Multi-layer framework intelligent operation and maintenance and predictive maintenance method and system for rail transit platform door

Abstract

The invention provides a rail transit platform door multilayer architecture intelligent operation and maintenance and predictive maintenance method and system, and relates to the technical field of intelligent operation and maintenance, and the method comprises the steps: obtaining platform door operation parameters through a multilayer data collection system, and generating health state feature vectors; inputting the feature vectors into a deep neural network, and constructing a fault prediction model; based on the prediction result, a decision tree algorithm is adopted to evaluate the maintenance priority, and an intelligent maintenance scheme is generated; and pushing an instruction to maintenance personnel through the mobile terminal, and feeding back an optimization model by using new data to realize self-adaptive adjustment of a maintenance strategy.

Description

Technical Field

[0001] The present invention relates to intelligent operation and maintenance technology, and in particular to an intelligent operation and maintenance and predictive maintenance method and system for a multi-layer architecture of a rail transit platform door. Background Art

[0002] The traditional regular maintenance mode cannot detect potential faults in time, which easily causes unexpected equipment downtime and safety hazards. Due to the lack of real-time monitoring and intelligent analysis methods, it is difficult to accurately grasp the actual operating status of each component of the platform door, resulting in passive and delayed maintenance work.

[0003] Secondly, the existing fault diagnosis methods mainly rely on the experience and judgment of maintenance personnel, and lack the ability to comprehensively analyze multi-source heterogeneous data. The platform door system involves multiple fields such as mechanics, electrical, and control. A single fault feature is difficult to fully reflect the health status of the equipment, affecting the accuracy and reliability of fault prediction.

[0004] Finally, traditional maintenance decisions lack data support and intelligent optimization, making it difficult to achieve accurate maintenance and optimal resource allocation. Maintenance plans are often fixed and fail to be dynamically adjusted according to the actual status of the equipment and operational needs, resulting in a waste of maintenance resources and low maintenance efficiency.

[0005] Therefore, it is urgent to develop an intelligent and predictive platform door maintenance method to improve equipment reliability, reduce maintenance costs, and ensure operational safety. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for intelligent operation and maintenance and predictive maintenance of a multi-layer architecture of a rail transit platform door, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention, Provides multi-layer intelligent operation and maintenance and predictive maintenance methods for rail transit platform doors, including: Establish a multi-layer data collection system for rail transit platform doors. Vibration sensors, temperature sensors, displacement sensors and current sensors installed on the bottom mechanism, drive mechanism and door structure of the platform door are used to collect the operating parameters of the platform door. The operating parameters include vibration frequency, drive current, component temperature and door opening and closing displacement. The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing to generate the health status feature vector of the platform door. The platform door health status feature vector is input into a deep neural network including a convolutional layer, a pooling layer and a fully connected layer, and the feature vector is subjected to time series correlation analysis and multi-scale feature extraction, and the deep neural network is trained in combination with historical maintenance data to construct a platform door fault prediction model based on deep learning; Based on the output results of the platform door fault prediction model based on deep learning, a decision tree algorithm is used to dynamically evaluate the maintenance priority of platform door components, and an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources is generated. Maintenance instructions are pushed to on-site maintenance personnel through mobile terminals, and new data collected during the maintenance process are fed back to the deep learning model for online optimization and updating, thereby realizing adaptive adjustment of the platform door maintenance strategy.

[0008] The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing, and the platform door health status feature vector is generated, including: The platform door operation parameters are collected by a vibration sensor arranged on the bottom mechanism of the platform door, a current sensor arranged on the platform door driving mechanism, a temperature sensor and a displacement sensor arranged on the platform door body structure, wherein the sampling frequency of the vibration sensor is 1000Hz, the sampling frequency of the temperature sensor is 1Hz, the sampling frequency of the displacement sensor is 100Hz, and the sampling frequency of the current sensor is 500Hz; the vibration sensor, the temperature sensor, the displacement sensor and the current sensor are divided into a mechanical group, a driving group and a door body group, each group is configured with an independent CAN controller and a CAN transceiver, and a differential signal transmission method is adopted to transmit the operation parameters to the edge computing unit through time division multiplexing at a transmission rate of 250kbps; The edge computing unit adopts a dual-core ARM processor, uses the db4 wavelet basis function to perform a three-layer decomposition of the operating parameters, filters out high-frequency noise through a soft threshold method, identifies and removes outliers based on the 3σ criterion, and uses a piecewise linear interpolation method to align the operating parameters to a 100Hz time axis; extracts time domain features and frequency domain features of the operating parameters, the time domain features include mean, standard deviation, peak factor and margin factor, the frequency domain features include main frequency component and frequency band energy distribution, and calculates the correlation coefficient between the operating parameters; The velocity curve characteristics, acceleration characteristics and displacement characteristics in the process of opening and closing the platform door are extracted as the operating condition characteristics. The envelope spectrum of the vibration signal and the harmonic ratio of the current signal are calculated through a sliding time window as the performance degradation characteristics. The environmental impact characteristics are established by combining the correlation analysis of the temperature change trend and the vibration response. The operating condition characteristics, the performance degradation characteristics and the environmental impact characteristics are combined to form a 145-dimensional feature vector. The principal component analysis method is used to reduce the dimension of the 145-dimensional feature vector, and the principal components with a cumulative contribution rate of 95% are selected to generate the platform door health status feature vector.

[0009] The platform door fault prediction model based on deep learning includes time series feature branch, spatial feature branch and attention feature branch; The temporal feature branch includes three one-dimensional convolutional layers connected in series, the convolution kernel sizes of the three one-dimensional convolutional layers are 3, 5, and 7 respectively, the output of each one-dimensional convolutional layer is connected to a maximum pooling layer, the pooling window size of the maximum pooling layer is 2, and the output of the temporal feature branch is used to characterize the temporal features of the health status feature vector; The spatial feature branch includes four two-dimensional convolutional layers connected in series, each of the four two-dimensional convolutional layers uses a 4×4 convolution kernel, the number of channels of the four two-dimensional convolutional layers is 32, 64, 128, and 256, respectively, and the output of the spatial feature branch is used to characterize the spatial features of the health status feature vector; The attention feature branch includes a self-attention calculation unit, which calculates the association weights between the dimensions of the health status feature vector, and normalizes the association weights by a softmax function to obtain normalized weights, and the output of the attention feature branch is used to characterize the attention features of the health status feature vector; The temporal feature, the spatial feature and the attention feature are fused to obtain a fused feature, and the fused feature is used for platform door failure prediction.

[0010] Building a deep learning-based platform door failure prediction model includes: Obtain the platform door health state feature vector, and perform recursive graph analysis on the platform door health state feature vector, map the platform door health state feature vector to a high-dimensional phase space by setting the phase space reconstruction parameters with an embedding dimension of 5 and a time delay of 2 to obtain feature trajectory points, and calculate the Euclidean distance between the feature trajectory points to construct a recursive matrix; The platform door health state feature vector is decomposed into time-frequency state by using complex Morlet wavelet, and the correlation coefficients at different scales are calculated; the platform door health state feature vector is decomposed into multiple integral mode functions by using local mode decomposition method, and the instantaneous frequency and instantaneous amplitude of the integral mode function are calculated to obtain micro-scale characteristics; The time-frequency energy spectrum of the platform door health status feature vector is constructed by combining empirical mode decomposition with Hilbert transform, and the marginal spectrum and instantaneous frequency spectrum are calculated to obtain the meso-scale characteristics. The platform door health status feature vector is decomposed into trend terms, periodic terms and residual terms through variational mode decomposition algorithm to obtain macro-scale characteristics. The micro-scale characteristics, the meso-scale characteristics and the macro-scale characteristics are input into a deep neural network for training to obtain a platform door fault prediction model.

[0011] Based on the output results of the platform door fault prediction model based on deep learning, a decision tree algorithm is used to dynamically evaluate the maintenance priority of platform door components, and an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources is generated, including: Based on the remaining life prediction value and the probability distribution of fault types output by the deep learning model, a maintenance priority evaluation index is constructed. The health status index is obtained by segmenting the remaining life prediction value according to the time interval. The fault risk index is calculated based on the product of the probability distribution of the fault type and the weight matrix of the degree of harm of the fault type. The operational impact index is calculated by the hierarchical analysis method in combination with the passenger flow at the platform gate, the station location and the backup resource status; The health status indicator, the fault risk indicator and the operation impact indicator are input into a CART decision tree, the minimum sample split number of the CART decision tree is set to 50, the minimum node purity threshold is set to 0.85, and the pruning parameter is determined to be 0.02 through cross-validation to obtain the maintenance priority evaluation result of the platform door component; Based on the maintenance priority evaluation result, an optimization objective function of time cost and risk cost is established, and the maintenance time window is calculated by genetic algorithm; based on the probability distribution of fault types, a maintenance plan is selected from the maintenance process library, and the maintenance project combination is determined by mining historical maintenance records through association rules; the maintenance time window and maintenance project combination of the maintenance plan are simulated and verified by discrete event simulation method to obtain an intelligent maintenance plan for platform doors.

[0012] Based on the maintenance priority evaluation result, an optimization objective function of time cost and risk cost is established, and the maintenance time window calculated by genetic algorithm includes: Based on the maintenance priority evaluation result, a time cost function and a risk cost function are constructed, wherein the time cost function includes a weighted sum of the maintenance operation time cost, the operation delay cost and the equipment downtime cost, and the risk cost function includes the product of the failure probability, the failure loss cost and the exponential decay function of the maintenance time margin; The maintenance start time point is encoded by chromosome using real number encoding, the population size is set to 100, the evolutionary generation is set to 200, and the individual selection probability is calculated by the rank-based roulette selection method, which is calculated by the ratio of the current individual ranking to the total ranking number; an adaptive arithmetic crossover operation is used for the selected individuals, and the crossover probability of the adaptive arithmetic crossover operation decreases linearly with the increase of the evolutionary generation; A Gaussian mutation operation is performed on the individuals after the crossover, and the mutation probability and mutation amplitude of the Gaussian mutation operation decay exponentially with the increasing number of evolutionary generations. The five percent individuals with the highest fitness in each generation are retained and directly enter the next generation, and the simulated annealing algorithm is periodically used to perform local search optimization on the five percent individuals with the highest fitness to obtain the optimal maintenance time window for the platform door.

[0013] According to a second aspect of the embodiments of the present invention, Provides a multi-layer intelligent operation and maintenance system for rail transit platform doors and predictive maintenance, including: The first unit is used to establish a multi-layer data collection system for rail transit platform doors. The platform door operating parameters are collected through vibration sensors, temperature sensors, displacement sensors and current sensors installed on the bottom mechanism, driving mechanism and door structure of the platform door. The operating parameters include vibration frequency, driving current, component temperature and door opening and closing displacement. The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing to generate a platform door health status feature vector; The second unit is used to input the platform door health status feature vector into a deep neural network including a convolution layer, a pooling layer and a fully connected layer, perform time series correlation analysis and multi-scale feature extraction on the feature vector, train the deep neural network in combination with historical maintenance data, and construct a platform door fault prediction model based on deep learning; The third unit is used to dynamically evaluate the maintenance priority of platform door components based on the output results of the platform door fault prediction model based on deep learning, using a decision tree algorithm, generate an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources, push maintenance instructions to on-site maintenance personnel through mobile terminals, and feed back new data collected during the maintenance process to the deep learning model for online optimization and updating, so as to achieve adaptive adjustment of the platform door maintenance strategy.

[0014] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0016] The beneficial effects of this application are as follows: 1. Multi-layer data collection and preprocessing: By installing multiple sensors at key locations of the platform door, a comprehensive data collection system has been established to monitor the operating status of the platform door in real time. The edge computing unit is used for data preprocessing to generate health status feature vectors, providing high-quality input data for subsequent fault prediction, thereby improving the accuracy and real-time performance of fault prediction.

[0017] 2. Deep learning fault prediction model: A deep neural network consisting of convolutional layers, pooling layers, and fully connected layers was used to perform time series correlation analysis and multi-scale feature extraction on the feature vector of the platform door health status. A fault prediction model based on deep learning was constructed by combining historical maintenance data for training. This method can effectively capture the complex patterns of the platform door operating status, improve the accuracy and predictability of fault prediction, and provide a reliable decision-making basis for preventive maintenance.

[0018] 3. Intelligent maintenance plan generation and adaptive optimization: Based on the output of the fault prediction model, a decision tree algorithm is used to dynamically evaluate maintenance priorities and generate intelligent maintenance plans. Maintenance instructions are pushed to maintenance personnel through mobile terminals, which improves maintenance efficiency. At the same time, new data collected during maintenance is fed back to the deep learning model to achieve online optimization and update of the model, so that the maintenance strategy can be adaptively adjusted. This closed-loop optimization mechanism ensures the continuous improvement of maintenance strategies and improves the overall reliability and operating efficiency of the platform door system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of a method for intelligent operation and maintenance and predictive maintenance of a multi-layer architecture of a rail transit platform door according to an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a multi-layer architecture intelligent operation and maintenance and predictive maintenance system for rail transit platform doors according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0022] Figure 1 FIG. 1 is a flow chart of a method for intelligent operation and maintenance and predictive maintenance of a multi-layer architecture of a rail transit platform door according to an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Establish a multi-layer data collection system for rail transit platform doors, collect platform door operating parameters through vibration sensors, temperature sensors, displacement sensors and current sensors installed on the bottom mechanism, drive mechanism and door structure of the platform door, the operating parameters include vibration frequency, drive current, component temperature and door opening and closing displacement, transmit the operating parameters to the edge computing unit through the field bus for data preprocessing, and generate a platform door health status feature vector; S102. Input the platform door health status feature vector into a deep neural network including a convolutional layer, a pooling layer, and a fully connected layer, perform time series correlation analysis and multi-scale feature extraction on the feature vector, train the deep neural network in combination with historical maintenance data, and construct a platform door fault prediction model based on deep learning; S103. Based on the output results of the platform door fault prediction model based on deep learning, a decision tree algorithm is used to dynamically evaluate the maintenance priority of the platform door components, and an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources is generated. Maintenance instructions are pushed to on-site maintenance personnel through mobile terminals, and new data collected during the maintenance process are fed back to the deep learning model for online optimization and updating, so as to achieve adaptive adjustment of the platform door maintenance strategy.

[0023] The specific implementation methods of the multi-layer architecture intelligent operation and maintenance and predictive maintenance method of rail transit platform doors are as follows: First, a multi-layer data collection system for rail transit platform doors is established. Vibration sensors, temperature sensors, displacement sensors and current sensors are installed on the bottom mechanism, drive mechanism and door structure of the platform door respectively. The vibration sensor collects the vibration frequency data of each component of the platform door, ranging from 0-1000Hz. The temperature sensor collects the operating temperature of key components such as the drive motor and controller, ranging from -20℃ to 80℃. The displacement sensor collects the displacement data during the opening and closing of the platform door with an accuracy of 0.1mm. The current sensor collects the working current of the drive motor, ranging from 0-20A. These sensors transmit the collected data to the edge computing unit on site in real time through the RS485 bus.

[0024] The edge computing unit preprocesses the received raw data, including denoising, filtering, normalization and other operations. Then it extracts key features, such as the mean, variance, peak value and other statistical quantities of vibration frequency, the maximum value and average value of temperature, the maximum value and rate of change of displacement, the effective value and harmonic content of current, etc. These features are combined to form the health status feature vector of the platform door, and a feature vector data is generated every 5 minutes.

[0025] Next, the generated health status feature vector is input into the pre-trained deep neural network. The network contains 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The convolutional layer uses a 3x3 convolution kernel to perform convolution operations on the input features to extract local features. The pooling layer uses a 2x2 maximum pooling to reduce the feature dimension. The fully connected layer maps the features to the fault type space. The input of the network is a 72x1 feature vector (144 sampling points in 24 hours), and the output is the probability distribution of 9 common fault types of platform doors.

[0026] During network training, 500,000 platform door operation data and corresponding maintenance records accumulated over three years were used as training sets. The network parameters were optimized using the stochastic gradient descent method, with the initial learning rate of 0.01, which decreased by 10% every 50 epochs. After 200 epochs of training, the fault prediction accuracy on the test set reached 92%.

[0027] Based on the trained deep learning model, the real-time input platform door health status feature vector is used to predict faults and obtain the probability of various faults occurring in the next 24 hours. The prediction results are input into the decision tree algorithm, which considers factors such as fault probability, fault impact, and maintenance difficulty to quantitatively evaluate the maintenance priority of each component. The evaluation result is a score of 1-10, with 10 indicating the highest priority.

[0028] According to the maintenance priority output by the decision tree, the system automatically generates an intelligent maintenance plan. The plan includes a specific maintenance schedule, a list of maintenance items, and the required maintenance resources. For example, for a drive motor failure with a priority of 9, the system arranges preventive maintenance within 24 hours. Maintenance items include replacing bearings, cleaning coils, etc. The required resources include 2 maintenance workers, 1 set of special tools, etc.

[0029] The generated maintenance plan is pushed to the on-site maintenance personnel through the mobile APP. The APP interface intuitively displays the priority, specific content and required tools of the maintenance task. The maintenance personnel can feedback the fault phenomenon and maintenance operation during the maintenance process through the APP. These newly collected data will be used to further optimize the deep learning model, such as adding new fault samples, adjusting model parameters, etc., so as to continuously improve the prediction accuracy of the model.

[0030] The beneficial effects of this method are mainly reflected in the following three aspects: 1. It realizes intelligent monitoring and predictive maintenance of the platform door throughout its life cycle, greatly improving equipment reliability and operating efficiency. By using multiple sensors to monitor key parameters in real time and combining deep learning algorithms for fault prediction, potential faults can be discovered in a timely manner to avoid serious consequences caused by equipment running with problems.

[0031] 2. Significantly reduced the maintenance cost and manpower requirements of platform doors. The traditional regular maintenance mode has been changed to predictive maintenance based on equipment status, reducing unnecessary maintenance work. The push of intelligent maintenance solutions has greatly improved maintenance efficiency and reduced dependence on highly skilled maintenance personnel.

[0032] 3. A sustainable and optimized intelligent operation and maintenance system has been built. By continuously collecting operation data and maintenance feedback, the deep learning model is continuously optimized to continuously improve the fault prediction capability. The adaptive adjustment of the system ensures that the maintenance strategy always remains in the optimal state, providing strong support for the long-term safe and efficient operation of rail transit.

[0033] In an optional implementation, the operating parameters are transmitted to the edge computing unit via the field bus for data preprocessing, and generating the platform door health status feature vector includes: The platform door operation parameters are collected by a vibration sensor arranged on the bottom mechanism of the platform door, a current sensor arranged on the platform door driving mechanism, a temperature sensor and a displacement sensor arranged on the platform door body structure, wherein the sampling frequency of the vibration sensor is 1000Hz, the sampling frequency of the temperature sensor is 1Hz, the sampling frequency of the displacement sensor is 100Hz, and the sampling frequency of the current sensor is 500Hz; the vibration sensor, the temperature sensor, the displacement sensor and the current sensor are divided into a mechanical group, a driving group and a door body group, each group is configured with an independent CAN controller and a CAN transceiver, and a differential signal transmission method is adopted to transmit the operation parameters to the edge computing unit through time division multiplexing at a transmission rate of 250kbps; The edge computing unit adopts a dual-core ARM processor, uses the db4 wavelet basis function to perform a three-layer decomposition of the operating parameters, filters out high-frequency noise through a soft threshold method, identifies and removes outliers based on the 3σ criterion, and uses a piecewise linear interpolation method to align the operating parameters to a 100Hz time axis; extracts time domain features and frequency domain features of the operating parameters, the time domain features include mean, standard deviation, peak factor and margin factor, the frequency domain features include main frequency component and frequency band energy distribution, and calculates the correlation coefficient between the operating parameters; The velocity curve characteristics, acceleration characteristics and displacement characteristics in the process of opening and closing the platform door are extracted as the operating condition characteristics. The envelope spectrum of the vibration signal and the harmonic ratio of the current signal are calculated through a sliding time window as the performance degradation characteristics. The environmental impact characteristics are established by combining the correlation analysis of the temperature change trend and the vibration response. The operating condition characteristics, the performance degradation characteristics and the environmental impact characteristics are combined to form a 145-dimensional feature vector. The principal component analysis method is used to reduce the dimension of the 145-dimensional feature vector, and the principal components with a cumulative contribution rate of 95% are selected to generate the platform door health status feature vector.

[0034] The specific implementation is as follows: First, a variety of sensors are installed in the platform door system to collect operating parameters. A vibration sensor is installed on the bottom mechanism of the platform door with a sampling frequency of 1000Hz to monitor the vibration of the door body during operation. A current sensor is installed on the platform door drive mechanism with a sampling frequency of 500Hz to monitor the current changes of the drive motor. A temperature sensor is installed on the platform door body structure with a sampling frequency of 1Hz to monitor the temperature changes of the door body. At the same time, a displacement sensor is installed on the door body with a sampling frequency of 100Hz to monitor the position changes of the door body during opening and closing.

[0035] The above sensors are divided into three groups: the mechanical group includes vibration sensors, the drive group includes current sensors, and the door group includes temperature sensors and displacement sensors. Each group is equipped with an independent CAN controller and CAN transceiver, and differential signal transmission is used to improve anti-interference ability. Using time division multiplexing technology, the operating parameters collected by each sensor are transmitted to the edge computing unit through the field bus at a transmission rate of 250kbps.

[0036] The edge computing unit uses a dual-core ARM processor with strong real-time data processing capabilities. First, the received operating parameters are preprocessed. The db4 wavelet basis function is used to perform a three-layer wavelet decomposition on the operating parameters to separate different frequency components. Then, the soft threshold method is used to perform threshold processing on the high-frequency components to filter out high-frequency noise. Outliers are identified based on the 3σ criterion, and data points that exceed the range of plus or minus 3 times the standard deviation of the mean value are marked as outliers and removed.

[0037] Next, the piecewise linear interpolation method is used to align the operating parameters of different sampling frequencies to the 100Hz time axis. For example, for a vibration signal sampled at 1000Hz, one point is taken for every 10 data points; for a temperature signal sampled at 1Hz, 99 data points are obtained by linear interpolation between two adjacent sampling points. This allows different parameters to be aligned in time, facilitating subsequent analysis.

[0038] Then, the time domain and frequency domain features are extracted from the preprocessed operating parameters. The time domain features include: mean, which is used to characterize the overall level of the signal; standard deviation, which is used to characterize the degree of fluctuation of the signal; peak factor, which is used to characterize the impact of the signal; margin factor, which is used to characterize the stability of the signal. The frequency domain features include: main frequency component, that is, the frequency component with the largest amplitude in the signal spectrum; frequency band energy distribution, which divides the spectrum into several frequency bands and calculates the energy proportion of each frequency band. At the same time, the correlation coefficient between different operating parameters is calculated to analyze the relationship between the parameters.

[0039] The characteristic parameters of the platform door opening and closing process are extracted as working condition characteristics. The speed curve characteristics include maximum speed, average speed, etc.; the acceleration characteristics include maximum acceleration, acceleration time, etc.; the displacement characteristics include door opening time, door closing time, etc. By setting a sliding time window, such as 5 seconds, the envelope spectrum of the vibration signal in the window is calculated, and the characteristic frequency and amplitude in the envelope spectrum are extracted. The ratio of each harmonic to the fundamental wave of the current signal is calculated as the harmonic ratio feature. These characteristics can reflect the performance degradation of the equipment.

[0040] Combined with the correlation analysis of temperature change trend and vibration response, the environmental impact characteristics are established. For example, the correlation coefficient between the temperature change rate and the root mean square value of vibration is calculated to reflect the degree of influence of temperature on vibration. The operating condition characteristics, performance degradation characteristics and environmental impact characteristics are combined to form a 145-dimensional feature vector.

[0041] Finally, the principal component analysis method is used to reduce the dimensionality of the 145-dimensional eigenvector. The covariance matrix of the eigenvector is calculated, and its eigenvalues ​​and eigenvectors are solved. The eigenvalues ​​are sorted from large to small, and the first few principal components with a cumulative contribution rate of 95% are selected. For example, if the cumulative contribution rate of the first 10 principal components reaches 95%, the 145-dimensional eigenvector is reduced to 10 dimensions to generate the platform door health status eigenvector.

[0042] The beneficial effects of this method include: 1. Through the coordinated monitoring and group transmission of multiple sensors, the platform door operation parameters are comprehensively collected, which improves the integrity and reliability of the data.

[0043] 2. The edge computing unit is used for data preprocessing and feature extraction, which reduces the amount of data transmission and improves the real-time and scalability of the system.

[0044] 3. Combining the time domain and frequency domain characteristics, operating condition characteristics, performance degradation characteristics and environmental impact characteristics, a multi-dimensional health status feature vector is constructed to comprehensively reflect the health status of the platform door, providing an effective basis for subsequent fault diagnosis and predictive maintenance.

[0045] In an optional implementation, the deep learning platform door fault prediction model includes a time series feature branch, a spatial feature branch, and an attention feature branch; The temporal feature branch includes three one-dimensional convolutional layers connected in series, the convolution kernel sizes of the three one-dimensional convolutional layers are 3, 5, and 7 respectively, the output of each one-dimensional convolutional layer is connected to a maximum pooling layer, the pooling window size of the maximum pooling layer is 2, and the output of the temporal feature branch is used to characterize the temporal features of the health status feature vector; The spatial feature branch includes four two-dimensional convolutional layers connected in series, each of the four two-dimensional convolutional layers uses a 4×4 convolution kernel, the number of channels of the four two-dimensional convolutional layers is 32, 64, 128, and 256, respectively, and the output of the spatial feature branch is used to characterize the spatial features of the health status feature vector; The attention feature branch includes a self-attention calculation unit, which calculates the association weights between the dimensions of the health status feature vector, and normalizes the association weights by a softmax function to obtain normalized weights, and the output of the attention feature branch is used to characterize the attention features of the health status feature vector; The temporal feature, the spatial feature and the attention feature are fused to obtain a fused feature, and the fused feature is used for platform door failure prediction.

[0046] The platform door fault prediction model based on deep learning includes time series feature branch, spatial feature branch and attention feature branch. First, the health status feature vector of the platform door is obtained, which contains the data collected by multiple sensors in a certain period of time.

[0047] The temporal feature branch is used to extract the temporal features of the health status feature vector. This branch contains three one-dimensional convolutional layers in series, with convolution kernel sizes of 3, 5, and 7 respectively. The output of each one-dimensional convolutional layer is connected to a maximum pooling layer with a pooling window size of 2. In specific implementation, the first one-dimensional convolutional layer uses three one-dimensional convolution kernels of size 3 to convolve the input features, and then downsamples them through the maximum pooling layer. The second one-dimensional convolutional layer uses five one-dimensional convolution kernels of size 5 for convolution, and then undergoes maximum pooling. The third one-dimensional convolutional layer uses seven one-dimensional convolution kernels of size 7 for convolution, and finally undergoes maximum pooling. This multi-scale convolution kernel design can capture feature patterns of different time spans.

[0048] The spatial feature branch is used to extract the spatial features of the health status feature vector. This branch contains four cascaded two-dimensional convolutional layers, all of which use 4×4 convolution kernels, and the number of channels is 32, 64, 128, and 256, respectively. In specific implementation, the first two-dimensional convolutional layer uses 32 4×4 convolution kernels to convolve the input feature map and outputs a 32-channel feature map. The second layer uses 64 4×4 convolution kernels and outputs a 64-channel feature map. The third layer uses 128 4×4 convolution kernels and outputs a 128-channel feature map. The fourth layer uses 256 4×4 convolution kernels and finally outputs a 256-channel feature map. This design of increasing the number of channels layer by layer can gradually extract higher-level spatial features.

[0049] The attention feature branch is used to calculate the correlation between the dimensions of the health status feature vector. This branch contains a self-attention calculation unit, which first calculates the correlation weights between the dimensions of the feature vector. In specific implementation, the feature vector can be converted into a query matrix, a key matrix, and a value matrix, and the attention score is calculated by matrix multiplication. Then, the softmax function is used to normalize the correlation weights to obtain the normalized attention weights. Finally, the attention weights are multiplied by the value matrix to obtain the attention feature.

[0050] The temporal features, spatial features and attention features are fused to obtain fused features. In specific implementation, the three features can be flattened into a one-dimensional vector, then concatenated together, and then fused through a fully connected layer to obtain the final fused feature representation.

[0051] Platform door fault prediction is performed based on fused features. The fused features can be input into a classifier (such as a softmax classifier) ​​to predict the fault type and fault probability of the platform door. The fused features can also be input into a regression model to predict the remaining service life of the platform door.

[0052] In specific applications, sensor data of platform doors in normal operation and various fault states can be collected to build a training data set. The deep learning model is trained with labeled data so that it can learn the characteristic patterns under different states. In actual prediction, the health status feature vector of the current platform door is input, and the model can output the fault prediction result.

[0053] This technical solution has the following beneficial effects: 1. Multi-branch feature extraction fully exploits the temporal, spatial and attention information of health status features and improves the feature representation capability.

[0054] 2. The multi-scale convolution kernel design of the temporal feature branch can capture patterns of different time spans and enhance the ability to model temporal dependencies.

[0055] 3. The attention mechanism can adaptively focus on important features, improving the interpretability and prediction accuracy of the model.

[0056] In an optional implementation, the platform door failure prediction model based on deep learning includes: Obtain the platform door health state feature vector, and perform recursive graph analysis on the platform door health state feature vector, map the platform door health state feature vector to a high-dimensional phase space by setting the phase space reconstruction parameters with an embedding dimension of 5 and a time delay of 2 to obtain feature trajectory points, and calculate the Euclidean distance between the feature trajectory points to construct a recursive matrix; The platform door health state feature vector is decomposed into time-frequency state by using complex Morlet wavelet, and the correlation coefficients at different scales are calculated; the platform door health state feature vector is decomposed into multiple integral mode functions by using local mode decomposition method, and the instantaneous frequency and instantaneous amplitude of the integral mode function are calculated to obtain micro-scale characteristics; The time-frequency energy spectrum of the platform door health status feature vector is constructed by combining empirical mode decomposition with Hilbert transform, and the marginal spectrum and instantaneous frequency spectrum are calculated to obtain the meso-scale characteristics. The platform door health status feature vector is decomposed into trend terms, periodic terms and residual terms through variational mode decomposition algorithm to obtain macro-scale characteristics. The micro-scale characteristics, the meso-scale characteristics and the macro-scale characteristics are input into a deep neural network for training to obtain a platform door fault prediction model.

[0057] The specific implementation is as follows: First, the characteristic vector of the platform door health status is obtained. The characteristic vector contains multi-dimensional data such as door position, door speed, door acceleration, motor current, and air pressure. The characteristic vector is analyzed by recursive graph, and the phase space reconstruction parameters with an embedding dimension of 5 and a time delay of 2 are set. The characteristic vector is mapped to the 5-dimensional phase space to obtain the characteristic trajectory points. The Euclidean distance between the characteristic trajectory points is calculated to construct a recursive matrix. For example, for the characteristic vector [10, 2, 0.5, 1.2, 80] at a certain moment, the trajectory points [10, 2, 0.5, 1.2, 80; 2, 0.5, 1.2, 80, 11; 0.5, 1.2,80, 11, 2.1; ...] are obtained after phase space reconstruction. The Euclidean distance between these trajectory points is calculated to obtain the recursive matrix.

[0058] Then, the complex Morlet wavelet is used to perform time-frequency decomposition of the eigenvector. Five different scales are selected for decomposition, and five groups of coefficients are obtained. The mutual correlation coefficients between these five groups of coefficients are calculated to obtain a 5x5 mutual correlation matrix. Then, the local mode decomposition method is used to decompose the eigenvector into five integral mode functions. The instantaneous frequency and instantaneous amplitude are calculated for each integral mode function to obtain the microscopic scale characteristics. For example, for a certain integral mode function, the instantaneous frequency sequence [10, 12, 11, 13, ...] Hz and the instantaneous amplitude sequence [0.5, 0.6, 0.4, 0.7, ...] are calculated.

[0059] Then, the eigenvector is constructed into a time-frequency energy spectrum by combining empirical mode decomposition with Hilbert transform. The marginal spectrum and instantaneous frequency spectrum are calculated to obtain the mesoscale features. For example, the marginal spectrum [100, 80, 60,40, 20] and the instantaneous frequency spectrum [15, 14, 13, 12,11] Hz are calculated. Then, the variational mode decomposition algorithm is used to decompose the eigenvector into a trend term, three periodic terms and a residual term to obtain the macroscale features. For example, the trend term is [10,11, 12, 13,...], the periodic term 1 is [1, 0,-1, 0, ...], the periodic term 2 is [0.5, 0, -0.5,0, ...], the periodic term 3 is [0.2,0, -0.2, 0,...], and the residual term is [0.1, -0.1, 0.2, -0.2, ...].

[0060] Finally, the micro-scale features, meso-scale features, and macro-scale features are used as input and sent to a 5-layer deep neural network for training. The network structure is: 64 nodes in the input layer, 3 hidden layers with 32, 16, and 8 nodes respectively, and 2 nodes in the output layer represent the failure probability and the remaining service life respectively. The Adam optimizer is used with a learning rate of 0.001 and 500 rounds of training. Finally, the platform door fault prediction model is obtained, which can predict the probability of failure and the remaining service life.

[0061] The beneficial effects of this method include: 1) Through multi-scale feature extraction, the health status of the platform door system is comprehensively characterized, and the accuracy of fault prediction is improved.

[0062] 2) The nonlinear dynamic characteristics of the platform door system are effectively extracted by using methods such as recursive graph and time-frequency analysis, which enhances the generalization ability of the model.

[0063] 3) Combined with deep learning methods, end-to-end fault prediction is achieved, which has stronger adaptability and robustness than traditional methods.

[0064] In an optional implementation, based on the output of the platform door fault prediction model of deep learning, a decision tree algorithm is used to dynamically evaluate the maintenance priority of the platform door components, and an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources is generated, including: Based on the remaining life prediction value and the probability distribution of fault types output by the deep learning model, a maintenance priority evaluation index is constructed. The health status index is obtained by segmenting the remaining life prediction value according to the time interval. The fault risk index is calculated based on the product of the probability distribution of the fault type and the weight matrix of the degree of harm of the fault type. The operational impact index is calculated by the hierarchical analysis method in combination with the passenger flow at the platform gate, the station location and the backup resource status; The health status indicator, the fault risk indicator and the operation impact indicator are input into a CART decision tree, the minimum sample split number of the CART decision tree is set to 50, the minimum node purity threshold is set to 0.85, and the pruning parameter is determined to be 0.02 through cross-validation to obtain the maintenance priority evaluation result of the platform door component; Based on the maintenance priority evaluation result, an optimization objective function of time cost and risk cost is established, and the maintenance time window is calculated by genetic algorithm; based on the probability distribution of fault types, a maintenance plan is selected from the maintenance process library, and the maintenance project combination is determined by mining historical maintenance records through association rules; the maintenance time window and maintenance project combination of the maintenance plan are simulated and verified by discrete event simulation method to obtain an intelligent maintenance plan for platform doors.

[0065] The present invention provides a method for generating an intelligent maintenance plan for platform doors based on deep learning and decision trees. The method first uses a deep learning model to predict the remaining life and probability distribution of fault types of platform doors, then constructs a maintenance priority evaluation index based on the prediction results, evaluates the maintenance priority of platform door components through a decision tree algorithm, and finally generates an intelligent maintenance plan including maintenance time, items, and resources. The specific implementation steps are as follows: First, based on the remaining life prediction value and fault type probability distribution output by the deep learning model, a maintenance priority evaluation index is constructed. The specific steps include: (1) The predicted remaining life span is segmented and mapped according to the time interval to obtain the health status index. For example, the predicted remaining life span can be divided into five intervals: [0, 30 days), [30 days, 90 days), [90 days, 180 days), [180 days, 365 days), [365 days, +∞), corresponding to health status indicators 1, 2, 3, 4, and 5, respectively.

[0066] (2) The fault risk index is calculated based on the product of the fault type probability distribution and the fault type hazard weight matrix. First, the fault type hazard weight matrix is ​​established. For example, for the five common fault types of platform doors (door leaf stuck, door leaf not in place, door leaf leakage, door lock failure, emergency unlocking device failure), weights are assigned 0.3, 0.2, 0.1, 0.3, and 0.1 respectively. Then, the fault type probability distribution is multiplied by the weight matrix to obtain the fault risk index.

[0067] (3) Combined with the passenger flow at the platform gate, the location of the station and the status of spare resources, the operational impact index is calculated through the hierarchical analysis method. First, a hierarchical structure model is constructed, with passenger flow, station location and spare resource status as three first-level indicators, and several second-level indicators. Then, the weight of each indicator is determined through expert scoring, and finally the comprehensive operational impact index is calculated.

[0068] Next, the health status index, fault risk index and operation impact index are input into the CART decision tree to evaluate the maintenance priority of the platform door components. The minimum sample split number of the CART decision tree is set to 50, the minimum node purity threshold is set to 0.85, and the pruning parameter is determined to be 0.02 through 10-fold cross validation. The output of the decision tree is the maintenance priority of the platform door components, which can be divided into three levels: "high", "medium" and "low".

[0069] Then, based on the maintenance priority evaluation results, the optimization objective function of time cost and risk cost is established, and the maintenance time window is calculated by genetic algorithm. The objective function can be expressed as the weighted sum of time cost and risk cost, where time cost considers the impact of maintenance time on operation, and risk cost considers the possible failure loss caused by delayed maintenance. The population size of the genetic algorithm is set to 100, the crossover probability is 0.8, the mutation probability is 0.1, and the number of iterations is 1000.

[0070] Next, the maintenance plan is selected from the maintenance process library based on the probability distribution of fault types, and the maintenance project combination is determined by mining historical maintenance records through association rules. First, the corresponding maintenance plan is selected from the maintenance process library according to the probability distribution of fault types. Then, the Apriori algorithm is used to mine association rules on historical maintenance records, with the support threshold set to 0.02 and the confidence threshold set to 0.6, to obtain the association relationship between maintenance projects, thereby determining a reasonable combination of maintenance projects.

[0071] Finally, the discrete event simulation method is used to simulate and verify the maintenance time window and maintenance project combination of the maintenance plan, and the intelligent maintenance plan for the platform door is obtained. During the simulation process, resource constraints such as maintenance personnel, equipment, and materials are considered, and random events in the maintenance process, such as fluctuations in maintenance time and maintenance quality, are simulated. Through multiple simulation experiments, the effects of different maintenance plans are compared, and the optimal intelligent maintenance plan is selected.

[0072] The beneficial effects of this method are mainly reflected in the following three aspects: (1) Improved the scientificity and accuracy of maintenance decisions. The deep learning model is used to predict the remaining life of the platform door and the probability distribution of the fault type. Combined with multi-dimensional evaluation indicators, the decision tree algorithm is used to evaluate the maintenance priority of the platform door components, which effectively improves the scientificity and accuracy of maintenance decisions.

[0073] (2) The optimal configuration of maintenance resources is achieved. Based on the maintenance priority evaluation results, the maintenance time window is optimized through genetic algorithm, and the reasonable combination of maintenance projects is determined by association rule mining, which achieves the optimal configuration of maintenance resources and improves maintenance efficiency.

[0074] (3) Enhanced the reliability and adaptability of the maintenance plan. The discrete event simulation method is used to verify the maintenance plan, taking into account the random factors in the actual maintenance process, which enhances the reliability and adaptability of the maintenance plan and is conducive to improving the overall operational stability of the platform door system.

[0075] In an optional implementation, the maintenance priority evaluation result establishes an optimization objective function of time cost and risk cost, and the maintenance time window calculated by the genetic algorithm includes: Based on the maintenance priority evaluation result, a time cost function and a risk cost function are constructed, wherein the time cost function includes a weighted sum of the maintenance operation time cost, the operation delay cost and the equipment downtime cost, and the risk cost function includes the product of the failure probability, the failure loss cost and the exponential decay function of the maintenance time margin; The maintenance start time point is encoded by chromosome using real number encoding, the population size is set to 100, the evolutionary generation is set to 200, and the individual selection probability is calculated by the rank-based roulette selection method, which is calculated by the ratio of the current individual ranking to the total ranking number; an adaptive arithmetic crossover operation is used for the selected individuals, and the crossover probability of the adaptive arithmetic crossover operation decreases linearly with the increase of the evolutionary generation; A Gaussian mutation operation is performed on the individuals after the crossover, and the mutation probability and mutation amplitude of the Gaussian mutation operation decay exponentially with the increasing number of evolutionary generations. The five percent individuals with the highest fitness in each generation are retained and directly enter the next generation, and the simulated annealing algorithm is periodically used to perform local search optimization on the five percent individuals with the highest fitness to obtain the optimal maintenance time window for the platform door.

[0076] The specific implementation is as follows: First, the time cost function and risk cost function are constructed based on the maintenance priority evaluation results. The time cost function consists of three parts: maintenance operation time cost, operation delay cost and equipment downtime cost. These three parts of cost are weighted and summed according to different weights to obtain the total time cost. Among them, the maintenance operation time cost mainly considers labor cost and material cost, the operation delay cost mainly considers the economic loss caused by train delays due to maintenance, and the equipment downtime cost considers the opportunity cost of equipment stopping operation. The risk cost function is obtained by multiplying three factors: the probability of failure, the failure loss cost and the exponential decay function related to the maintenance time margin. The probability of failure can be obtained through historical data analysis, the failure loss cost is estimated according to the severity of different failure types, and the exponential decay function of the maintenance time margin is used to reflect the increased risk brought by delayed maintenance.

[0077] Next, a genetic algorithm is used to optimize the maintenance time window. First, chromosome encoding is performed, and the maintenance start time point is encoded using real number encoding. For example, if the maintenance window of a device is 0-24 hours, the chromosome can be encoded as a real number between 0 and 24. Set the population size to 100 and the evolutionary generation to 200 generations. In the selection operation, a rank-based roulette selection method is used. Specifically, the individuals in the population are first sorted from large to small according to the fitness value, and then the selection probability of each individual is calculated. The selection probability is equal to the ratio of the current individual ranking to the total ranking number. For example, if an individual ranks 10th and there are 100 individuals in total, its selection probability is 10 / 100=0.1.

[0078] After the selection is completed, the selected individuals are crossovered. Adaptive arithmetic crossover is used here, and the crossover probability decreases linearly with the increase of evolutionary generations. For example, the initial crossover probability can be set to 0.9, and the termination crossover probability can be set to 0.6, then the crossover probability of the tth generation can be expressed as 0.9-0.3t / 200. During the crossover process, two parent individuals are randomly selected, and a random number α between 0 and 1 is generated. If α is less than the crossover probability, a crossover operation is performed. The crossover method is arithmetic crossover, that is, the gene values ​​of the two offspring individuals are the weighted averages of the corresponding gene values ​​of the two parent individuals.

[0079] After the crossover is completed, the individuals are mutated. Here, Gaussian mutation operation is used, and the mutation probability and mutation amplitude both decay exponentially with the increase of evolutionary generations. For example, the initial mutation probability can be set to 0.1, and the termination mutation probability can be set to 0.001, then the mutation probability of the tth generation can be expressed as 0.1×0.99^t. The mutation amplitude also decays exponentially, the initial value can be set to 1, and the termination value can be set to 0.1. During the mutation process, each gene position is judged, and if the random number is less than the mutation probability, it is mutated. The mutation method is to superimpose a Gaussian random number with a mean of 0 and a standard deviation of the current mutation amplitude on the original gene value.

[0080] In order to improve the convergence performance of the algorithm, the elite retention strategy is adopted, that is, the 5% individuals with the highest fitness in each generation are retained and directly enter the next generation. At the same time, in order to avoid falling into the local optimum, the simulated annealing algorithm is periodically used to perform local search optimization on these elite individuals. Specifically, the elite individuals are optimized by simulated annealing every 10 generations. In the simulated annealing process, a neighborhood solution is generated with the current solution as the center. If the neighborhood solution is better than the current solution, it is accepted, otherwise the worse solution is accepted with a certain probability. The initial temperature is set to 100, the termination temperature is set to 1, and the cooling coefficient is 0.95.

[0081] Through the above steps, the optimal maintenance time window of the platform door can be obtained. For example, suppose a platform door system contains 10 key components, and the maintenance time window of each component is 0-24 hours. After optimization, the optimal maintenance time window may be: [2.5, 8.3, 12.7, 15.9, 18.2, 20.1, 21.8, 22.9, 23.5, 23.8]. This means that the optimal maintenance start time of the 10 key components is 2:30 am, 8:18 am, 12:42 pm, 3:54 pm, 6:12 pm, 8:06 pm, 9:48 pm, 10:54 pm, 11:30 pm and 11:48 pm.

[0082] The beneficial effects of this method are mainly reflected in the following three aspects: 1. By constructing the optimization objective function of time cost and risk cost, the maintenance cost and potential risks are fully considered, which can achieve a good balance between cost and safety and improve the scientificity and rationality of maintenance decisions.

[0083] 2. The genetic algorithm is used for optimization and solution, and combined with improvement strategies such as adaptive crossover and Gaussian mutation, it can effectively avoid falling into the local optimum, improve the global search ability and convergence speed of the algorithm, and thus obtain a better maintenance time window solution.

[0084] 3. The introduction of elite retention strategy and simulated annealing local search enhances the ability to develop the optimal solution while ensuring population diversity, further improves the quality of the optimization results, and makes the maintenance time window solution more reliable and efficient.

[0085] Figure 2 FIG. 1 is a schematic diagram of the structure of a multi-layer architecture intelligent operation and maintenance and predictive maintenance system for rail transit platform doors according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to establish a multi-layer data collection system for rail transit platform doors. The platform door operating parameters are collected through vibration sensors, temperature sensors, displacement sensors and current sensors installed on the bottom mechanism, driving mechanism and door structure of the platform door. The operating parameters include vibration frequency, driving current, component temperature and door opening and closing displacement. The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing to generate a platform door health status feature vector; The second unit is used to input the platform door health status feature vector into a deep neural network including a convolution layer, a pooling layer and a fully connected layer, perform time series correlation analysis and multi-scale feature extraction on the feature vector, train the deep neural network in combination with historical maintenance data, and construct a platform door fault prediction model based on deep learning; The third unit is used to dynamically evaluate the maintenance priority of platform door components based on the output results of the platform door fault prediction model based on deep learning, using a decision tree algorithm, generate an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources, push maintenance instructions to on-site maintenance personnel through mobile terminals, and feed back new data collected during the maintenance process to the deep learning model for online optimization and updating, so as to achieve adaptive adjustment of the platform door maintenance strategy.

[0086] According to a third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0087] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0088] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-layer architecture intelligent operation and maintenance and predictive maintenance method for rail transit platform doors, characterized in that: include: Establish a multi-layer data collection system for rail transit platform doors. Vibration sensors, temperature sensors, displacement sensors and current sensors installed on the bottom mechanism, drive mechanism and door structure of the platform door are used to collect the operating parameters of the platform door. The operating parameters include vibration frequency, drive current, component temperature and door opening and closing displacement. The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing to generate the health status feature vector of the platform door. The platform door health status feature vector is input into a deep neural network including a convolutional layer, a pooling layer and a fully connected layer, and the feature vector is subjected to time series correlation analysis and multi-scale feature extraction, and the deep neural network is trained in combination with historical maintenance data to construct a platform door fault prediction model based on deep learning; Based on the output results of the platform door fault prediction model based on deep learning, a decision tree algorithm is used to dynamically evaluate the maintenance priority of platform door components, and an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources is generated. Maintenance instructions are pushed to on-site maintenance personnel through mobile terminals, and new data collected during the maintenance process are fed back to the deep learning model for online optimization and updating, thereby realizing adaptive adjustment of the platform door maintenance strategy.

2. The method according to claim 1, characterized in that The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing, and the platform door health status feature vector is generated, including: The platform door operation parameters are collected by a vibration sensor arranged on the bottom mechanism of the platform door, a current sensor arranged on the platform door driving mechanism, a temperature sensor and a displacement sensor arranged on the platform door body structure, wherein the sampling frequency of the vibration sensor is 1000Hz, the sampling frequency of the temperature sensor is 1Hz, the sampling frequency of the displacement sensor is 100Hz, and the sampling frequency of the current sensor is 500Hz; the vibration sensor, the temperature sensor, the displacement sensor and the current sensor are divided into a mechanical group, a driving group and a door body group, each group is configured with an independent CAN controller and a CAN transceiver, and a differential signal transmission method is adopted to transmit the operation parameters to the edge computing unit through time division multiplexing at a transmission rate of 250kbps; The edge computing unit adopts a dual-core ARM processor, uses the db4 wavelet basis function to perform a three-layer decomposition of the operating parameters, filters out high-frequency noise through a soft threshold method, identifies and removes outliers based on the 3σ criterion, and uses a piecewise linear interpolation method to align the operating parameters to a 100Hz time axis; extracts time domain features and frequency domain features of the operating parameters, the time domain features include mean, standard deviation, peak factor and margin factor, the frequency domain features include main frequency component and frequency band energy distribution, and calculates the correlation coefficient between the operating parameters; The velocity curve characteristics, acceleration characteristics and displacement characteristics in the process of opening and closing the platform door are extracted as the operating condition characteristics. The envelope spectrum of the vibration signal and the harmonic ratio of the current signal are calculated through a sliding time window as the performance degradation characteristics. The environmental impact characteristics are established by combining the correlation analysis of the temperature change trend and the vibration response. The operating condition characteristics, the performance degradation characteristics and the environmental impact characteristics are combined to form a 145-dimensional feature vector. The principal component analysis method is used to reduce the dimension of the 145-dimensional feature vector, and the principal components with a cumulative contribution rate of 95% are selected to generate the platform door health status feature vector.

3. The method according to claim 1, characterized in that: The platform door fault prediction model based on deep learning includes time series feature branch, spatial feature branch and attention feature branch; The temporal feature branch includes three one-dimensional convolutional layers connected in series, the convolution kernel sizes of the three one-dimensional convolutional layers are 3, 5, and 7 respectively, the output of each one-dimensional convolutional layer is connected to a maximum pooling layer, the pooling window size of the maximum pooling layer is 2, and the output of the temporal feature branch is used to characterize the temporal features of the health status feature vector; The spatial feature branch includes four two-dimensional convolutional layers connected in series, each of the four two-dimensional convolutional layers uses a 4×4 convolution kernel, the number of channels of the four two-dimensional convolutional layers is 32, 64, 128, and 256, respectively, and the output of the spatial feature branch is used to characterize the spatial features of the health status feature vector; The attention feature branch includes a self-attention calculation unit, which calculates the association weights between the dimensions of the health status feature vector, and normalizes the association weights by a softmax function to obtain normalized weights, and the output of the attention feature branch is used to characterize the attention features of the health status feature vector; The temporal feature, the spatial feature and the attention feature are fused to obtain a fused feature, and the fused feature is used for platform door failure prediction.

4. The method according to claim 1, characterized in that: Building a deep learning-based platform door failure prediction model includes: Obtain the platform door health state feature vector, and perform recursive graph analysis on the platform door health state feature vector, map the platform door health state feature vector to a high-dimensional phase space by setting the phase space reconstruction parameters with an embedding dimension of 5 and a time delay of 2 to obtain feature trajectory points, and calculate the Euclidean distance between the feature trajectory points to construct a recursive matrix; The platform door health state feature vector is decomposed into time-frequency state by using complex Morlet wavelet, and the correlation coefficients at different scales are calculated; the platform door health state feature vector is decomposed into multiple integral mode functions by using local mode decomposition method, and the instantaneous frequency and instantaneous amplitude of the integral mode function are calculated to obtain micro-scale characteristics; The time-frequency energy spectrum of the platform door health status feature vector is constructed by combining empirical mode decomposition with Hilbert transform, and the marginal spectrum and instantaneous frequency spectrum are calculated to obtain the meso-scale characteristics. The platform door health status feature vector is decomposed into trend terms, periodic terms and residual terms through variational mode decomposition algorithm to obtain macro-scale characteristics. The micro-scale characteristics, the meso-scale characteristics and the macro-scale characteristics are input into a deep neural network for training to obtain a platform door fault prediction model.

5. The method according to claim 1, characterized in that Based on the output results of the platform door fault prediction model based on deep learning, a decision tree algorithm is used to dynamically evaluate the maintenance priority of platform door components, and an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources is generated, including: Based on the remaining life prediction value and the probability distribution of fault types output by the deep learning model, a maintenance priority evaluation index is constructed. The health status index is obtained by segmenting the remaining life prediction value according to the time interval. The fault risk index is calculated based on the product of the probability distribution of the fault type and the weight matrix of the degree of harm of the fault type. The operational impact index is calculated by the hierarchical analysis method in combination with the passenger flow at the platform gate, the station location and the backup resource status; The health status indicator, the fault risk indicator and the operation impact indicator are input into a CART decision tree, the minimum sample split number of the CART decision tree is set to 50, the minimum node purity threshold is set to 0.85, and the pruning parameter is determined to be 0.02 through cross-validation to obtain the maintenance priority evaluation result of the platform door component; Based on the maintenance priority evaluation result, an optimization objective function of time cost and risk cost is established, and the maintenance time window is calculated by genetic algorithm; based on the probability distribution of fault types, a maintenance plan is selected from the maintenance process library, and the maintenance project combination is determined by mining historical maintenance records through association rules; the maintenance time window and maintenance project combination of the maintenance plan are simulated and verified by discrete event simulation method to obtain an intelligent maintenance plan for platform doors.

6. The method according to claim 5, characterized in that Based on the maintenance priority evaluation result, an optimization objective function of time cost and risk cost is established, and the maintenance time window calculated by genetic algorithm includes: Based on the maintenance priority evaluation result, a time cost function and a risk cost function are constructed, wherein the time cost function includes a weighted sum of the maintenance operation time cost, the operation delay cost and the equipment downtime cost, and the risk cost function includes the product of the failure probability, the failure loss cost and the exponential decay function of the maintenance time margin; The maintenance start time point is encoded by chromosome using real number encoding, the population size is set to 100, the evolutionary generation is set to 200, and the individual selection probability is calculated by the rank-based roulette selection method, which is calculated by the ratio of the current individual ranking to the total ranking number; an adaptive arithmetic crossover operation is used for the selected individuals, and the crossover probability of the adaptive arithmetic crossover operation decreases linearly with the increase of the evolutionary generation; A Gaussian mutation operation is performed on the individuals after the crossover, and the mutation probability and mutation amplitude of the Gaussian mutation operation decay exponentially with the increasing number of evolutionary generations. The five percent individuals with the highest fitness in each generation are retained and directly enter the next generation, and the simulated annealing algorithm is periodically used to perform local search optimization on the five percent individuals with the highest fitness to obtain the optimal maintenance time window for the platform door.

7. A multi-layered intelligent operation and maintenance and predictive maintenance system for rail transit platform doors, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to establish a multi-layer data collection system for rail transit platform doors. The platform door operating parameters are collected through vibration sensors, temperature sensors, displacement sensors and current sensors installed on the bottom mechanism, driving mechanism and door structure of the platform door. The operating parameters include vibration frequency, driving current, component temperature and door opening and closing displacement. The operating parameters are transmitted to the edge computing unit through the field bus for data preprocessing to generate a platform door health status feature vector; The second unit is used to input the platform door health status feature vector into a deep neural network including a convolution layer, a pooling layer and a fully connected layer, perform time series correlation analysis and multi-scale feature extraction on the feature vector, train the deep neural network in combination with historical maintenance data, and construct a platform door fault prediction model based on deep learning; The third unit is used to dynamically evaluate the maintenance priority of platform door components based on the output results of the platform door fault prediction model based on deep learning, using a decision tree algorithm, generate an intelligent maintenance plan including maintenance time, maintenance items and maintenance resources, push maintenance instructions to on-site maintenance personnel through mobile terminals, and feed back new data collected during the maintenance process to the deep learning model for online optimization and updating, so as to achieve adaptive adjustment of the platform door maintenance strategy.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images