map calibrating the location of high-risk areas and the anomaly level.

[0109] Heat map generation needs to solve the mapping from discrete grids to continuous space: Spatial Interpolation: Bicubic Interpolation: Expand the 5×3 grid into a 50×30 pixel image to smoothly transition the score values.

[0110] The score of grid B2 is 0.8 → Assign the corresponding pixel area (x10 - 20, y20 - 30) a value of 80 (0.8 × 100); Adjacent grid A2 score is 0.3 → pixel gradient at the junction (e.g., value at coordinate (19, 21) = (0.8×0.7 + 0.3×0.3) = 0.65).

[0111] ‌Thermal Diffusion Algorithm: Physical simulation of abnormal propagation: Diffusion source: grids with an abnormal score > 0.7 are regarded as "heat sources" (e.g., C3 score is 0.92).

[0112] Diffusion equation: The score value is diffused to surrounding pixels every 0.1 seconds, and the attenuation coefficient = 0.05 / pixel distance.

[0113] Example: Diffusion from C3 (coordinate 25, 15) to B3 (coordinate 20, 15): Pixel distance = 5 → Attenuated score increment = 0.92×(0.05 / 5) = 0.0092; New score of B3 = original value 0.75 + 0.0092 = 0.7592.

[0114] ‌Dynamic trend visualization: The area of the red region (score > 0.7) expands over time, representing the spread of the anomaly; Arrows indicate the diffusion direction (e.g., spreading from back to front).

[0115] ‌Output of the partitioned anomaly probability heat map: ‌Color coding: Blue (0 - 0.3): Safe; Yellow (0.3 - 0.6): Warning; Red (0.6 - 1.0): High risk.

[0116] ‌Calibration of high - risk areas: When the continuous red area > 0.4㎡, it is marked as a high - risk area (e.g., the 1.2㎡ area at the rear); Anomaly level = average score × area of the region (e.g., 0.82×1.2 = 0.984 → Level 1 high risk).

[0117] ‌Application example: Due to passenger congestion at the rear of the elevator, CO2 accumulates, and in the generated heat map, the rear 1 / 3 is a red patch, with the level marked as "LV2".

[0118] Fuse environmental health data with the visually recognized passenger distribution information, analyze the mutual influence relationship of environmental parameters in different regions through a graph neural network. For example, the increase in CO2 in areas with a high density of passengers may affect adjacent regions. Finally, generate a heat map that intuitively shows the risk distribution, clearly mark the location and severity of abnormal areas, achieve precise positioning and grading of environmental anomalies, and provide a basis for targeted disposal. The heat map visually presents the risk distribution, facilitating the rapid identification of problem areas and improving the response efficiency.

[0119] S204. Based on the partition anomaly probability heat map, dynamically allocate monitoring resources through an adaptive attention mechanism, preferentially enhance the sensor sampling frequency in high-risk areas, and superimpose and display real-time environmental parameters, the partition anomaly probability heat map, and risk warning prompts on the car display screen; Specifically, the monitoring priority can be divided according to the anomaly level in the partition anomaly probability heat map, and the resource allocation coefficient of high-risk areas can be dynamically calculated using the attention weight function. The value of the resource allocation coefficient has an exponential relationship with the anomaly level; The system divides the car area into four levels of monitoring priority by parsing the color coding and numerical labels of the heat map: Priority division rules: Blue area (anomaly level 0 - 0.3): Marked as P4 level (lowest priority), and the resource allocation coefficient is fixed at 1.0 (i.e., the reference sampling frequency).

[0120] Yellow area (0.3 - 0.6): Marked as P3 level (medium attention), and the coefficient increases linearly (e.g., anomaly level 0.4 → coefficient 1.4).

[0121] Orange area (0.6 - 0.8): Marked as P2 level (high priority), triggering exponential calculation (e.g., level 0.7 → coefficient = 1.5^2 = 2.25).

[0122] Red area (>0.8): Marked as P1 level (highest risk), and the coefficient = 1.5^(level × 10). For example, when the anomaly level is 0.85 → coefficient = 1.5^8.5 ≈ 25.6 (but limited by the hardware upper limit to 3.0).

[0123] Example of the operation of the attention weight function: When the heat map shows that grid 2 in the rear of the car is red (anomaly level 0.88) and grid 4 on the side wall is orange (0.72): Rear resource allocation coefficient = 1.5^(0.88×10) ≈ 1.5^8.8 ≈ 28.7 → take 3.0 after hardware limiting; Side wall coefficient = 1.5^(0.72×10) = 1.5^7.2 ≈ 17.8 → take 3.0 after limiting; The coefficient is sent to the sensor control module through the CAN bus (Controller Area Network, in-vehicle communication protocol).

[0124] Adjust the sampling strategy of the sensor array, start the super-resolution sampling mode for high-risk areas, and increase the sampling frequency of temperature / humidity sensors to 3 times the reference value; Dynamic sampling strategy adjustment is divided into two stages: hardware reconfiguration and data flow optimization: Hardware reconfiguration: The default sampling frequency of the temperature / humidity sensor (model SHT40) is 1 Hz (once per second). When the resource allocation coefficient ≥ 2.0 is received: Switch to Turbo mode: By rewriting the I 2 C register (Inter-Integrated Circuit, two-wire communication protocol), the sampling rate is increased to 3 Hz.

[0125] Power consumption management: Only boost the power supply of the sensors in the high-risk grid (from 3.3V to 5V) to avoid a sharp increase in the energy consumption of the entire column of sensors.

[0126] Example: When the temperature sensor of Grid No. 2 at the rear is at P1 level, the sampling interval is compressed from 1000 milliseconds to 333 milliseconds, and the single-sampling time is shortened from 20 milliseconds to 5 milliseconds (high-speed mode).

[0127] Data flow optimization: High-frequency sampling generates three times the amount of data (e.g., the temperature value increases from 16 points per second to 48 points), and a dedicated buffer is required for temporary storage: An independent memory partition (with a capacity of 8KB) is opened up, and a ring buffer structure is used to store temporary data.

[0128] Data compression: Differential encoding (Delta Encoding) is performed on consecutive sampled values, and only the change amount is stored (e.g., a 0.1°C change in temperature is recorded as +01).

[0129] The Edge Computing Unit filters valid data in real time: Transient interferences (such as a 0.2°C instantaneous fluctuation caused by a passenger passing by) are eliminated, and continuous abnormal signals are retained.

[0130] Embed a heat map overlay module in the rendering engine of the car display screen, and perform pixel-level fusion of real-time environmental parameters, the heat map of the partition anomaly probability, and the camera video stream through a transparency gradient algorithm; Multi-source information fusion display relies on a 10-inch LCD display screen (resolution 1280×720) on the top of the car, and the rendering engine adopts a three-layer overlay architecture: Underlying video stream: The video stream of the camera (2 million pixels, fish-eye lens) is transmitted at 30 fps (Frames Per Second, frame rate), and after distortion correction, it is mapped to the bottom layer of the display screen.

[0131] Middle-layer heat map overlay: Alpha Blending algorithm: Each pixel of the heat map is assigned an alpha value (opacity value): Alpha = 0.3 (semi-transparent) for the blue area and Alpha = 0.8 (high coverage) for the red area.

[0132] Blending formula: The final pixel color = video pixel color × (1 - Alpha) + heat map pixel color × Alpha.

[0133] Example: The pixel of the rear seat in the video (RGB = 120, 80, 60) is overlaid with the pixel of the red heat map (RGB = 255, 0, 0), Alpha = 0.8 → The blended RGB = (120×0.2 + 255×0.8, 80×0.2 + 0×0.8, 60×0.2 + 0×0.8) = (222, 16, 12).

[0134] Dynamic update: The heat map is refreshed every 0.5 seconds and accelerated for rendering through the GPU (Graphics Processing Unit).

[0135] Top-level environmental parameters: Suspending and displaying key values at the four corners of the screen: Upper left corner: average car temperature (e.g., 24.3°C), humidity (52%RH); lower right corner: highest abnormal area level (e.g., "rear LV2"); the font changes color dynamically: normal value in green (temperature < 26°C), red and flashing when exceeding the threshold (CO2 > 1000 ppm).

[0136] Construct a dynamic annotation system for risk warning. When the abnormal level of a certain area exceeds the preset level, a flashing warning box and voice prompt will be automatically superimposed at the corresponding position.

[0137] The multi-modal warning system consists of visual annotation and voice broadcast: Visual annotation: Flashing warning box: Using SVG vector graphics (Scalable Vector Graphics), it is drawn in real time according to the coordinates of the high-risk areas of the heat map: LV1 warning (0.6 - 0.8): yellow square box, flashing at a frequency of 1 Hz (0.5 seconds on and 0.5 seconds off); LV2 and above (> 0.8): red octagon box, flashing at a high frequency of 2 Hz (0.25 seconds on / 0.25 seconds off).

[0138] Intelligent obstacle avoidance algorithm: When the selected area overlaps with the passenger's face (detected by face detection), it automatically offsets 50 pixels to avoid occlusion.

[0139] Voice prompt: Hierarchical broadcast content: LV1 triggers prerecorded voice: "Attention, local environment is abnormal" (volume 60dB); LV2 triggers TTS synthesis (Text-to-Speech): "Warning! High temperature and high risk at the rear" (volume 75dB).

[0140] Sound source orientation technology: Through the four-microphone array on the top of the car, the voice sound image is positioned in the direction of the abnormal area (for example, when warning at the rear, the sound source is positioned behind the display screen).

[0141] Emergency braking linkage: When the same area remains in the LV2 state for 10 seconds, a CAN signal is automatically sent to the elevator control system to trigger the following responses: The rotation speed of the car fan is increased to 3000 RPM (Revolutions Per Minute); An "environmental emergency" event code (Event Code 0xE3) is sent to the building management system.

[0142] According to the risk level of the heat map, the monitoring strategy is automatically adjusted: The sampling frequency is increased for high-risk areas to ensure data real-time; At the same time, environmental data and the risk heat map are superimposed and displayed on the car display screen, and high-risk areas are highlighted through color changes, flashing prompts, etc., to achieve intuitive information transmission, optimize resource allocation, and concentrate monitoring efforts to address major risks. Real-time visualization improves the risk perception ability of passengers and maintenance personnel and promotes rapid response.

[0143] S205, according to the duration and spatial diffusion trend of the zonal anomaly probability heat map, an environmental deterioration warning signal is generated through a spatio-temporal prediction model. When the level of the environmental deterioration warning signal reaches the preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, forming a closed-loop monitoring response.

[0144] Specifically, the spatio-temporal change characteristics of the zonal anomaly probability heat map can be extracted, and the duration, diffusion speed, and adjacent area correlation strength of the abnormal area are modeled through a spatio-temporal graph convolutional network to output a spatio-temporal feature vector; The system collects the zonal anomaly probability heat map (resolution 50×30 pixels) continuously for 10 minutes at a rate of 1 frame per second, and extracts three types of key features through the Spatio-Temporal Graph Convolutional Network (ST-GCN); Duration Quantification: For each high-risk pixel block (such as the red area with coordinates from (35, 20) to (45, 25)), calculate the duration (in seconds) that it remains in an abnormal state continuously. For example, if a high-temperature area has not been eliminated for 120 seconds, it is marked as "long-term abnormal", and the feature value is normalized to 0.85 (120 seconds / the maximum set threshold of 141 seconds).

[0145] Dynamic Decay Mechanism: If the abnormal area is interrupted during the diffusion process (such as the disappearance of the red area due to ventilation and cooling), the duration feature value decays at a rate of 5% per second (for example, the original value of 0.8 drops to 0.8×(0.95^3)≈0.69 after 3 seconds of interruption).

[0146] Diffusion Speed Modeling: Detect the abnormal spread trend through the difference between adjacent heatmap frames: Diffusion Speed (Speed) = New Abnormal Pixel Area ÷ Time Interval (seconds).

[0147] Example: A VOC pollution source spreads from 5 pixels to 50 pixels in 5 seconds → Speed = (50 - 5) / 5 = 9 pixels / second.

[0148] Direction Encoding: Combining with the car space coordinate system, quantify the diffusion direction as an angular value (0° - 360°). For example, if the abnormality spreads from the back to the front (along the positive Y-axis), it is recorded as 90°, and if it spreads to the front left, it is recorded as 45°.

[0149] Region Association Strength Analysis: Use the Graph Attention Network (GAT) to calculate the coupling coefficient (CouplingCoefficient) between the abnormal source and the affected area: The association strength from the high-temperature source area (A) to the adjacent area (B) = Abnormal value of A × Heat conduction coefficient × Spatial distance attenuation factor; Example: The high-temperature area at the rear (abnormal value 0.9) is 2 meters away from the front area (attenuation factor 0.5), and the conduction coefficient is 0.8 → Association strength = 0.9×0.8×0.5 = 0.36; Output a 256-dimensional spatio-temporal feature vector (including sub-features such as duration, diffusion speed, direction angle, association strength, etc.) for use by the subsequent prediction model.

[0150] Input the spatio-temporal feature vector into a pre-trained LSTM-Transformer hybrid prediction model to predict the environmental deterioration trend within a preset number of minutes in the future and generate a probabilistic warning signal; The LSTM-Transformer hybrid model consists of a Long Short-Term Memory (LSTM) network and a Transformer encoder, and processes spatio-temporal features in stages: ‌LSTM time series modeling stage‌: Input 100 consecutive spatiotemporal feature vectors (covering the past 100 seconds) and extract the temporal dependency pattern through two layers of LSTM units (128 neurons per layer): The first layer of LSTM outputs a hidden state to capture short-term fluctuations (such as the change in abnormal area every 10 seconds); The second layer of LSTM outputs the cell state (Cell State), which memorizes long-term trends (such as continuous temperature rise for more than 3 minutes).

[0151] Key parameter example: When the hidden state value exceeds the threshold of 0.7, it is judged as "accelerated deterioration" (for example, the CO2 concentration in a certain area increases by 200 ppm per minute).

[0152] ‌Transformer spatial dependency modeling phase‌: Input the temporal features output by LSTM into the Transformer encoder (4-layer structure, 8 attention heads per layer): The self-attention mechanism calculates the mutual influence of features in each region: the attention weight of the features in the rear high-temperature area and the central area is 0.85 (strong correlation); the weight of the low-temperature area on the side wall is only 0.12 (weak correlation); positional encoding preserves spatial topology information (for example, the coordinate encoding value of the rear of the car is higher than that of the front).

[0153] Output a comprehensive spatiotemporal feature matrix (128×100) to characterize the future risk evolution pattern.

[0154] Probabilistic early warning signal generation: The fully connected layer maps the feature matrix into three types of warning probabilities: probability of deterioration within the next 2 minutes (e.g., 82%); probability of deterioration within the next 5 minutes (e.g., 64%); and probability of deterioration within the next 10 minutes (e.g., 28%).

[0155] Decision rule: If any probability value is ≥ 70%, a graded warning signal is generated (e.g., "Level 1 warning: 82% probability of deterioration of rear high temperature within 2 minutes").

[0156] A three-level gradient alarm mechanism is designed. When the probabilistic warning signal reaches level one, an audible and visual alarm is triggered in the elevator car. When it reaches level two, the ventilation system is activated to enhance ventilation. When it reaches level three, an emergency elevator dispatch request is sent to the building management system. The three-level alarm mechanism adopts a cumulative triggering strategy, and the response intensity increases with the warning level: Level 1 alarm (warning probability ≥ 70%): Acoustic and optical warning: The LED strip on the top of the car is switched to a red breathing flashing mode (frequency 1 Hz, brightness 1000 lumens).

[0157] Play prerecorded voice: "Abnormal environment detected, please pay attention to safety" (volume 65 decibels, repeated 3 times).

[0158] Physical positioning: The warning signal source points to the direction of the high-risk area on the heat map (for example, when the rear is abnormal, the sound emits from the rear speaker).

[0159] Secondary alarm (warning probability ≥ 85%): Ventilation enhancement: Send instructions to the fan controller through the Modbus protocol (industrial equipment communication standard) to increase the speed from 1500 RPM to 3000 RPM (Revolutions Per Minute).

[0160] Start the emergency ventilation mode: The opening degree of the fresh air valve increases from 30% to 100%, and the ventilation volume increases by 3 times.

[0161] Local purification: If the VOC exceeds the standard, activate the nano filter above the high-risk area (such as grid B3 - C3 at the rear).

[0162] Tertiary alarm (warning probability ≥ 95%): Elevator dispatching linkage: Generate event code 0xE7 (environmental emergency evacuation instruction) and send it to the building management system through the OPC UA protocol (industrial automation communication protocol).

[0163] Building system execution strategy: Dispatch the target elevator to stop at the nearest floor (for example, if it was originally scheduled to go up to the 20th floor, change to open the door at the 15th floor); adjacent elevators are stopped (to prevent passengers from entering the dangerous car).

[0164] Escape guidance: The car display screen is switched to a green arrow to indicate the path to leave the cabin (such as "Please move forward to the front door").

[0165] Establish a closed-loop verification mechanism, compare the actual environmental change data with the prediction results, dynamically update the parameters of the hybrid prediction model through reinforcement learning, continuously optimize the warning accuracy, and form a closed-loop monitoring response.

[0166] The closed-loop verification system performs model optimization once every 24 hours, and the process is as follows: Data comparison and analysis: Extract historical prediction records (such as 10 "primary warnings") and actual sensor data (such as the temperature rise amplitude): Correct prediction: The actual deterioration occurs within the warning time window (e.g., if deterioration is predicted within 2 minutes and actually triggers at the 110th second); Prediction failure: The actual situation does not deteriorate or the deterioration exceeds the time limit (e.g., if deterioration is predicted in 5 minutes but actually occurs in 10 minutes).

[0167] Quantitative error metrics: Time error = |Predicted deterioration time - Actual deterioration time| (e.g., an error of 120 seconds is recorded as -2 minutes); Spatial error = Distance from the predicted position to the center point of the actual position (e.g., a deviation of 0.5 meters).

[0168] Reinforcement learning parameter update: Use Proximal Policy Optimization (PPO algorithm) to adjust the model weights: Reward function design: Prediction successful and time error < 30 seconds: Reward value +1.0; Prediction failure but spatial error < 0.3 meters: Reward value -0.5; Complete false alarm (no actual deterioration): Reward value -1.5.

[0169] Policy network update: If the predicted time error > 60 seconds for three consecutive times, reduce the weight of the LSTM forget gate by 10%; If the spatial error persists at a high level, increase the weight of the Transformer position encoding layer by 15%; After the model is updated, it is deployed to the edge computing unit to replace the old version.

[0170] Closed-loop response verification: Simulate environmental mutation scenarios (such as injecting high-temperature gas) to test the performance of the new model: Before optimization: The average predicted deterioration time error is 45 seconds; After optimization: The error is reduced to 12 seconds and the warning accuracy rate is increased to 93%; Real-time linkage verification: After the three-level alarm command is issued to the building system, the measured elevator dispatching delay is shortened from 20 seconds to 8 seconds.

[0171] By analyzing the duration and diffusion speed of the abnormal area, predict the future environmental change trend. Trigger responses at different levels according to the prediction results: Primary warning to prompt passengers, intermediate linkage with the ventilation system, and high-level adjustment of the elevator operation strategy, such as preferentially dispatching the elevator to the nearest floor for ventilation treatment, forming a complete closed loop from detection to disposal, realizing the upgrade from passive monitoring to active prediction. The graded response mechanism ensures that the disposal intensity matches the risk level. System linkage realizes intelligent management and effectively prevents the deterioration of environmental accidents.

[0172] It can be seen that by using a multi-modal sensor array to collect the environmental data in the elevator car in real time, a dynamic environmental feature map is generated; based on the dynamic environmental feature map, an environmental health index with high confidence is output; according to the environmental health index and the passenger density detection data, a heat map of the abnormal probability of each area is generated; based on the heat map of the abnormal probability of each area, real-time environmental parameters, the heat map of the abnormal probability of each area and risk warning prompts are superimposed and displayed on the car display screen; according to the duration and spatial diffusion trend of the heat map of the abnormal probability of each area, an environmental deterioration warning signal is generated through a spatio-temporal prediction model. When the level of the environmental deterioration warning signal reaches the preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, so that the accuracy and response efficiency of elevator environmental safety management can be improved through multi-modal data fusion, dynamic risk assessment and predictive warning.

[0173] Another embodiment of the present invention provides an elevator environment anomaly monitoring system. Refer to Figure 3 , the system may include: A fusion module 301, configured to collect the environmental data in the elevator car in real time through a multi-modal sensor array, and use a spatio-temporal convolutional network to fuse the spatial distribution characteristics and time series fluctuation characteristics of the sensor data to generate a dynamic environmental feature map, wherein the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds and voiceprint data; A reconstruction module 302, configured to reconstruct and denoise an environmental state model based on the dynamic environmental feature map through a generative adversarial network, wherein the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the environmental anomaly probability in combination with a historical normal environmental database and outputs an environmental health index with high confidence; An association module 303, configured to construct a multi-modal anomaly scoring model according to the environmental health index and the passenger density detection data, and use a graph neural network to associate the environmental parameter correlation between elevator car areas to generate a heat map of the abnormal probability of each area, wherein the heat map of the abnormal probability of each area marks the positions of high-risk areas and the abnormal levels; A display module 304, configured to dynamically allocate monitoring resources based on the heat map of the abnormal probability of each area through an adaptive attention mechanism, preferentially enhance the sensor sampling frequency of high-risk areas, and superimpose and display real-time environmental parameters, the heat map of the abnormal probability of each area and risk warning prompts on the car display screen; A warning module 305, configured to generate an environmental deterioration warning signal through a spatio-temporal prediction model according to the duration and spatial diffusion trend of the heat map of the abnormal probability of each area. When the level of the environmental deterioration warning signal reaches the preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy to form a closed-loop monitoring response.

[0174] It can be seen that by collecting the environmental data in the car in real time through a multimodal sensor array, a dynamic environmental feature map is generated; based on the dynamic environmental feature map, an environmental health index with high confidence is output; according to the environmental health index and the passenger density detection data, a heat map of the abnormal probability of each area is generated; based on the heat map of the abnormal probability of each area, real-time environmental parameters, the heat map of the abnormal probability of each area and risk warning prompts are superimposed and displayed on the car display screen; according to the duration and spatial diffusion trend of the heat map of the abnormal probability of each area, an environmental deterioration warning signal is generated through a spatio-temporal prediction model. When the level of the environmental deterioration warning signal reaches the preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator dispatching strategy, so that the accuracy and response efficiency of elevator environmental safety management can be improved through multimodal data fusion, dynamic risk assessment and predictive warning.

[0175] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the steps in the above-mentioned embodiment of the elevator environmental anomaly monitoring method when running.

[0176] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for executing the following steps: S201, collect the environmental data in the car in real time through a multimodal sensor array, and use a spatio-temporal convolutional network to fuse the spatial distribution characteristics and time series fluctuation characteristics of the sensor data to generate a dynamic environmental feature map, where the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds and voiceprint data; S202, based on the dynamic environmental feature map, reconstruct and denoise the environmental state model through a generative adversarial network. The generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the environmental anomaly probability in combination with the historical normal environmental database and outputs an environmental health index with high confidence; S203, according to the environmental health index and the passenger density detection data, construct a multimodal anomaly scoring model, and use a graph neural network to associate the environmental parameter correlation between car areas to generate a heat map of the abnormal probability of each area, where the heat map of the abnormal probability of each area marks the location and abnormal level of high-risk areas; S204, based on the heat map of the abnormal probability of each area, dynamically allocate monitoring resources through an adaptive attention mechanism, preferentially enhance the sensor sampling frequency in high-risk areas, and superimpose and display real-time environmental parameters, the heat map of the abnormal probability of each area and risk warning prompts on the car display screen; S205: Based on the duration and spatial diffusion trend of the partitioned anomaly probability heat map, an environmental deterioration warning signal is generated through a spatiotemporal prediction model. When the environmental deterioration warning signal level reaches a preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, forming a closed-loop monitoring response.

[0177] It can be seen that the environmental data in the cabin is collected in real time through the multimodal sensor array to generate a dynamic environmental feature map; based on the dynamic environmental feature map, a high-confidence environmental health index is output; based on the environmental health index and passenger density detection data, a partition abnormality probability heat map is generated; based on the partition abnormality probability heat map, real-time environmental parameters, partition abnormality probability heat map and risk warning prompts are superimposed on the cabin display screen; based on the duration and spatial diffusion trend of the partition abnormality probability heat map, an environmental deterioration warning signal is generated through the spatiotemporal prediction model. When the environmental deterioration warning signal level reaches the preset signal level, the gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, thereby improving the accuracy and response efficiency of elevator environmental safety management through multimodal data fusion, dynamic risk assessment and predictive warning.

[0178] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps of the embodiment of the elevator environment abnormality monitoring method.

[0179] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0180] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program: S201, collecting environmental data in the car in real time through a multimodal sensor array, and fusing the spatial distribution characteristics of the sensor data with the time series fluctuation characteristics using a spatiotemporal convolutional network to generate a dynamic environmental feature map, wherein the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint data; S202, based on the dynamic environmental feature map, reconstructing a denoised environmental state model through a generative adversarial network, wherein the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the probability of environmental anomaly based on a historical normal environment database, and outputs a high-confidence environmental health index; S203: Based on the environmental health index and passenger density detection data, a multimodal anomaly scoring model is constructed, and a graph neural network is used to correlate the environmental parameter correlations between cabin areas to generate a partitioned anomaly probability heat map, wherein the partitioned anomaly probability heat map identifies the location of high-risk areas and the anomaly level; S204: Based on the partitioned abnormality probability heat map, monitoring resources are dynamically allocated through an adaptive attention mechanism, the sensor sampling frequency in high-risk areas is prioritized, and real-time environmental parameters, the partitioned abnormality probability heat map, and risk warning prompts are superimposed on the car display screen; S205: Based on the duration and spatial diffusion trend of the partitioned anomaly probability heat map, an environmental deterioration warning signal is generated through a spatiotemporal prediction model. When the environmental deterioration warning signal level reaches a preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, forming a closed-loop monitoring response.

[0181] It can be seen that the environmental data in the cabin is collected in real time through the multimodal sensor array to generate a dynamic environmental feature map; based on the dynamic environmental feature map, a high-confidence environmental health index is output; based on the environmental health index and passenger density detection data, a partition abnormality probability heat map is generated; based on the partition abnormality probability heat map, real-time environmental parameters, partition abnormality probability heat map and risk warning prompts are superimposed on the cabin display screen; based on the duration and spatial diffusion trend of the partition abnormality probability heat map, an environmental deterioration warning signal is generated through the spatiotemporal prediction model. When the environmental deterioration warning signal level reaches the preset signal level, the gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, thereby improving the accuracy and response efficiency of elevator environmental safety management through multimodal data fusion, dynamic risk assessment and predictive warning.

[0182] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. An elevator environment anomaly monitoring method, characterized in that, The method comprises: The system collects environmental data in the cabin in real time through a multimodal sensor array, and uses a spatiotemporal convolutional network to fuse the spatial distribution characteristics of the sensor data with the time series fluctuation characteristics to generate a dynamic environmental feature map. The environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint data. Based on the dynamic environmental feature map, a denoised environmental state model is reconstructed through a generative adversarial network. The generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the probability of environmental anomalies based on a historical normal environment database and outputs a high-confidence environmental health index. Based on the environmental health index and passenger density detection data, a multimodal anomaly scoring model is constructed, and a graph neural network is used to correlate the environmental parameter correlations between cabin areas to generate a partitioned anomaly probability heat map, wherein the partitioned anomaly probability heat map identifies the location of high-risk areas and the level of anomaly; Based on the partition abnormality probability heat map, monitoring resources are dynamically allocated through an adaptive attention mechanism, sensor sampling frequency in high-risk areas is prioritized, and real-time environmental parameters, partition abnormality probability heat map and risk warning prompts are superimposed on the car display screen; Based on the duration and spatial diffusion trend of the partitioned anomaly probability heat map, an environmental deterioration warning signal is generated through a spatiotemporal prediction model. When the environmental deterioration warning signal level reaches the preset signal level, the gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, forming a closed-loop monitoring response.

2. The method according to claim 1, wherein The multimodal sensor array is used to collect environmental data in the car in real time, and the spatial distribution characteristics and time series fluctuation characteristics of the sensor data are integrated using a spatiotemporal convolutional network to generate a dynamic environmental feature map, wherein the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds and voiceprint data, including: Based on the original environmental data stream collected by the multimodal sensor array, a sliding time window is used to segment the temperature, humidity, CO2 concentration, volatile organic compounds and voiceprint signals to generate time-aligned multimodal data slices; The multimodal data slices are spatially encoded, the car is divided into gridded areas, and the topological relationships of the sensor nodes are annotated to construct a spatiotemporal correlation tensor. The spatiotemporal correlation tensor is input into a 3D convolution-long short-term memory hybrid network. The local region features are extracted through the spatial convolution kernel, and the fluctuation pattern across the time window is captured by combining the time axis LSTM. The feature vector set that integrates the spatiotemporal features is output. Regional features are aggregated based on the feature vector set, and a graph pooling algorithm is used to generate a dynamic environmental feature map covering the entire area of the car. The graph nodes encode the regional environmental status, and the edge weights represent the correlation between environmental parameters between regions.

3. The method according to claim 2, wherein Based on the dynamic environmental feature map, a denoised environmental state model is reconstructed through a generative adversarial network, wherein the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the probability of environmental anomaly based on a historical normal environment database and outputs a high-confidence environmental health index, including: Input the dynamic environmental feature map into the generator network, and reconstruct the denoised environmental state map through the multi-scale convolutional layer with residual connections. The generator outputs the estimated values of environmental parameters after noise suppression; Input the estimated values of environmental parameters and the standard state map in the historical normal environment database into the discriminator network, and calculate the distribution difference degree between the two through spectral normalization processing, and output the environmental anomaly probability score; In the adversarial training stage, the generator reversely optimizes the parameters according to the anomaly probability feedback by the discriminator, and uses the Wasserstein distance to constrain the authenticity of the generated data, and iteratively improves the denoising ability; Fuse the parameter deviation degrees of each node in the denoised environmental state map, calculate the comprehensive influence weights of temperature, humidity, CO2 concentration, and volatile organic compounds through the entropy weight method, and generate a multi-dimensional environmental health index; Normalize and calibrate the multi-dimensional environmental health index, and combine the anomaly probability output by the discriminator to generate a high-confidence environmental health index in the range of 0-1.

4. The method according to claim 3, characterized in that, Construct a multi-modal anomaly scoring model according to the environmental health index and the passenger density detection data, and use the graph neural network to associate the environmental parameter correlations between car areas to generate a heat map of the partition anomaly probability. Among them, the heat map of the partition anomaly probability calibrates the positions and anomaly levels of high-risk areas, including: Align the environmental health index and the crowd density detection data based on the camera in space and time, and dynamically allocate the environmental health weight and passenger density weight of different areas through the attention mechanism to generate a fusion weight matrix; Construct a graph neural network model, map the car area to graph nodes, use the environmental parameter correlation as the edge attribute, and model the abnormal propagation path between areas through the graph attention layer; Update the graph node features according to the fusion weight matrix, use the graph convolutional network to iteratively propagate the abnormal signal, calculate the abnormal cumulative value of each node, and generate the initial abnormal score distribution; Perform spatial interpolation processing on the initial abnormal score, use the heat diffusion algorithm to simulate the propagation trend of the anomaly in the car, and output the heat map of the partition anomaly probability that calibrates the positions and anomaly levels of high-risk areas.

5. The method according to claim 4, wherein Based on the heat map of the partition anomaly probability, dynamically allocate monitoring resources through the adaptive attention mechanism, preferentially enhance the sensor sampling frequency in high-risk areas, and superimpose and display the real-time environmental parameters, the heat map of the partition anomaly probability and the risk warning prompt on the car display screen, including: Divide the monitoring priorities according to the anomaly levels in the heat map of the partition anomaly probability, and dynamically calculate the resource allocation coefficient of high-risk areas using the attention weight function. The value of the resource allocation coefficient has an exponential relationship with the anomaly level; Adjust the sampling strategy of the sensor array, start the super-resolution sampling mode for high-risk areas, and increase the sampling frequency of temperature / humidity sensors to 3 times the reference value; Embed a heat map overlay module in the rendering engine of the car display screen, and perform pixel-level fusion of the real-time environmental parameters, the heat map of the partition anomaly probability and the camera video stream through the transparency gradient algorithm; Construct a risk warning dynamic annotation system. When the anomaly level of a certain area exceeds the preset level, automatically superimpose a flashing warning box and voice prompt at the corresponding position.

6. The method according to claim 5, characterized in that, According to the duration and spatial diffusion trend of the partition abnormality probability heat map, an environmental deterioration warning signal is generated through a spatiotemporal prediction model. When the environmental deterioration warning signal level reaches a preset signal level, a gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, forming a closed-loop monitoring response, including: Extract the spatiotemporal variation characteristics of the partitioned anomaly probability heat map, model the duration, diffusion speed, and correlation strength of the abnormal area through the spatiotemporal graph convolutional network, and output the spatiotemporal feature vector; The spatiotemporal feature vectors are input into a pre-trained LSTM-Transformer hybrid prediction model to predict environmental deterioration trends within a preset number of minutes and generate probabilistic early warning signals. A three-level gradient alarm mechanism is designed. When the probabilistic warning signal reaches level one, an audible and visual alarm is triggered in the elevator car. When it reaches level two, the ventilation system is activated to enhance ventilation. When it reaches level three, an emergency elevator dispatch request is sent to the building management system. Establish a closed-loop verification mechanism to compare actual environmental change data with prediction results, dynamically update the hybrid prediction model parameters through reinforcement learning, continuously optimize the accuracy of early warnings, and form a closed-loop monitoring response.

7. An elevator environment abnormal monitoring system, characterized in that, The system comprises: A fusion module is used to collect environmental data in the car in real time through a multimodal sensor array, and use a spatiotemporal convolutional network to fuse the spatial distribution characteristics of the sensor data with the time series fluctuation characteristics to generate a dynamic environmental feature map, where the environmental data includes temperature, humidity, CO2 concentration, volatile organic compounds, and voiceprint data; A reconstruction module is used to reconstruct a denoised environmental state model based on the dynamic environmental feature map through a generative adversarial network, wherein the generator eliminates sensor noise interference through adversarial training, and the discriminator calculates the probability of environmental anomaly based on a historical normal environment database and outputs a high-confidence environmental health index; a correlation module, configured to construct a multimodal anomaly scoring model based on the environmental health index and passenger density detection data, and utilize a graph neural network to correlate the environmental parameter correlations between cabin zones to generate a partitioned anomaly probability heat map, wherein the partitioned anomaly probability heat map identifies the location of high-risk areas and the level of anomaly; A display module is used to dynamically allocate monitoring resources based on the partition abnormality probability heat map through an adaptive attention mechanism, prioritize increasing the sensor sampling frequency in high-risk areas, and superimpose real-time environmental parameters, the partition abnormality probability heat map, and risk warning prompts on the car display screen; The early warning module is used to generate an environmental deterioration early warning signal through a spatiotemporal prediction model based on the duration and spatial diffusion trend of the partitioned anomaly probability heat map. When the environmental deterioration early warning signal level reaches the preset signal level, the gradient alarm mechanism is triggered and the building management system is linked to adjust the elevator scheduling strategy, forming a closed-loop monitoring response.

8. The system according to claim 7, wherein The fusion module is specifically used to: Based on the original environmental data stream collected by the multimodal sensor array, a sliding time window is used to segment the temperature, humidity, CO2 concentration, volatile organic compounds and voiceprint signals to generate time-aligned multimodal data slices; Perform spatial position encoding on multi-modal data slices, divide the car into grid regions and label the topological relationships of sensor nodes to construct a spatio-temporal correlation tensor; Input the spatio-temporal correlation tensor into a three-dimensional convolutional-long short-term memory hybrid network, extract local region features through spatial convolution kernels, and combine the time-axis LSTM to capture the fluctuation laws across time windows, and output a set of feature vectors that fuse spatio-temporal features; Perform regional feature aggregation based on the set of feature vectors, and use the graph pooling algorithm to generate a dynamic environment feature map covering the entire car area. Among them, the map nodes encode the regional environment state, and the edge weights represent the correlation of environmental parameters between regions.

9. A storage medium, characterized in that, A computer program is stored in the storage medium, wherein the computer program is set to execute the method described in any one of claims 1-6 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is set to run the computer program to execute the method described in any one of claims 1-6.

Images