Platform screen door for rail transit

Abstract

A platform screen door for rail transit. By means of providing a monitoring sensor (12) and a fault early-warning device (13) for a screen door, real-time operation data of the screen door is acquired, and the real-time operation data is then input into a pre-constructed prediction model; on the basis of the prediction model, multi-round cyclic segmented prediction is performed on the real-time operation data, so as to output an operation prediction result at each time node within a target time period, wherein one piece of segmented-prediction data for the target time period is output by means of each round of prediction, and the obtained segmented-prediction data and the real-time operation data are used as model input data for the next round of prediction; and the operation prediction result is input into a pre-constructed fault monitoring model, and on the basis of the fault monitoring model, a fault type monitoring result at a corresponding time node is output. By means of the technical means, a potential fault type at a corresponding time node can be detected in a timely manner, such that accurate and efficient fault monitoring is realized, thereby improving the reliability and stability of operation of a screen door.

Description

Rail transit platform shielding door

[0001] The present application claims priority to the Chinese patent application No. 2024112963244 filed on September 18, 2024 with the China Patent Office, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] Embodiments of the present application relate to the technical field of shielding doors, and in particular to a rail transit platform shielding door. BACKGROUND

[0003] Currently, in the running process of the rail transit platform shielding door, the automatic opening and closing of the shielding door body are realized through a control module. In the shielding door operation and maintenance scene of the platform shielding door, the maintenance mode is mainly to monitor the running state of the shielding door, and then perform fault maintenance or maintenance operations. In order to give an early warning of the shielding door failure, a fault prediction model is used to analyze the equipment reliability according to the state information of the shielding door, and when the equipment reliability decreases to a set value, a warning of the equipment running failure is given, and potential running failures are monitored in time to ensure the stability of the shielding door operation.

[0004] However, the current shielding door failure monitoring method based on simple reliability analysis cannot accurately give the time and type of shielding door failure, resulting in a certain error in shielding door failure monitoring, and maintenance personnel cannot accurately determine the fault type and preventive maintenance time node according to the fault warning. The maintenance assistance effect of the shielding door is poor, which easily leads to over-maintenance or maintenance lag, and thus affects the reliability of the shielding door operation. SUMMARY

[0005] Embodiments of the present application provide a rail transit platform shielding door, which can accurately predict its own running failure and monitor potential running failure in time, and solve the problem of shielding door failure monitoring error.

[0006] In a first aspect, embodiments of the present application provide a rail transit platform shielding door, comprising:

[0007] a controller configured to control the shielding door body to perform opening and closing operations;

[0008] a monitoring sensor configured to collect real-time running data of the shielding door based on the opening and closing operations of the shielding door body;

[0009] The fault early warning device is used for acquiring the real-time operation data, inputting the real-time operation data into a pre-constructed prediction model, performing multi-round cyclic segmented prediction on the real-time operation data based on the prediction model, outputting an operation prediction result of each time node of a target period, taking the real-time operation data as initial model input data of the prediction model in the multi-round cyclic segmented prediction process, outputting one segmented prediction data of the target period in each round of prediction, and taking the predicted segmented prediction data and the real-time operation data as model input data of the next round of prediction, inputting the operation prediction result into a pre-constructed fault monitoring model, and outputting a fault type monitoring result of a corresponding time node based on the fault monitoring model, wherein the fault monitoring model is pre-trained based on fault operation data of a shielding door, and the fault operation data is pre-labeled with a corresponding fault type label.

[0010] The training process of the prediction model comprises:

[0011] The historical operation data of the shielding door is collected to construct a training sample, and the training sample is split into a first sequence and a second sequence in time sequence;

[0012] The first sequence is input into the prediction model to perform multi-round cyclic segmented prediction, and a corresponding prediction sequence is obtained;

[0013] A loss function is calculated according to the prediction sequence and the second sequence, and model parameters of the prediction model are iteratively adjusted based on the loss function until the prediction model converges.

[0014] Further, the historical operation data comprises shielding door opening and closing data, door body displacement data and door body speed data collected at different historical time nodes.

[0015] Further, the prediction model comprises a first multi-layer perception module, a decoder network and a second multi-layer perception module connected in sequence;

[0016] The first multi-layer perception module is used for converting model input data of the prediction model into a vector of a set shape;

[0017] The decoder network is used for generating an initial prediction sequence based on the vector of the set shape;

[0018] The second multi-layer perception module is used for converting the initial prediction sequence into the segmented prediction data according to a shape of the model input data.

[0019] Further, a calculation formula of the first multi-layer perception module is represented as: x out = relu(w·x+b)

[0020] wherein w represents the weight of the hidden layer of the first multi-layer perception module, b represents the bias term of the hidden layer, relu is a nonlinear activation function, x out is the output of the first multi-layer perception module, and x is the input of the first multi-layer perception module.

[0021] Further, before the real-time operation data is input into the pre-constructed prediction model, the method further comprises:

[0022] The real-time operation data is subjected to missing value filling and outlier screening processing, and the processed real-time operation data is subjected to standardization processing.

[0023] Further, the standardization processing is used to standardize the real-time operation data into sequence data with a mean value of 0 and a standard deviation of 1.

[0024] Further, the monitoring sensor is a monitoring camera, which determines the real-time operation data of the shield door by target detection and positioning on the shield door body, based on the detected shield door body position.

[0025] In a second aspect, the embodiments of the present application provide a fault monitoring method of a shield door, applied to the track transit platform shield door as described in the first aspect, comprising:

[0026] Obtaining real-time operation data of the shield door, and inputting the real-time operation data into a pre-constructed prediction model;

[0027] Based on the prediction model, the real-time operation data is subjected to multi-round cycle segmented prediction, and an operation prediction result of each time node of a target period is output, in which the real-time operation data is used as the initial model input data of the prediction model in the multi-round cycle segmented prediction process, one segmented prediction data of the target period is output in each round of prediction, and the predicted segmented prediction data and the real-time operation data are used as the model input data of the next round of prediction;

[0028] The operation prediction result is input into a pre-constructed fault monitoring model, and a fault type monitoring result of the corresponding time node is output based on the fault monitoring model, the fault monitoring model is pre-trained based on the fault operation data of the shield door, and the fault operation data is pre-labeled with a corresponding fault type label;

[0029] The training process of the prediction model comprises:

[0030] The historical operation data of the shield door is collected to construct a training sample, and the training sample is split into a first sequence and a second sequence in time sequence;

[0031] input the first sequence into the prediction model for multi-round cyclic segmented prediction to obtain a corresponding prediction sequence;

[0032] calculate a loss function according to the prediction sequence and the second sequence, and iteratively adjust model parameters of the prediction model based on the loss function until the prediction model converges.

[0033] In a third aspect, an embodiment of the present application provides a fault monitoring device of a shield door, comprising:

[0034] an input module configured to acquire real-time running data of the shield door and input the real-time running data into a pre-constructed prediction model;

[0035] a data prediction module configured to perform multi-round cyclic segmented prediction on the real-time running data based on the prediction model, output running prediction results of each time node of a target time period, use the real-time running data as initial model input data of the prediction model in the multi-round cyclic segmented prediction process, output one segmented prediction data of the target time period in each round of prediction, and use the predicted segmented prediction data and the real-time running data as model input data for the next round of prediction;

[0036] a fault monitoring module configured to input the running prediction results into a pre-constructed fault monitoring model, output fault type monitoring results of a corresponding time node based on the fault monitoring model, and pre-train the fault monitoring model based on fault running data of the shield door, wherein the fault running data is pre-labeled with corresponding fault type labels;

[0037] The training process of the prediction model comprises:

[0038] collecting historical running data of the shield door to construct training samples, and splitting the training samples into a first sequence and a second sequence according to time sequence;

[0039] inputting the first sequence into the prediction model for multi-round cyclic segmented prediction to obtain a corresponding prediction sequence;

[0040] calculating a loss function according to the prediction sequence and the second sequence, and iteratively adjusting model parameters of the prediction model based on the loss function until the prediction model converges.

[0041] In a fourth aspect, an embodiment of the present application provides a storage medium containing computer executable instructions, which are used to execute the fault monitoring method of the shield door as described in the second aspect when executed by a computer processor.

[0042] The embodiment of the application sets the monitoring sensor and the fault early warning device of the shielding door, obtains real-time running data of the shielding door, inputs the real-time running data into a pre-constructed prediction model, performs multi-round cycle segmented prediction on the real-time running data based on the prediction model, and outputs running prediction results of each time node in a target period. In the multi-round cycle segmented prediction process, the real-time running data is used as initial model input data of the prediction model, each round of prediction outputs segmented prediction data of the target period, and the predicted segmented prediction data and the real-time running data are used as model input data for the next round of prediction. The running prediction results are input into a pre-constructed fault monitoring model, and the fault monitoring model outputs fault type monitoring results of the corresponding time node based on the fault monitoring model. The fault monitoring model is pre-trained based on fault running data of the shielding door, and the fault running data is pre-labeled with corresponding fault type labels. By using the above technical means, the running prediction results of the shielding door at each time node in the future target period are predicted by cycle segmentation, and then the fault monitoring is performed based on the running prediction results, the potential fault type at the corresponding time node is found in time, the accurate and efficient fault monitoring is realized, and the reliability and stability of the shielding door running are improved. BRIEF DESCRIPTION OF DRAWINGS

[0043] Fig. 1 is a structural connection schematic diagram of a rail transit platform shielding door provided by an embodiment of the application;

[0044] Fig. 2 is a flowchart of a fault monitoring method of a shielding door provided by an embodiment of the application;

[0045] Fig. 3 is a training flowchart of a prediction model in an embodiment of the application;

[0046] Fig. 4 is a sequence prediction schematic diagram of a prediction model in an embodiment of the application;

[0047] Fig. 5 is a fault monitoring flowchart in an embodiment of the application;

[0048] Fig. 6 is a structural schematic diagram of a fault monitoring device of a shielding door provided by an embodiment of the application;

[0049] Fig. 7 is a structural schematic diagram of an electronic device provided by an embodiment of the application. DETAILED DESCRIPTION

[0050] In order to make the purposes, technical solutions and advantages of the present application clearer, the specific embodiments of the present application are further described in detail below in conjunction with the drawings. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only parts related to the present application are shown in the drawings, but not all. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowchart describes each operation (or step) as a sequential process, many of the operations can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operations are completed, but can also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, etc.

[0051] Embodiment one:

[0052] Figure 1 shows a rail transit platform shielding door provided by the embodiment one of the present application, referring to figure 1, the rail transit platform shielding door comprises:

[0053] The controller 11 is used for controlling the shielding door body 10 to perform opening and closing operations;

[0054] The monitoring sensor 12 is used for collecting real-time running data of the shielding door based on the opening and closing operations of the shielding door body 10;

[0055] The fault early warning device 13 is used for obtaining the real-time running data, inputting the real-time running data into a pre-constructed prediction model, performing multi-round cycle segmented prediction on the real-time running data based on the prediction model, outputting running prediction results of each time node of a target period, taking the real-time running data as initial model input data of the prediction model in the multi-round cycle segmented prediction process, outputting a segmented prediction data of the target period for each round of prediction, and taking the predicted segmented prediction data and the real-time running data as model input data for the next round of prediction, inputting the running prediction results into a pre-constructed fault monitoring model, outputting fault type monitoring results of the corresponding time node based on the fault monitoring model, and pre-training the fault monitoring model based on fault running data of the shielding door, and pre-annotating corresponding fault type labels for the fault running data.

[0056] The monitoring sensor and the fault early warning device are arranged on the corresponding shield door body to monitor the running state of the shield door in real time and give an early warning when a fault is monitored. The monitoring sensor 12 detects the opening and closing operation of the shield door body, and then obtains the corresponding real-time running data of the shield door. The real-time running data is analyzed by the fault early warning device 13 to achieve the effect of fault early warning.

[0057] Specifically, FIG. 2 shows a flowchart of a shield door fault monitoring method provided by the embodiment of the present application. The shield door fault monitoring method provided in the embodiment can be executed by a shield door fault early warning device.

[0058] The following describes an example in which the shield door fault early warning device is the main body of the shield door fault monitoring method. Referring to FIG. 2, the shield door fault monitoring method specifically includes the following steps.

[0059] S110, obtaining shield door real-time running data and inputting the real-time running data into a pre-constructed prediction model.

[0060] When monitoring the fault of the shield door, the real-time running data of the shield door is collected, and then the real-time running data is used to predict the running state of the shield door in the future period by using the pre-constructed prediction model, so as to accurately and timely monitor the fault according to the predicted running state of the shield door.

[0061] The monitoring sensor 12 is a monitoring camera. The monitoring camera detects and locates the shield door body, and determines the real-time running data of the shield door based on the detected position of the shield door body. The real-time running data can be shield door opening and closing data, door body displacement data, door body speed data, etc. The monitoring camera is installed at a position that can clearly capture the movement track of the shield door body, such as the side or top of the shield door. The focal length, exposure, frame rate and other parameters of the camera are adjusted according to the monitoring requirements to ensure clear and stable images and reduce noise interference. The monitoring camera collects video streams corresponding to the shield door body in real time. The video streams captured by the camera are first subjected to image processing, including denoising, contrast enhancement, edge detection and other steps, to improve the image quality and facilitate subsequent analysis. Then, the processed images are subjected to target detection by using computer vision technology (such as convolutional neural network CNN in deep learning), to identify the position of the shield door body. Before that, a target detection model is trained to enable the model to identify specific objects (i.e. the shield door body) in the image. After detecting the shield door body, the position of the shield door body in the three-dimensional space is calculated based on camera calibration and three-dimensional reconstruction technology, by using the pixel coordinates in the image and the geometric parameters (such as focal length, viewing angle, etc.) of the camera.

[0062] By monitoring the position change of the door body, it can be judged whether the shielding door is in an open state or a closed state. When the door body moves from the closed position to the open position (or vice versa), it can be recorded as one switching operation; by continuously monitoring the position change of the door body, the displacement amount of the door body in a specific time period can be calculated; by comparing the position change of the door body in the two consecutive images, combined with the time interval, the instantaneous speed of the door body can be calculated. The shielding door switching data, door body displacement data, door body speed data and other real-time running data are obtained, and the real-time running data processed by the monitoring camera are transmitted to the fault early warning device in real time.

[0063] Optionally, in the application, the running data of the shielding door can also be collected in real time by the infrared sensor, the displacement sensor, the processor of the shielding door and other nodes. The shielding door switching data can be the motor current data, voltage data, infrared sensor data and the like of the switching door. According to the actual fault monitoring requirements, the real-time running data can also be the door body position, switching door time, current voltage, temperature, vibration and other parameters. The specific type of real-time running data is not limited in this application, and will not be described here.

[0064] The collected real-time running data is input into the pre-trained prediction model, so as to realize the prediction of the future running state of the shielding door. Before that, the historical running data of the shielding door is collected to train the prediction model based on the historical running data.

[0065] Referring to FIG. 3, the training process of the prediction model includes:

[0066] S1001, collecting the historical running data of the shielding door to construct a training sample, and splitting the training sample into a first sequence and a second sequence according to time sequence;

[0067] S1002, inputting the first sequence into the prediction model for multi-round cycle segmentation prediction to obtain a corresponding prediction sequence;

[0068] S1003, calculating a loss function according to the prediction sequence and the second sequence, and iteratively adjusting the model parameters of the prediction model based on the loss function until the prediction model converges.

[0069] The historical operation data is collected according to actual historical operation data of the shielding door, the type of the historical operation data corresponds to the type of the real-time operation data, and the data type is set according to actual requirements. Optionally, the historical operation data includes shielding door opening and closing data, door body displacement data and door body speed data collected at different historical time nodes. These historical operation data are preprocessed in the order of the historical time nodes, such as data cleaning, data formatting (unifying data format, dimension normalization) and data division (dividing the data set into a training set, a validation set and a test set). Further, the historical operation data are divided into a plurality of sample pairs in time sequence, each sample pair includes a plurality of groups of operation data sorted by time nodes (each group of data records shielding door opening and closing data, door body displacement data and door body speed data once), and each sample pair is divided into an input sequence (a first sequence) and a target sequence (a second sequence). The first sequence is used for model prediction, and the second sequence is used for calculating the model prediction error.

[0070] After the training sample construction is completed, the prediction model is initialized, the initial parameters of the model are set, and then the training sample is used for multi-round cycle segmented prediction. For each sample pair of the training sample, the first sequence is input into the prediction model, and then multi-round cycle segmented prediction is performed inside the model. Each round of prediction outputs a prediction result for a time period, and then each round of prediction takes the prediction result obtained by the prediction as the input part of the next round of prediction, and splices the first sequence as the input of the next round of the model. The above process is repeated until the prediction of the entire sequence is completed, and a complete prediction sequence is obtained.

[0071] Further, based on the prediction sequence, a set loss function (such as mean square error, cross entropy, etc.) is used to calculate the difference between the prediction sequence and the second sequence, which reflects the accuracy of the model prediction. According to the value of the loss function, the parameters of the prediction model can be adjusted iteratively using the corresponding optimization algorithm. After completing the prediction of each group of sample pairs, the loss function is calculated and the model parameters are adjusted. By monitoring the change of the loss function, when the loss function value reaches the set value, or reaches the preset training round, time or other conditions, it is determined that the prediction model has converged, and the training process is ended.

[0072] Optionally, during the actual training process, the validation set and the test set can also be used to evaluate the performance of the trained prediction model. The validation set is used to adjust the model hyperparameters during the training process to prevent overfitting; the test set is used to evaluate the generalization ability of the model. By training a prediction model that can accurately predict the future operation state of the shielding door, the future operation state of the shielding door can be accurately predicted, and then the future operation state of the shielding door is used for fault monitoring to timely discover potential fault types.

[0073] In particular, the prediction model of the present application comprises a first multi-layer perception module, a decoder network and a second multi-layer perception module connected in sequence;

[0074] The first multi-layer perception module is configured to convert the model input data of the prediction model into a vector of a set shape;

[0075] The decoder network is configured to generate an initial prediction sequence based on the vector of the set shape,

[0076] The second multi-layer perception module is configured to convert the initial prediction sequence into segmented prediction data according to the shape of the model input data.

[0077] As shown in FIG. 4, by adding a multi-layer perception module to the decoder network, the model can adapt to different input sequence shapes. The model generates the final prediction sequence in a multi-cycle manner. Since the input sequence of each cycle concatenates the sequence predicted in the previous cycle, the length of the sequence in the second dimension is increased. Therefore, in order to solve the problem of inconsistent input sequence shapes, the present application processes the data through the first multi-layer perception module before inputting the data into the model, so that the shape of the data input into the decoder network is consistent.

[0078] The first multi-layer perception module comprises an input layer, an output layer and multiple hidden layers, which convert the first sequence of the input model into a set shape through nonlinear transformation. The second multi-layer perception module converts the output of the decoder network into the shape of the original sequence, which is opposite in structure to the first multi-layer perception module. The calculation formula of the first multi-layer perception module is represented as: out = relu(w x + b)

[0079] wherein w represents the weight of the hidden layer of the first multi-layer perception module, b represents the bias term of the hidden layer, relu is a nonlinear activation function, x out is the output of the first multi-layer perception module, and x is the input of the first multi-layer perception module.

[0080] As shown in FIG. 4, the shape of the input sequence of the first multi-layer perception module is [l, m + (n-1)·k], where l is the length of a single sequence, m is the running data of the initial input sequence, such as the number of switch door actions of the shield door, k is the predicted running data, and n is the number of cycles in one prediction. In a phased prediction manner, the K times of running data to be predicted are divided into n groups. In the first stage of prediction, the initial sequence (at this time, the first sequence) is input into the model to obtain the next k = K / n times of running data. In the second stage of prediction, the initial sequence and the predicted k times of running data are input into the model together to predict the next [k, 2k] actions. The cycle is repeated until the K times of running data are predicted. The shape of the input sequence of the model in each cycle is [l, m], [l, m+k], …, [l, m + (n-1)·k]. By using the above method, the prediction model only needs to cycle n times to complete all the predictions, which can ensure the accuracy of long-term prediction and improve the inference speed.

[0081] Based on the above trained prediction model, in the process of monitoring the failure of the shield door, the real-time running data of the shield door is input into the prediction model, and the running state of the shield door in the future period can be determined through model inference.

[0082] Optionally, before the real-time running data is input into the pre-constructed prediction model, the following steps are further included:

[0083] The real-time running data is subjected to missing value filling and outlier screening processing, and the processed real-time running data is subjected to standardization processing.

[0084] It can be understood that, in the process of collecting the running data, a small part of abnormal data and missing data may be collected due to sensor failure or communication failure, and this part of data may seriously affect the performance of the model and needs to be processed. Based on this, the application adopts a data filling method to process the missing values in the running data, and the filling value can be a historical average value or a set value. For example, the average value of 10 values before and after the missing position is filled into the missing position. For the abnormal value, the entire sequence containing the abnormal value is deleted.

[0085] After the processing is completed, the real-time running data is subjected to standardization processing to meet the input and inference requirements of the model. The standardization processing is used to standardize the real-time running data into sequence data with a mean value of 0 and a standard deviation of 1.

[0086] Standardization refers to standardizing a sequence of data into a standard normal distribution form with a mean value of 0 and a standard deviation of 1. The formula of the standardization processing is as follows:

[0087] wherein x0 is the original value of the running data, μ is the mean value, σ is the standard deviation, and x' is the value after standardization. After the above processing, the real-time running data can be input into the prediction model for model inference.

[0088] It should be noted that, in the prediction model training process, in order to accelerate the convergence of the model in subsequent model training, the historical running data also needs to be subjected to the above missing value filling, abnormal value deletion and standardization processing. The pre-processing process of the running data of the present application is not fixedly limited, and will not be described here.

[0089] S120, based on the prediction model, performing multi-round cyclic segmented prediction on the real-time running data, and outputting the running prediction result of each time node in the target period. In the multi-round cyclic segmented prediction process, the real-time running data is used as the initial model input data of the prediction model, one segmented prediction data of the target period is output in each round of prediction, and the predicted segmented prediction data and the real-time running data are used as the model input data for the next round of prediction.

[0090] Further, in the inference process of the prediction model, the real-time running data input is subjected to multi-round cyclic segmented prediction with reference to the segmented prediction process in the model training. A future period to be predicted is defined as a target period. The real-time running data is used as the initial input sequence, the running data of the first segment of the target period is predicted in the first stage, which is defined as the first segment prediction data; the first segment prediction data and the real-time running data are spliced as the input sequence for the second stage prediction, the second segment prediction data of the target period is predicted in the second stage; the first segment prediction data, the second segment prediction data and the real-time running data are spliced as the input sequence for the third stage prediction, and the third stage prediction inference is performed. Similarly, until the prediction of all segmented prediction data of the target period is completed, all segmented prediction data are spliced in time sequence to obtain the running prediction result of each time node in the target period.

[0091] Exemplarily, assuming that the running state of the shield door in the next 24 hours (target period) is to be predicted, the prediction is made by inputting real-time running data (such as real-time running data in the current one hour). By dividing the target period into multiple equal-length time periods, each time period has a fixed length (such as one hour). In the multi-round segmented prediction process of the prediction model, the first round of prediction uses the initial real-time running data as input, and the running data of the first hour in the future is predicted to obtain the prediction result of the time period. Save this prediction result as part of the input in the next round of prediction. In the second round of prediction, the prediction result of the first time period obtained by the first round of prediction is used as input together with the real-time running data to make a prediction to obtain the prediction result of the second hour. Repeat the above process, each time use the predicted prediction result and real-time running data as input to make the next round of prediction, until the running data of the 24th hour is predicted, and all the predicted running data in the next 24 hours is output, thereby completing the prediction of the running state of the shield door.

[0092] In S130, the running prediction result is input into a pre-constructed fault monitoring model, and a fault type monitoring result of a corresponding time node is output based on the fault monitoring model. The fault monitoring model is pre-trained based on fault running data of the shield door, and the fault running data is pre-labeled with corresponding fault type labels.

[0093] Finally, the running prediction result is input into the pre-constructed fault monitoring model, and the potential fault type occurring at the corresponding time node in the future period can be diagnosed. The fault monitoring model can be built using a CNN convolutional neural network. CNN can effectively extract local features in the data and accelerate model training through local receptive fields and parameter sharing mechanisms. By inputting the running prediction result into the fault monitoring model, the model processes through convolution layers, pooling layers, and fully connected layers, and finally outputs the fault monitoring result of the shield door at each time node. If a fault is monitored, the fault monitoring result contains the corresponding fault type.

[0094] Before that, the device is adjusted to normal and various fault states by manual adjustment of the device, and the running data during that period is collected to obtain the running data of the shield door in each fault type state. Then, the collected fault running data is labeled, that is, each fault running data is configured with a corresponding fault type label. Then, the labeled fault running data is used as a training set to train the fault monitoring model. During the training process, the model learns to extract features from the fault running data and establish a mapping relationship between the features and the fault types.

[0095] After the model training is completed, based on the operation prediction results of the target period of the input fault monitoring model at different time nodes, the fault monitoring model performs fault diagnosis on the operation prediction results of each time node in chronological order, determines whether there is an operation fault at the time node, and if so, outputs the corresponding fault type. By analyzing the results output by the fault monitoring model, it is determined whether there is a potential fault risk. If the model predicts that there is a risk of a corresponding fault type at a certain time node, further attention is needed. For example, according to the analysis result, corresponding fault response measures are formulated. Including arranging maintenance plan in advance, increasing monitoring frequency, adjusting operation parameters, etc. If the predicted fault risk is very high, emergency measures can be taken immediately to prevent the fault from occurring. In this way, by discovering potential operation faults of the shield door in the future period in time, corresponding measures can be taken in time to ensure the reliability and stability of the shield door operation.

[0096] Referring to FIG. 5, by using no-label data to build a prediction model data set, a prediction model is built for model training to output operation prediction results based on input real-time operation data. On the other hand, a fault monitoring model data set is built by using labeled data. In this process, by simulating the operation state under various fault states of the shield door, accurate and highly available operation data can be collected to build the fault monitoring model data set, and only a small amount of data is needed for model training. Based on the trained fault monitoring model, by inputting the operation prediction results output by the above prediction model, the fault monitoring model diagnoses whether the shield door will fail in the future period and the possible fault category. According to the prediction and diagnosis results, a predictive maintenance report is formed, which can assist workers to complete the shield door inspection and maintenance. By predicting in advance, the number of shield door downtime maintenance is reduced. And by predicting the fault category, the difficulty of fault screening can be reduced, effectively improving the work efficiency.

[0097] The above, by setting the monitoring sensor of the shielding door and the fault early warning device, real-time running data of the shielding door is acquired, and the real-time running data is input into a pre-constructed prediction model; the real-time running data is subjected to multi-round cyclic segmented prediction based on the prediction model, and running prediction results of each time node in a target period are output; in the multi-round cyclic segmented prediction process, the real-time running data is used as initial model input data of the prediction model, one segmented prediction data of the target period is output for each round of prediction, and the segmented prediction data that has been predicted and the real-time running data are used as model input data for the next round of prediction; the running prediction results are input into a pre-constructed fault monitoring model, and fault type monitoring results of the corresponding time node are output based on the fault monitoring model; the fault monitoring model is pre-trained based on fault running data of the shielding door, and the fault running data is pre-labeled with corresponding fault type labels. By using the above technical means, the running prediction results of the shielding door at each time node in the future target period are subjected to cyclic segmented prediction, and then the fault monitoring is performed based on the running prediction results, so that the potential fault type at the corresponding time node is discovered in a timely manner, accurate and timely fault monitoring is realized, and the reliability and stability of the shielding door operation are improved.

[0098] Embodiment Two

[0099] On the basis of the above embodiment, FIG. 6 is a structural schematic diagram of a shielding door fault monitoring device provided in Embodiment Two of the present application. Referring to FIG. 6, the shielding door fault monitoring device provided in the present embodiment specifically comprises: an input module 21, a data prediction module 22, and a fault monitoring module 23.

[0100] The input module 21 is configured to acquire real-time running data of the shielding door and input the real-time running data into a pre-constructed prediction model;

[0101] The data prediction module 22 is configured to perform multi-round cyclic segmented prediction on the real-time running data based on the prediction model, output running prediction results of each time node in a target period, and in the multi-round cyclic segmented prediction process, use the real-time running data as initial model input data of the prediction model, output one segmented prediction data of the target period for each round of prediction, and use the segmented prediction data that has been predicted and the real-time running data as model input data for the next round of prediction.

[0102] The fault monitoring module 23 is configured to input the running prediction results into a pre-constructed fault monitoring model, output fault type monitoring results of the corresponding time node based on the fault monitoring model, and pre-train the fault monitoring model based on fault running data of the shielding door, and pre-label the fault running data with corresponding fault type labels.

[0103] Specifically, the training process of the prediction model comprises:

[0104] The historical operation data of the shielding door is collected to construct a training sample, and the training sample is split into a first sequence and a second sequence in time sequence;

[0105] The first sequence is input into a prediction model for multi-round cycle segmentation prediction to obtain a corresponding prediction sequence;

[0106] A loss function is calculated according to the prediction sequence and the second sequence, and model parameters of the prediction model are iteratively adjusted based on the loss function until the prediction model converges.

[0107] Specifically, the historical operation data includes shielding door opening and closing data, door body displacement data and door body speed data collected at different historical time nodes.

[0108] Specifically, the prediction model includes a first multi-layer perception module, a decoder network and a second multi-layer perception module connected in sequence;

[0109] The first multi-layer perception module is used to convert model input data of the prediction model into a vector of a set shape;

[0110] The decoder network is used to generate an initial prediction sequence based on the vector of the set shape;

[0111] The second multi-layer perception module is used to convert the initial prediction sequence into segmentation prediction data according to the shape of the model input data.

[0112] Specifically, the calculation formula of the first multi-layer perception module is represented as: x out = relu(w·x+b)

[0113] wherein w represents the weight of the hidden layer of the first multi-layer perception module, b represents the bias term of the hidden layer, relu is a nonlinear activation function, x out is the output of the first multi-layer perception module, and x is the input of the first multi-layer perception module.

[0114] Specifically, before inputting the real-time operation data into the pre-constructed prediction model, it further includes:

[0115] The real-time operation data is subjected to missing value filling and outlier screening processing, and the processed real-time operation data is subjected to standardization processing.

[0116] Specifically, the standardization processing is used to standardize the real-time operation data into sequence data with a mean of 0 and a standard deviation of 1.

[0117] The monitoring sensor of the shielding door and the fault early warning device are arranged, real-time running data of the shielding door is acquired, and the real-time running data is input into a pre-constructed prediction model; the real-time running data is subjected to multi-round cyclic segmented prediction based on the prediction model, and running prediction results of each time node in a target period are output; in the multi-round cyclic segmented prediction process, the real-time running data is used as initial model input data of the prediction model, one segmented prediction data of the target period is output in each round of prediction, and the segmented prediction data that has been predicted and the real-time running data are used as model input data for the next round of prediction; the running prediction results are input into a pre-constructed fault monitoring model, and fault type monitoring results of the corresponding time node are output based on the fault monitoring model; the fault monitoring model is pre-trained based on fault running data of the shielding door, and the fault running data is pre-labeled with corresponding fault type labels. By using the above technical means, the running prediction results of the shielding door at each time node in the future target period are subjected to cyclic segmented prediction, and then the fault monitoring is performed based on the running prediction results, so that the potential fault type at the corresponding time node is found in a timely manner, accurate and timely fault monitoring is realized, and the reliability and stability of the shielding door running are improved.

[0118] The fault monitoring device of the shielding door provided in the second embodiment of the present application can be used to execute the fault monitoring method of the shielding door provided in the first embodiment, and has corresponding functions and advantages.

[0119] Embodiment three:

[0120] The electronic device provided in the third embodiment of the present application includes a processor 31, a memory 32, a communication module 33, an input device 34, and an output device 35. The number of processors in the electronic device can be one or more, and the number of memories in the electronic device can be one or more. The processor, the memory, the communication module, the input device, and the output device of the electronic device can be connected through a bus or other means.

[0121] The memory, as a computer readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules of the fault monitoring method of the shielding door (for example, the input module, the data prediction module and the fault monitoring module in the fault monitoring device of the shielding door) according to any embodiment of the present application. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; the data storage area can store data created according to the use of the device and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device or other non-volatile solid-state memory device. In some examples, the memory can further include a memory remotely arranged with respect to the processor, and these remote memories can be connected to the device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0122] The communication module is used for data transmission.

[0123] The processor executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory, that is, implements the above-mentioned fault monitoring method of the shielding door.

[0124] The input device can be used to receive input digital or character information, and generate key signal input related to user settings and function control of the device. The output device can include a display device such as a display screen.

[0125] The electronic device provided above can be used to execute the fault monitoring method of the shielding door provided in the first embodiment, and has corresponding functions and advantages.

[0126] Embodiment four:

[0127] The embodiment of the present application also provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to execute a shield door fault monitoring method, the shield door fault monitoring method comprising: acquiring shield door real-time running data, and inputting the real-time running data into a pre-constructed prediction model; performing multi-round cyclic segmented prediction on the real-time running data based on the prediction model, and outputting a running prediction result of each time node of a target period, wherein in the multi-round cyclic segmented prediction process, the real-time running data is used as initial model input data of the prediction model, one segmented prediction data of the target period is output in each round of prediction, and the predicted segmented prediction data and the real-time running data are used as model input data of the next round of prediction; inputting the running prediction result into a pre-constructed fault monitoring model, and outputting a fault type monitoring result of the corresponding time node based on the fault monitoring model, wherein the fault monitoring model is pre-trained based on shield door fault running data, and the fault running data is pre-labeled with a corresponding fault type label.

[0128] Storage medium - any of various types of memory devices or storage devices. The term "storage medium" is intended to include an installation medium, e.g., a CD-ROM, floppy disks, or tape device; a computer system memory or random access memory such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; or a non-volatile memory such as a magnetic medium (e.g., a hard disk or floppy disk), optical medium (e.g., a CD-ROM), flash memory, or other similar type of memory; a register or other similar type of memory element. The storage medium can also include a non-tangible medium in which data is stored and retrieved, e.g., a

[0129] Of course, the storage medium provided by the embodiment of the present application comprises computer executable instructions, which are not limited to the shield door fault monitoring method as described above, and can also execute the related operations in the shield door fault monitoring method provided by any embodiment of the present application.

[0130] The shield door fault monitoring device, the storage medium and the electronic equipment provided in the above embodiments can execute the shield door fault monitoring method provided by any embodiment of the present application, and the technical details not described in detail in the above embodiments can be referred to the shield door fault monitoring method provided by any embodiment of the present application.

[0131] The above merely provides preferred embodiments of the present application and the applied technical principles. The present application is not limited to the specific embodiments described herein, and various obvious changes, modifications and replacements made by those skilled in the art without departing from the scope of the present application shall not be excluded. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and more other equivalent embodiments can be included without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A rail transit platform screen door, wherein, The application relates to a shield door fault early warning method and device. The controller is used for controlling a shield door body to perform opening and closing operations. The monitoring sensor is used for collecting real-time operation data of the shield door based on the opening and closing operations of the shield door body. The fault early warning device is used for acquiring the real-time operation data, inputting the real-time operation data into a pre-constructed prediction model, performing multi-round cyclic segmented prediction on the real-time operation data based on the prediction model, outputting operation prediction results of each time node in a target period, taking the real-time operation data as initial model input data of the prediction model in the multi-round cyclic segmented prediction process, outputting segmented prediction data of the target period in each round of prediction, and taking the predicted segmented prediction data and the real-time operation data as model input data of the next round of prediction. The training process of the prediction model comprises the following steps: Collecting historical operation data of the shield door to construct training samples, and splitting the training samples into a first sequence and a second sequence according to time sequences. The first sequence is input into the prediction model to perform multi-round cyclic segmented prediction, and a corresponding prediction sequence is obtained. According to the prediction sequence and the second sequence, a loss function is calculated, and model parameters of the prediction model are iteratively adjusted based on the loss function until the prediction model converges.

2. The rail transit platform screen door according to claim 1, wherein, The historical operation data comprises shield door opening and closing data, door body displacement data and door body speed data collected at different historical time nodes.

3. The rail transit platform screen door of claim 1, wherein, The prediction model comprises a first multi-layer perception module, a decoder network and a second multi-layer perception module connected in sequence. The first multi-layer perception module is used for converting model input data of the prediction model into a vector with a set shape. The decoder network is used for generating an initial prediction sequence based on the vector with the set shape. The second multi-layer perception module is used for converting the initial prediction sequence into the segmented prediction data according to the shape of the model input data.

4. The rail transit platform screen door of claim 3, wherein, The calculation formula of the first multi-layer perception module is represented as: x out = relu(w·x+b) wherein w represents weights of hidden layers of the first multi-layer perception module, b represents bias terms of the hidden layers, relu is a nonlinear activation function, x out is an output of the first multi-layer perception module, and x is an input of the first multi-layer perception module.

5. The rail transit platform screen door of claim 1, wherein, Before the real-time operation data is input into the pre-constructed prediction model, the fault early warning device further comprises the following steps: The real-time operation data is subjected to missing value filling and abnormal value screening processing, and the processed real-time operation data is subjected to standardization processing.

6. The rail transit platform screen door of claim 5, wherein, The standardization processing is used for standardizing the real-time operation data into sequence data with a mean value of 0 and a standard deviation of 1.

7. The rail transit platform screen door of claim 1, wherein, The monitoring sensor is a monitoring camera, which determines the real-time operation data of the shield door by performing target detection and positioning on the shield door body based on the detected shield door body position.

8. A method for monitoring the failure of a shield door applied to the platform screen door of any one of claims 1-7, wherein, The application relates to a shield door fault early warning method and device. The application relates to a shield door fault early warning method and device. The application relates to a shield door fault early warning method and device. based on the prediction model, the real-time operation data is subjected to multi-round cyclic segmented prediction, and operation prediction results of each time node of a target period are output; in the multi-round cyclic segmented prediction process, the real-time operation data is taken as initial model input data of the prediction model, one segmented prediction data of the target period is output for each round of prediction, and the segmented prediction data that has been predicted and the real-time operation data are taken as model input data for the next round of prediction; the operation prediction results are input into a pre-constructed fault monitoring model, and fault type monitoring results of a corresponding time node are output based on the fault monitoring model; the fault monitoring model is pre-trained based on fault operation data of the shielding door, and the fault operation data is pre-labeled with a corresponding fault type label; the training process of the prediction model comprises: historical operation data of the shielding door is collected to construct training samples, and the training samples are split into a first sequence and a second sequence in chronological order; the first sequence is input into the prediction model to perform multi-round cyclic segmented prediction, and a corresponding prediction sequence is obtained; a loss function is calculated according to the prediction sequence and the second sequence, and model parameters of the prediction model are iteratively adjusted based on the loss function until the prediction model converges.

9. A malfunction monitoring apparatus of a shielding door, wherein, comprises: an input module configured to acquire real-time operation data of the shielding door and input the real-time operation data into a pre-constructed prediction model; a data prediction module configured to perform multi-round cyclic segmented prediction on the real-time operation data based on the prediction model, and output operation prediction results of each time node of a target period; in the multi-round cyclic segmented prediction process, the real-time operation data is taken as initial model input data of the prediction model, one segmented prediction data of the target period is output for each round of prediction, and the segmented prediction data that has been predicted and the real-time operation data are taken as model input data for the next round of prediction; a fault monitoring module configured to input the operation prediction results into a pre-constructed fault monitoring model, and output fault type monitoring results of a corresponding time node based on the fault monitoring model; the fault monitoring model is pre-trained based on fault operation data of the shielding door, and the fault operation data is pre-labeled with a corresponding fault type label; the training process of the prediction model comprises: historical operation data of the shielding door is collected to construct training samples, and the training samples are split into a first sequence and a second sequence in chronological order; the first sequence is input into the prediction model to perform multi-round cyclic segmented prediction, and a corresponding prediction sequence is obtained; a loss function is calculated according to the prediction sequence and the second sequence, and model parameters of the prediction model are iteratively adjusted based on the loss function until the prediction model converges.

10. A storage medium containing computer-executable instructions, wherein, The computer executable instructions, when executed by the computer processor, are used to perform the fault monitoring method of the shielding door according to claim 8. The computer executable instructions, when executed by the computer processor, are used to perform the fault monitoring method of the shielding door according to claim 8.

Images