Automatic detection and early warning method for door status of JSQ6 type concave bottom double-decker transport vehicle

Abstract

This invention discloses an automatic detection and early warning method for the door status of a JSQ6 type concave-bottom double-decker transport vehicle, belonging to the field of intelligent detection technology for railway transportation. The method includes: inputting door status signals into a multimodal data fusion model; generating a door status feature vector through feature extraction and weight calculation; and combining this vector with visual camera data, piezoelectric tactile sensor data, and environmental monitoring data for feature fusion and pattern recognition to generate a comprehensive door status evaluation vector. The method further involves classifying and mapping the door risk scores to generate door status early warning information; and integrating and analyzing the early warning information to generate a real-time door status early warning report. Through risk level classification and multi-dimensional time-series trend analysis, real-time early warning information and door status early warning reports are generated, thereby achieving intelligent, real-time monitoring of door status and reliable safety early warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology for railway transportation, and in particular to an automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-decker transport car. Background Technology

[0002] With the continuous development of railway freight vehicles, double-decker transport vehicles have gradually become important equipment for improving transport capacity and adapting to diversified transport needs. In particular, the JSQ6 concave-bottom double-decker transport vehicle, due to its compact structure, high loading efficiency, and strong adaptability, is widely used for long-distance transport of bulk goods. Real-time detection and monitoring of the vehicle door status has always been a core concern in the industry regarding the operational safety of transport vehicles. Traditional door detection methods mainly rely on mechanical limit switches, electrical contacts, and single sensors to collect open and closed status data. These methods are relatively simple in structure and can meet basic opening and closing judgment requirements.

[0003] In existing applications of related technologies, the status recognition of double-decker transport vehicle doors still mainly focuses on "single-dimensional" monitoring methods, often emphasizing mechanical position feedback or the collection and analysis of data from single-type sensors. This approach is easily affected by complex operating environments. For example, changes in ambient lighting can affect the stability of visual data, sensor errors can cause deviations in door status judgments, and there is a lack of unified modeling and fusion processing of multi-dimensional information. This results in room for improvement in the timeliness and accuracy of abnormal door status detection, especially in double-decker concave-bottom vehicle structures where door operation is constrained by the spatial environment, making incomplete data collection or delayed judgments more likely. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an automatic detection and early warning method for the door status of the JSQ6 type concave bottom double-layer transport vehicle, which solves the problems of incomplete collection of multi-dimensional status information of the door and untimely detection of abnormal status.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an automatic detection and early warning method for the door status of a JSQ6 type concave-bottom double-decker transport vehicle, which includes real-time acquisition of comprehensive door status data of the JSQ6 type concave-bottom double-decker transport vehicle, synchronization, noise reduction, standardization and feature integration, and generation of door status signals;

[0008] The door status signal is input into the multimodal data fusion model. Through feature extraction and weight calculation, a door status feature vector is generated. Combined with visual camera data, piezoelectric tactile sensor data and environmental monitoring data, feature fusion and pattern recognition are performed to generate a comprehensive evaluation vector of the door status.

[0009] The comprehensive evaluation vector of the car door status is input into a convolutional neural network. Through data compression and information integration operations, an abstract feature representation is obtained. Then, fully connected layer processing and activation function calculation are performed to generate a car door risk score.

[0010] The risk scores of car doors are classified and mapped according to risk levels to generate early warning information on car door status.

[0011] The system integrates and analyzes door status warning information to generate real-time door status warning reports.

[0012] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 concave-bottom double-decker transport vehicle described in this invention, the steps of real-time acquisition of comprehensive door status data of the JSQ6 concave-bottom double-decker transport vehicle, synchronization, noise reduction, standardization, and feature integration to generate a door status signal are as follows.

[0013] Real-time acquisition of door opening and closing angle, vibration response, and local temperature of the frame; through fixed-frequency recording and synchronous processing, comprehensive door status data is generated; and interpolation and alignment processing is performed to generate a synchronized door data stream.

[0014] The door synchronization data stream is subjected to noise suppression and de-jitter processing to generate a clean door signal sequence. The door feature vector is obtained through multi-dimensional feature extraction and structured coding.

[0015] The door feature vectors are aligned and fused over time to generate door status signals.

[0016] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the steps of inputting the door status signal into a multimodal data fusion model and generating a door status feature vector through feature extraction and weight calculation are as follows.

[0017] The door status signal is input into the multimodal data fusion model, and feature extraction and integration calculation are performed to generate the door enhancement sequence. The door enhancement sequence is then used for time-frequency feature processing and nonlinear integration calculation to generate the enhancement signal.

[0018] The enhanced signal is subjected to time-weighted integration and feature accumulation processing to obtain an integrated signal sequence. Time integration and composite feature operation are then performed to form a door state feature vector.

[0019] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the steps of combining visual camera data, piezoelectric tactile sensor data, and environmental monitoring data to perform feature fusion and pattern recognition to generate a comprehensive evaluation vector for the door status are as follows.

[0020] Real-time acquisition of visual camera data, piezoelectric tactile sensor data, and environmental monitoring data, followed by time synchronization and preprocessing, yields multimodal data of the vehicle door.

[0021] Visual, tactile, and environmental features were extracted from the multimodal data of the car doors.

[0022] Visual features, tactile features, and environmental features are fused with the door status feature vector through weighted interaction within a unified time window to obtain a fused score scalar sequence. Then, a status classification judgment is performed to generate a comprehensive evaluation vector of the door status.

[0023] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the specific steps for extracting visual features, tactile features, and environmental features from the multimodal data of the door are as follows.

[0024] Time series segmentation and synchronization alignment are performed on the multimodal data of the car doors to obtain time window data slices;

[0025] Edge and contour processing is performed on the image data in the time window data slice to extract visual features;

[0026] The tactile signals in the time window data slices are filtered and energy is calculated to extract the tactile features;

[0027] The environmental monitoring signals in the time window data slices are subjected to perturbation analysis to extract environmental features.

[0028] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the step of inputting the comprehensive evaluation vector of the door status into a convolutional neural network, and obtaining an abstract feature representation through data compression and information integration operations, is as follows:

[0029] The comprehensive evaluation vector of the car door status is input into the convolutional neural network, and convolution operation and local pattern extraction are performed to form a convolutional feature tensor. Then, normalization processing is performed to obtain a normalized convolutional feature tensor.

[0030] The normalized convolutional feature tensor is subjected to dimensionality compression and feature concatenation operations to obtain preliminary convolutional features, and then pooling is performed to reduce the dimensionality to obtain the activation feature tensor.

[0031] The activation feature tensor is compressed and integrated to form an abstract feature representation.

[0032] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the specific steps for performing fully connected layer processing and activation function calculation to generate a door risk score are as follows.

[0033] The abstract feature representation is flattened in one dimension to form a standardized input vector, and then a fully connected transformation is performed to obtain a fully connected output vector.

[0034] The output vector of the fully connected layer is subjected to composite function feature enhancement processing to obtain the activation vector, and then mapping and scoring calculation operations are performed to generate the door risk score.

[0035] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the specific steps for performing risk level classification and mapping processing on the door risk score to generate door status early warning information are as follows.

[0036] The risk score of the car door is normalized to obtain a normalized risk index, and then nonlinear mapping is performed to form a risk coefficient.

[0037] The risk coefficients are classified and compared at the boundary to generate risk level labels. The time series data is then processed and formatted to generate door status warning information.

[0038] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the specific steps for integrating and analyzing the door status early warning information to generate a real-time door status early warning report are as follows.

[0039] The door status warning information is summarized and processed in chronological order to generate a warning dataset, and then formatted to generate a formatted warning dataset.

[0040] Multidimensional data analysis and time-series trend calculation are performed on the formatted early warning information dataset to generate time-series trend analysis results.

[0041] The time-series trend analysis results are subjected to pattern recognition and anomaly detection operations to obtain anomaly pattern recognition results. The results are then combined with door status warning information and risk level labels to generate a real-time door status warning report.

[0042] As a preferred embodiment of the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle described in this invention, the specific steps for performing multi-dimensional data analysis and time-series trend calculation on the formatted early warning information dataset to form a time-series trend analysis result are as follows.

[0043] The formatted early warning information dataset is arranged according to time series, and the features of each early warning information are extracted and processed to form a multi-dimensional feature matrix.

[0044] The multidimensional feature matrix is ​​processed by composite function mapping to generate enhanced feature vectors, and then the time series is processed by difference and accumulation to obtain the door risk trend sequence.

[0045] The risk trend sequence of car doors is integrated and summarized in chronological order to form the time-series trend analysis results.

[0046] The beneficial effects of this invention are as follows: By real-time acquisition of multimodal data of the doors of the JSQ6 concave-bottom double-layer transport vehicle, and performing synchronization, noise reduction, and feature integration, high-quality door status signals are generated. Through multimodal feature extraction, weighted fusion, and pattern recognition, a comprehensive door status evaluation vector is formed, achieving a comprehensive and multi-dimensional understanding of the door's operating status. The comprehensive evaluation vector is input into a convolutional neural network for feature abstraction and fully connected layer processing to generate a quantified door risk score, achieving high-precision identification of abnormal states. Through risk level classification and multi-dimensional time-series trend analysis, real-time early warning information and door status early warning reports are generated, thereby realizing intelligent and real-time monitoring of door status and reliable safety early warning. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of an automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle.

[0049] Figure 2 This is a flowchart of the data acquisition and preprocessing process for the door status of the JSQ6 type concave-bottom double-layer transport vehicle.

[0050] Figure 3 This is a flowchart of the multimodal data fusion and evaluation process for the doors of the JSQ6 type concave-bottom double-layer transport vehicle.

[0051] Figure 4 This is a flowchart of the convolutional neural network processing for the door state of the JSQ6 type concave-bottom double-layer transport vehicle. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0055] Reference Figures 1-4 As an embodiment of the present invention, this embodiment provides an automatic detection and early warning method for the door status of a JSQ6 type concave-bottom double-layer transport vehicle, comprising the following steps:

[0056] S1. Real-time acquisition of comprehensive door status data of JSQ6 type concave bottom double-layer transport vehicle, synchronization, noise reduction, standardization and feature integration, and generation of door status signals.

[0057] For details, please refer to Figure 2 The steps are as follows:

[0058] S1.1 Real-time acquisition of door opening and closing angle, vibration response and local temperature of the frame, through fixed frequency recording and synchronous processing, to form comprehensive door status data, and perform interpolation and alignment processing to generate door synchronous data stream.

[0059] Furthermore, the system collects real-time data on door opening and closing angles, vibration response, and local frame temperature. A fixed-frequency recording and synchronization process is used to periodically sample each value and record a precise timestamp at each sampling point, forming structured comprehensive door status data under a unified time reference. During the fixed-frequency recording and synchronization process, timing deviations between different sampling sources are eliminated through timestamp sorting and clock synchronization or external time reference calibration. Acquisition channels with different sampling rates are resampled for subsequent alignment. Interpolation and alignment processing is performed on the comprehensive door status data. This process compensates for missing sampling points by applying existing interpolation methods such as linear interpolation or spline interpolation at predetermined time grid points and reconstructs the sampling values ​​of each channel at a unified time grid point, outputting a synchronized door data stream that satisfies timing consistency and a one-to-one correspondence between samples.

[0060] S1.2. Noise suppression and de-jitter processing are performed on the door synchronization data stream to generate a clean door signal sequence. The door feature vector is obtained through multi-dimensional feature extraction and structured coding operations.

[0061] Furthermore, noise suppression and de-jitter processing are implemented for the opening angle channel, vibration response channel, and local temperature channel of the frame, respectively: noise suppression uses a low-pass filter to suppress high-frequency measurement noise, a notch filter to remove power frequency interference, and a wavelet threshold denoising method to eliminate non-stationary noise; de-jitter processing uses amplitude threshold and duration judgment or a sliding median filter to suppress short-time pulses and mechanical jitter pseudo-signals, outputting a purified door signal sequence; after obtaining the purified door signal sequence, multi-dimensional feature extraction is performed on each channel within a unified time window, including time-domain features (such as mean, root mean square value, peak-to-peak value, variance, skewness, kurtosis, and energy) and frequency-domain features. (The dominant frequency, spectral energy distribution, and bandwidth are extracted using Fast Fourier Transform) and time-frequency features are extracted (scale energy or energy envelope statistics are extracted using Short-Time Fourier Transform or Continuous Wavelet Transform); the impact factor and spectral peak index are further extracted from the vibration response channel to reflect transient events; edge intensity statistics are extracted from the edge contour signal of the image source and incorporated into the feature set in numerical form; the features of each channel are structured and encoded according to a predetermined channel-feature item mapping rule; each feature item is first standardized based on a rolling window (e.g., mean-variance normalization or minimum-maximum normalization), and then concatenated in a fixed order to form a fixed-length vector, outputting the door feature vector.

[0062] S1.3. Align and fuse the door feature vectors over time to generate door status signals.

[0063] Furthermore, the door feature vectors are mapped to a unified time base according to timestamps. The timestamp sequence contained in the door feature vectors is analyzed and common time grid points are determined. The sampled values ​​of each door feature vector are reconstructed on the common time grid points through linear interpolation or spline interpolation to ensure time alignment. To compensate for the time delay and phase difference between channels, the optimal time delay of each pair of door feature vectors is calculated within the moving time window using a cross-correlation function, and the corresponding door feature vectors are adjusted by time shift to eliminate systematic lag. The reconstructed and aligned door feature vectors are standardized element by element using mean-variance normalization or minimum-maximum normalization to eliminate dimensional differences and ensure numerical comparability. Within a time window, local robustness weights are calculated based on inverse variance estimation or signal-to-noise ratio estimation within the sliding window. These local robustness weights reflect the reliability of information for each door feature vector within the current time window. Following a predetermined channel-feature mapping rule, each door feature vector is weighted and fused according to its corresponding robustness weight. Scalar features are weighted and summed to obtain a fused scalar, while vectorized features are first scaled dimension-wise according to weights and then concatenated in a fixed order to obtain a fused vector. The fused vector in the time series is smoothed using methods such as moving averages or low-pass filters to reduce high-frequency fluctuations, and anomalous abrupt changes are eliminated through differential constraints. The output door status signal meets the requirements of time consistency, dimensional uniformity, and robustness.

[0064] S2. Input the door status signal into the multimodal data fusion model, generate a door status feature vector through feature extraction and weight calculation, and combine it with visual camera data, piezoelectric tactile sensor data and environmental monitoring data to perform feature fusion and pattern recognition to generate a comprehensive evaluation vector of the door status.

[0065] For details, please refer to Figure 3 The steps are as follows:

[0066] It should be noted that existing methods typically process door status signals, visual camera data, piezoelectric tactile sensor data, and environmental monitoring data independently, and only use simple stitching or single weighting methods to fuse them in the later stage. The pattern recognition process mostly relies on a single feature dimension, which is easily affected by environmental interference and single-type sensor errors, resulting in insufficient accuracy of comprehensive door status assessment.

[0067] This invention uses a multimodal data fusion model to perform deep feature extraction and interactive weighted fusion of door status signals, visual camera data, piezoelectric tactile sensor data, and environmental monitoring data within the same time window. During the fusion process, the importance and interrelationship of different features are considered, and a comprehensive evaluation vector of door status is generated by combining multi-dimensional joint change patterns in the pattern recognition stage, thereby improving the accuracy and robustness of the comprehensive evaluation of door status.

[0068] S2.1 Input the door status signal into the multimodal data fusion model, perform feature extraction and integration calculation to generate the door enhancement sequence, and perform time-frequency feature processing and nonlinear integration calculation through the door enhancement sequence to generate the enhancement signal.

[0069] Furthermore, the door status signal is input into a multimodal data fusion model. Local convolution and pooling operations are performed on each channel of the door status signal in a parallel branch to extract local temporal patterns. Simultaneously, a short-time Fourier transform or continuous wavelet transform is performed on the door status signal to obtain its time-frequency representation, and spectral energy, dominant frequency, and energy envelope are calculated as frequency domain and time-frequency features. Pointwise nonlinear mapping and gated weighting operations are then applied to the time-domain and time-frequency features extracted in the parallel branch. Specifically, a pointwise nonlinear mapping function (e.g., ReLU or Sigmoid) is applied to each door feature vector, combined with a local signal-to-noise ratio-based... The weights are multiplied element-wise and then weighted and summed. The fusion results are spliced ​​together in time order to form the door enhancement sequence. Fine-grained time-frequency feature processing is performed on the door enhancement sequence. The enhanced spectrum is obtained by performing short-time Fourier transform or continuous wavelet transform on the door enhancement sequence. The spectral peak time trajectory, frequency band energy distribution and energy envelope statistics are extracted from the enhanced spectrum. After the extracted spectral peak time trajectory, frequency band energy distribution and energy envelope statistics are normalized dimension-wise, nonlinear mapping and time pooling operations (such as time max pooling or average pooling) are applied to suppress instantaneous noise and enhance persistent features, and the enhanced signal is output.

[0070] It should also be noted that the process of constructing the multimodal data fusion model involves time alignment and normalization of the door status signal to ensure a unified input dimension. Local features are extracted and redundant information is compressed through convolution operations. The convolution operation results are then input into a multi-layer neural network structure for nonlinear mapping to form a multi-layer abstract representation. This multi-layer neural network structure consists of an input layer, a hidden layer, and an output layer. The input layer receives the convolution operation results and performs dimension matching. The hidden layer consists of several fully connected layers and nonlinear activation functions to extract high-level abstract features and compress redundant information. The output layer integrates and compresses the abstract features generated by the hidden layer to form a fixed-dimensional representation. An attention mechanism is then used to enhance key information and suppress irrelevant features. Finally, the data is integrated and normalized through a fully connected layer to obtain the multimodal data fusion model.

[0071] The training process of a multi-layer neural network involves feeding the input feature vector into each layer of the network for forward propagation to generate the output vector, calculating the loss function using known labels or target values, updating the network weights and biases layer by layer using the backpropagation algorithm, and iterating and optimizing until the loss function converges to obtain a multi-layer neural network that can be used for feature mapping.

[0072] S2.2 Perform time-weighted integration and feature accumulation processing on the enhanced signal to obtain an integrated signal sequence, and perform time integration and composite feature operation to form a door state feature vector.

[0073] Furthermore, the enhanced signal undergoes time-series weighted integration and feature accumulation processing. Within a unified time window, weighting coefficients are assigned to the enhanced signal at each time step. These weighting coefficients can be calculated based on signal-to-noise ratio, local variance, or feature stability. The enhanced signal is combined with the weights through element-wise multiplication to obtain a weighted signal. Subsequently, the weighted signal is accumulated along the time dimension, for example, through stepwise summation or sliding window integration, to form an integrated signal sequence. A time integration operation is performed on the integrated signal sequence, accumulating the feature values ​​of each time step along the time axis to capture long-term trends. Simultaneously, composite feature operations are performed, arithmetically combining or weighted combining the different door feature vectors of the integrated signal sequence, such as calculating the mean, variance, or linear combination, to enhance feature representation capabilities. The output is a door state feature vector.

[0074] S2.3. Real-time acquisition of visual camera data, piezoelectric tactile sensor data, and environmental monitoring data, followed by time synchronization and preprocessing to obtain multimodal data of the car door.

[0075] Furthermore, when collecting real-time data from the visual camera, piezoelectric tactile sensor, and environmental monitoring, each type of sensor data is recorded at a fixed sampling frequency, and timestamp information is saved during the recording process to ensure synchronous reference of the sensor data. The visual camera data undergoes image denoising, color space standardization, and resolution unification processing; the piezoelectric tactile sensor data is filtered and de-jittered; and the environmental monitoring data undergoes outlier removal and normalization processing. After all types of preprocessing are completed, the three types of preprocessed data are time-aligned and interpolated using timestamps to ensure that all sensor data correspond at the same time step. The aligned visual camera data, piezoelectric tactile sensor data, and environmental monitoring data are then combined in a time series to generate multimodal data of the vehicle door.

[0076] S2.4 Perform time series segmentation and synchronization alignment operations on the multimodal data of the car door to obtain time window data slices.

[0077] Furthermore, the continuous time series is divided into fixed-length time windows to ensure that each time window contains complete visual camera data, piezoelectric tactile sensor data, and environmental monitoring data. Within each time window, the timestamps of the visual camera data, piezoelectric tactile sensor data, and environmental monitoring data are checked, and missing or misaligned time points are adjusted through linear interpolation or nearest neighbor interpolation to ensure that the visual camera data, piezoelectric tactile sensor data, and environmental monitoring data are precisely aligned at the same time step. After alignment, the multimodal signals within each time window are combined to form an independent time window data slice.

[0078] S2.5. Perform edge and contour processing on the image data in the time window data slice to extract visual features.

[0079] Furthermore, the color image is converted to a grayscale image to highlight brightness variation information. The Sobel or Canny operator is used to calculate the gradient of the grayscale image to obtain the brightness variation amplitude of each pixel in the horizontal and vertical directions. The edge pixel positions are determined based on the gradient amplitude, forming an edge information set. A contour tracking algorithm is used to continuously connect the edge pixels to generate a contour set, and the geometric features of each contour, such as length, shape, and spatial position, are extracted. After clarifying the source of the edge information and contour set, the edge pixel positions and their gradient amplitudes are associated with the geometric features of the contours and combined into a unified representation. The combined information is organized in vector or matrix form to form a structured code, resulting in visual features.

[0080] S2.6 Filter and calculate the energy of the tactile signal in the time window data slice to extract the tactile features.

[0081] Furthermore, a low-pass filter is applied to the tactile signal, such as a Butterworth filter with a cutoff frequency of 50 Hz, to remove high-frequency noise and jitter interference. The sum of squares or average of the squared amplitudes of the filtered tactile signal are then calculated at each time step to obtain the signal energy value, expressed as:

[0082] ;

[0083] in, Indicates time step The signal energy value, Indicates the current time step index. This indicates the number of sampling points used to calculate energy. Indicates the sampling point index. Indicates at time step The The amplitude of the filtered tactile signal at each sampling point;

[0084] Used to measure the intensity and trend of tactile response; after energy calculation is completed, the energy values ​​of each time step are organized in chronological order and combined with the instantaneous characteristics of the signal, such as peak value, mean value and standard deviation, to form a structured representation, and these structured tactile information are encoded as tactile features.

[0085] S2.7 Perform disturbance analysis on the environmental monitoring signals in the time window data slices to extract environmental features.

[0086] Furthermore, the environmental monitoring signals are smoothed, for example, by using moving averages or low-pass filtering to remove random noise and short-term fluctuations. The instantaneous rate of change and local variance at each time step are calculated to assess the degree of disturbance in the signal over a short period of time. The degree of disturbance is normalized and combined with the average direction of change and cumulative change of the environmental monitoring signals over a longer time scale to obtain a complete representation of the environmental characteristics. Finally, the degree of disturbance and the environmental characteristic representation are structured and encoded to obtain the environmental features.

[0087] S2.8. Visual features, tactile features, and environmental features are fused with the door status feature vector through weighted interaction within a unified time window to obtain a fused scoring scalar sequence, and then a status classification judgment is performed to generate a comprehensive evaluation vector of the door status.

[0088] Furthermore, when performing weighted interactive fusion calculations of visual features, tactile features, and environmental features with the door state feature vector within a unified time window, weights are assigned based on the importance of each type of feature (such as visual features, tactile features, environmental features, and door state feature vector). The contribution and stability of each feature to the overall door state information within the unified time window are differentiated. Visual features are evaluated using historical change amplitude or edge / contour information energy; tactile features are evaluated using instantaneous energy value or response intensity; environmental features are evaluated using the degree of disturbance and long-term trend reliability; and the door state feature vector is evaluated using historical state changes or statistical stability. Features with high contribution, high correlation, or high stability receive higher weights, while features with low contribution or low correlation receive lower weights. The weighted summation is performed at each time step, with visual features, tactile features, environmental features, and door state feature vectors weighted according to their respective weights. The interaction between features is considered, and this interaction is obtained by quantifying the joint changes of different feature components within the same time step. For example, element-wise multiplication of the visual and tactile feature vectors reflects the intensity of simultaneous changes in both types of features across different dimensions; or the covariance of visual and environmental features is calculated within a time window to reflect the joint change trend of the two types of features over time. This interactive information is integrated during the weighted fusion process, ensuring that the fused score scalar sequence reflects both the importance of individual features and the joint effect between multiple features. After fusion, the fused score scalar sequence is obtained, expressed as:

[0089] ;

[0090] in, Indicates at time step The fusion score scalar sequence, This represents the index of discrete time steps within a unified time window. Indicates the feature category index. Indicates the index of the interaction feature category. Indicates at time step The Numerical representation of class features Indicates at time step Assigned to the Importance weights of class features Indicates at time step The first result obtained through interactive computation Interactive features, This represents the weights assigned to the interactive features during the fusion process;

[0091] The fusion score scalar sequence can reflect the comprehensive door status information at each time step; finally, the fusion score scalar sequence is classified and judged, for example, according to the example classification interval or clustering method, the score sequence is mapped to the corresponding status category (such as normal status, slightly abnormal status, and severely abnormal status), forming a comprehensive evaluation vector of door status.

[0092] S3. Input the comprehensive evaluation vector of the car door status into the convolutional neural network. Through data compression and information integration operations, obtain the abstract feature representation, and perform fully connected layer processing and activation function calculation to generate the car door risk score.

[0093] For details, please refer to Figure 4 The steps are as follows:

[0094] It should be noted that existing methods typically input the comprehensive evaluation vector of the car door status directly into the convolutional neural network, mainly relying on the convolutional layer to extract local features, and then outputting risk-related values ​​through the fully connected layer. In the process, risk results are generated mainly through single-dimensional feature compression and classification.

[0095] This invention inputs the comprehensive evaluation vector of the vehicle door status into a convolutional neural network, performs multi-level feature compression and information integration through convolutional layers, and then combines fully connected layers and activation functions to generate a vehicle door risk score. Its advantage is that it can preserve the temporal correlation and local feature differences in the comprehensive evaluation vector of the vehicle door status during the convolution operation, thereby providing a more complete abstract feature representation for the risk score.

[0096] S3.1 Input the comprehensive evaluation vector of the car door status into the convolutional neural network for convolution operation and local pattern extraction to form a convolutional feature tensor, and perform normalization processing to obtain a normalized convolutional feature tensor.

[0097] Furthermore, the comprehensive evaluation vector of the car door state is arranged into an input matrix suitable for convolution operations according to the time series and feature dimensions. Under the action of each convolution kernel, the input matrix is ​​scanned locally to extract local pattern features. For example, the correlation between time and feature dimensions is captured by sliding window convolution. The convolution result forms a convolutional feature tensor, which contains the feature responses of the car door state at different local regions and scales. The convolutional feature tensor is normalized, for example, by batch normalization or local response normalization, to adjust the feature values ​​within the tensor to a uniform range, resulting in a normalized convolutional feature tensor.

[0098] It should also be noted that the training process of a convolutional neural network involves inputting the input feature tensor into the convolutional layer for convolution and activation operations to generate feature maps. After dimensionality reduction by the pooling layer, higher-level features are formed. The fully connected layer maps the high-level features to a prediction result consistent with the output class dimension. This prediction result is then compared with the known labels of the training samples. The error is calculated according to the selected loss function and used to guide the update of the network parameters. The backpropagation algorithm is used to adjust the convolutional kernel weights, pooling parameters, and fully connected weights layer by layer, iteratively optimizing until the loss function converges, thus obtaining a convolutional neural network that can be used for feature extraction and mapping.

[0099] S3.2 Perform dimensionality compression and feature concatenation operations on the normalized convolutional feature tensor to obtain preliminary convolutional features, and then perform pooling dimensionality reduction operations to obtain the activation feature tensor.

[0100] Furthermore, linear or nonlinear compression operations are performed on the temporal or channel dimensions of the tensor, such as through convolutional kernels or channel-wise weighted summation, to merge multidimensional local features into low-dimensional feature representations. The compressed features are then concatenated across different channels or scales to form preliminary convolutional features, in order to preserve multi-scale information and local pattern responses. Pooling operations, such as max pooling or average pooling, are then performed on the preliminary convolutional features to reduce the dimensionality of the features along the temporal or spatial dimensions, thereby reducing redundant information and feature size, while highlighting key activation responses (e.g., highlighting features at the time step with the highest amplitude, local maximum activation points, or significant responses in specific channels). The pooling results form the activation feature tensor.

[0101] S3.3. Compress and integrate the activation feature tensor to form an abstract feature representation.

[0102] Furthermore, the activation feature tensors are linearly or nonlinearly aggregated in the channel dimension or spatial dimension, for example, by global average pooling or channel-wise weighted summation, to compress multidimensional local response features into a low-dimensional representation; the compressed features are then spliced ​​or fused to retain important activation patterns and local response information, while removing redundant and invalid features, thus forming an abstract feature representation.

[0103] S3.4. Perform one-dimensional flattening on the abstract feature representation to form a standardized input vector, and perform fully connected transformation calculation to obtain a fully connected output vector.

[0104] Furthermore, the abstract feature representation is expanded into a continuous one-dimensional vector according to the channel or spatial dimension, ensuring that all features are arranged in a uniform order to form a standardized input vector. The standardized input vector is then input into the fully connected operation, and the fully connected transformation calculation is completed by performing a linear transformation with the fully connected weight matrix and superimposing biases, thus obtaining the fully connected output vector.

[0105] S3.5. Perform composite function feature enhancement processing on the output vector of the fully connected layer to obtain the activation vector, and perform mapping and scoring calculation operations to generate the door risk score.

[0106] Furthermore, the fully connected output vector is mapped element-wise through a non-linear activation function, such as the Sigmoid or ReLU function, to map the original linear combination information into activation vectors, thereby enhancing important feature responses and suppressing low-contribution features. After generating the activation vectors, they are combined with historical door state information through mapping methods such as weighted summation or element-wise multiplication to form a fused mapping vector. The fused mapping vector is then subjected to a scoring operation, such as through linear mapping or normalization, to convert the fused mapping vector into a numerical representation on a uniform scale, generating a door risk score.

[0107] S4. Perform risk level classification and mapping processing on the door risk score to generate door status warning information.

[0108] S4.1 Normalize the risk score of the car door to obtain a normalized risk index, and perform nonlinear mapping to form a risk coefficient.

[0109] Furthermore, the minimum and maximum values ​​of the car door risk score within the overall sample range are linearly mapped to obtain a normalized risk index, ensuring that the normalized risk index values ​​fall within a uniform range. The normalized risk index is then transformed through a nonlinear mapping function to enhance the ability to distinguish high-risk values ​​and suppress fluctuations in low-risk values, thus forming a risk coefficient.

[0110] S4.2 Perform a classification boundary comparison operation on the risk coefficient, generate risk level labels, and perform time series processing and formatting to generate door status warning information.

[0111] Furthermore, each risk coefficient is compared with a normalized risk indicator. Based on the distribution of normalized risk indicators in historical door status samples, the numerical range is divided into low risk, medium risk, and high risk. Each risk coefficient is compared with the boundaries of low risk, medium risk, and high risk. If it falls into the corresponding range, the risk level can be determined and a risk level label can be generated. According to the time sequence corresponding to the risk level label, the risk levels at each time point are organized into a continuous sequence, and each label is uniformly coded. For example, low risk is marked as "L", medium risk as "M", and high risk as "H", forming a standardized format. The organized label sequence is output as structured door status warning information.

[0112] S5. Perform information integration and analysis on the door status warning information to generate a real-time door status warning report.

[0113] S5.1 Summarize and process the door status warning information in chronological order to generate a warning dataset, and then format it to generate a formatted warning dataset.

[0114] Furthermore, the door status warning information is sorted chronologically. Each warning includes the occurrence time, risk level label, and corresponding door status information. The sorted warnings are then merged sequentially by occurrence time to form a continuous and complete record sequence, generating a warning dataset. For the generated warning dataset, the field names, time format, risk level label, and door status label for each record are standardized. This includes standardizing the time to the example format "YYYY-MM-DD HH:MM:SS", the risk level label to the example label "Low, Medium, High", and the door status label to the example code "Closed, Half-open, Open", resulting in a formatted warning dataset with a standardized structure and uniform format.

[0115] S5.2 Arrange the formatted early warning information dataset according to the time series, and extract the features of each early warning information in each dimension for organization and processing to form a multi-dimensional feature matrix.

[0116] Furthermore, the formatted early warning datasets are sorted chronologically to ensure each formatted dataset has a clear position in the time series. For each formatted early warning dataset, risk level labels, door status indicators, occurrence time, and other record fields are extracted sequentially as features of each dimension. The features of each formatted early warning dataset are arranged in a fixed order and stacked sequentially according to the time series to form a two-dimensional matrix structure. The rows of the two-dimensional matrix represent early warning information records at different time steps, and the columns represent features of each dimension, thus achieving the organization and unified representation of multi-dimensional feature information and obtaining a structured and continuous multi-dimensional feature matrix.

[0117] S5.3 Perform composite function mapping on the multidimensional feature matrix to generate enhanced feature vectors, and perform difference and accumulation processing on the time series to obtain the door risk trend sequence.

[0118] Furthermore, the eigenvectors of each row of the multidimensional feature matrix are input into a composite function for nonlinear mapping operations to generate enhanced feature vectors, thereby strengthening the numerical and relational representation of the original features in each dimension. First-order differencing is performed on the enhanced feature vectors sequentially in chronological order to characterize the changing trends between consecutive time steps. The differencing results are accumulated, and the changes at each time step are superimposed to form a continuous risk accumulation curve. The continuous sequence obtained through differencing and accumulation is used as the door risk trend sequence.

[0119] S5.4 Integrate and summarize the risk trend sequence of the car door in chronological order to form the time-series trend analysis results.

[0120] Furthermore, the risk trend sequence of the vehicle door is integrated in chronological order, and the risk change trends of consecutive time steps are arranged and summarized in turn. By comparing the differences in risk values ​​of consecutive time steps in turn, the risk is determined to be rising if it is higher than the previous time step, and decreasing if it is lower than the previous time step. If the change range is within the example range, the risk is determined to be stable. The risk change pattern is extracted, and the risk fluctuation amplitude and direction within the continuous time period are marked and classified to form the time series trend analysis results.

[0121] S5.5 Perform pattern recognition and anomaly detection operations on the time-series trend analysis results to obtain anomaly pattern recognition results. Combine the door status warning information and risk level labels to perform information integration operations and generate a real-time door status warning report.

[0122] Furthermore, the trend characteristics of each time step are scanned sequentially based on the time-series trend analysis results. Abnormal fluctuations or states that do not conform to the normal change pattern are identified through pattern matching methods, generating abnormal pattern recognition results. The abnormal pattern recognition results are then matched and integrated with the time, content, and corresponding risk level label in each door status warning message to form a structured record containing anomaly identification, warning information, and risk level, generating a real-time door status warning report.

[0123] This embodiment also provides a computer device applicable to the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-decker transport vehicle, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-decker transport vehicle as proposed in the above embodiment.

[0124] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0125] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the automatic detection and early warning method for the door status of the JSQ6 type concave-bottom double-layer transport vehicle as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0126] In summary, this invention achieves intelligent, real-time monitoring and reliable safety warnings for the doors of a JSQ6 type concave-bottom double-decker transport vehicle by: real-time acquisition of multimodal data from the doors, followed by synchronization, denoising, and feature integration; generating high-quality door status signals through multimodal feature extraction, weighted fusion, and pattern recognition; forming a comprehensive door status evaluation vector through multimodal feature extraction, weighted fusion, and pattern recognition; and generating a quantified door risk score through risk level classification and multidimensional time-series trend analysis. Furthermore, it generates real-time early warning information and door status warning reports through risk level classification and multidimensional time-series trend analysis.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for automatically detecting and warning the door state of JSQ6 concave-bottom double-deck transport vehicle, characterized in that: Comprise, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, Real-time acquisition of the door comprehensive state data of JSQ6 concave bottom double-layer transport vehicle, synchronization, denoising, standardization and feature integration, generation of door state signal, the specific steps are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, 2. The JSQ6 type concave bottom double layer transport vehicle door state automatic detection and early warning method according to claim 1, characterized in that: The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, 3. The JSQ6 type concave bottom double layer transport vehicle door state automatic detection and early warning method according to claim 1, characterized in that: The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, 4. The JSQ6 type concave bottom double layer transport vehicle door state automatic detection and early warning method of claim 3, characterized in that: The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights to generate the door state feature vector are as follows, The specific steps of inputting the door state signal into the multi-modal data fusion model, extracting features and calculating weights The environmental monitoring signal in the time window data piece is disturbed to extract the environmental characteristics.

5. The JSQ6 type double-deck flat bottom railcar door status automatic detection and early warning method of claim 3, characterized in that: The vehicle door state comprehensive evaluation vector is input into the convolutional neural network, convolution operation and local pattern extraction are performed, a convolution feature tensor is formed, and normalization processing is performed to obtain a normalized convolution feature tensor. The vehicle door state comprehensive evaluation vector is input into the convolutional neural network, convolution operation and local pattern extraction are performed, a convolution feature tensor is formed, and normalization processing is performed to obtain a normalized convolution feature tensor. The normalized convolution feature tensor is subjected to dimension compression and feature splicing operation to obtain a preliminary convolution feature, and is subjected to pooling dimension reduction operation to obtain an activation feature tensor. The activation feature tensor is compressed and integrated to form an abstract feature representation.

6. The JSQ6 type double-deck railcar door state automatic detection and early warning method of claim 1, wherein: The abstract feature representation is subjected to one-dimensional flattening processing to form a standardized input vector, and full connection transformation calculation is performed to obtain a full connection layer output vector. The abstract feature representation is subjected to one-dimensional flattening processing to form a standardized input vector, and full connection transformation calculation is performed to obtain a full connection layer output vector. The vehicle door risk score is subjected to risk level classification and mapping processing to generate vehicle door state early warning information, and the specific steps are as follows, 7. The JSQ6 type automatic detection and early warning method for the door state of a double-deck railcar with a concave bottom according to claim 6, characterized in that: The vehicle door risk score is subjected to risk level classification and mapping processing to generate vehicle door state early warning information, and the specific steps are as follows, The vehicle door risk score is subjected to risk level classification and mapping processing to generate vehicle door state early warning information, and the specific steps are as follows, The vehicle door state early warning information is subjected to information integration and analysis operation to generate a real-time vehicle door state early warning report, and the specific steps are as follows, 8. The JSQ6 type automatic detection and early warning method for the door state of a double-deck railcar with a concave bottom according to claim 7, characterized in that: The vehicle door state early warning information is subjected to information integration and analysis operation to generate a real-time vehicle door state early warning report, and the specific steps are as follows, The vehicle door state early warning information is subjected to information integration and analysis operation to generate a real-time vehicle door state early warning report, and the specific steps are as follows, The formatted early warning information data set is subjected to multi-dimensional data analysis and time series trend calculation to form a time series trend analysis result, and the specific steps are as follows, The formatted early warning information data set is subjected to multi-dimensional data analysis and time series trend calculation to form a time series trend analysis result, and the specific steps are as follows, 9. The JSQ6 type automatic detection and early warning method for the door state of a double-deck railcar with a concave bottom according to claim 8, characterized in that: The formatted early warning information data set is subjected to multi-dimensional data analysis and time series trend calculation to form a time series trend analysis result, and the specific steps are as follows, The formatted early warning information data set is subjected to multi-dimensional data analysis and time series trend calculation to form a time series trend analysis result, and the specific steps are as follows, The formatted early warning information data set is subjected to multi-dimensional data analysis and time series trend calculation to form a time series trend analysis result, and the specific steps are as follows, The formatted early warning information data set is subjected to multi-dimensional data analysis and time series trend calculation to form a time series trend analysis result, and the specific steps are as follows,

Images