Urban rail transit one-key switching station self-checking optimization method

Abstract

The application provides a kind of urban rail transit one-key switch station self-check optimization method, comprising the following steps: S1, construct time series dataset;S2, time series dataset preprocessing operation;The data in time series dataset is made into supervised learning dataset by feature engineering;S3, build causal convolutional neural network model, load training set data in causal convolutional neural network model, model training is carried out to causal convolutional neural network model, so that causal convolutional neural network model captures long time series time series features;S4, input validation set to causal convolutional neural network model verification, input the statistical value of the day before a week made by feature engineering to the neural network model trained, to predict the self-check operation record of the day.The application can provide one-time alarm information and its recommended operation mode prompt for the daily inspection plan of the whole station equipment, save operation time, improve the scientificity and efficiency of one-key switch station function.

Description

Technical Field

[0001] This invention belongs to the field of one-button switch station self-testing technology, and in particular relates to an optimized method for one-button switch station self-testing in urban rail transit. Background Technology

[0002] While the rapid development of subway construction in my country has brought convenience to people's lives, it has also brought huge challenges to the operation and management of subway stations. In order to alleviate the pressure on subway operation and management and ensure the level of operation and service, major cities across the country have successively launched the construction of smart subways.

[0003] Currently, research on smart stations has shifted from the functional verification stage to the value output stage. One-button station opening / closing is one of the most representative functions in smart stations. The commonly used one-button station opening / closing function is based on traditional integrated monitoring systems, automatically linking the systems and equipment involved in the station opening and closing process. This transforms traditional equipment monitoring into scene linkage, enabling one-button opening and closing through remote operation. The remote operation process includes issuing and confirming self-check commands for each device or system. Each step of these operations displays prompts, and most require manual confirmation, significantly increasing operation time and staff workload, making one-button station opening / closing relatively cumbersome. Therefore, this patent application designs an optimized self-check method for one-button station opening / closing in urban rail transit. Summary of the Invention

[0004] In view of this, the present invention aims to propose an optimized method for one-button station opening and closing self-inspection in urban rail transit, in order to solve the problem that the current one-button station opening and closing operation is relatively cumbersome, and that subway operators repeatedly click on various self-inspection equipment alarm prompts and confirmation prompts, which greatly increases the operation time and the workload of staff.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] A self-checking and optimization method for one-button station opening and closing in urban rail transit, the method comprising the following steps:

[0007] S1. Construct a time-series dataset; Encode historical self-check operation records, convert self-check operation categories from peer-level category information into numerical representations to form a time-series dataset;

[0008] In step S1, the self-test operation categories include no alarm information displayed, displaying alarm information and selecting confirmation, and displaying alarm information and selecting ignore;

[0009] The encoding transforms the original feature variables in the historical self-check operation records into multi-dimensional variables for original feature classification, and uses 0 / 1 on each dimension to quantify yes / no, present / absent;

[0010] Convert any undisplayed alarm information in the self-test operation category to [0,0];

[0011] Change the display alarm information and confirmation option in the self-test operation category to [1,0];

[0012] Change the displayed alarm information and select ignore in the self-test operation category to [1,1];

[0013] S2. Preprocessing operations on the time series dataset; Perform feature engineering on the data in the time series dataset to convert the time series dataset into a supervised learning dataset, and divide the supervised learning dataset into a training set, a test set, and a validation set;

[0014] S3. Build a causal convolutional neural network model, load training data containing time-series features into the causal convolutional neural network model, train the causal convolutional neural network model, and enable the causal convolutional neural network model to capture long-term time-series features.

[0015] S4. Input the data from the validation set into the trained causal convolutional neural network model to verify the model performance. Input the statistical values ​​of the previous week produced by feature engineering into the trained neural network model to predict the self-check operation records of the day.

[0016] The causal convolutional neural network model includes a residual block, a self-attention mechanism module, and a fully connected layer connected in sequence.

[0017] The number of residual blocks is multiple, and each residual block includes a causal convolutional layer, a normalization layer, an activation function, and a Dropout layer. Skip connections are also introduced into the residual blocks. The formula for each residual block is as follows:

[0018] ;

[0019] in, This is the output of the previous residual block. This is the computation layer function for the current residual block. This is the result of the current residual block.

[0020] Furthermore, in step S1, the historical self-check records represented by numerical values ​​are arranged in chronological order to form a time-series dataset, which is represented as follows:

[0021] ;

[0022] In the formula, It is a time-series dataset, consisting of historical self-inspection records arranged in chronological order and represented by numerical values.

[0023] Furthermore, in step S2, data statistics are performed on all the time-series datasets in step S1. The feature engineering includes the number of times ignored in the week before the current day, the number of times confirmed in the week before the current day, the number of times no warning was issued in the week before the current day, and the statistics of the most frequent self-test operation category in each device in the week of the current day.

[0024] , , and Input characteristics that constitute a day ,by A supervised learning dataset is constructed from historical self-inspection records arranged in chronological order and represented by numerical values. The supervised learning dataset is represented as follows:

[0025] ;

[0026] In the formula, For supervised learning datasets, These are the labeled samples.

[0027] Furthermore, the residual block construction method in step S3 is as follows: the residual block includes two residual modules with identical structures, which are connected sequentially. Each residual module includes:

[0028] (1) The first causal convolutional layer is divided into two layers with dilation coefficients of 2 and 4 respectively, and the kernel size of each layer is 3;

[0029] (2) The second normalization layer connects the input of the normalization layer to the output of the causal convolutional layer. The formula for this layer is:

[0030] , ;

[0031] in, and For the mean and variance of a batch of data, For minterms greater than 0, and These are the parameters for network learning.

[0032] (3) The third ReLU activation function layer, whose input is connected to the output of the normalization layer, has the following activation function formula:

[0033] ;

[0034] in, The output value of the planning layer. It is the ReLU activation function;

[0035] (4) The fourth layer is the Dropout layer, which discards half of the neuron connections. The input of the Dropout layer is connected to the output of the ReLU activation function layer, and the output of the Dropout layer is connected to the causal convolution layer of the next residual module.

[0036] Furthermore, the skip connection is to add a channel between the start and end of the residual block, which is then added to the result of the main channel via a 1*1 convolutional layer.

[0037] Furthermore, the self-attention mechanism includes three channels, wherein the first channel is formulated as follows:

[0038] ;

[0039] In the formula, The result after the last residual block and These are the max pooling and average pooling layers, respectively. A 1x1 convolutional layer, For the sigmoid activation function, "This indicates that a splicing operation is performed on the channel;

[0040] The second channel is passed through a 1x1 convolutional layer and a normalization layer, multiplied by the result of the first channel, and then passed through a Softmax layer. The specific formula is as follows:

[0041] ;

[0042] ;

[0043] in, This is the output of the first channel. This is a function for processing 1x1 size convolutional and normalization layers. This is the result of multiplying two channels. for Output after the SoftMax layer;

[0044] The third channel is passed through a 1x1 convolutional layer and multiplied with the results of the other two channels.

[0045] Furthermore, the self-test operation record includes all alarm information operation types of each self-testing device.

[0046] Compared with existing technologies, the self-inspection and optimization method for one-button station opening and closing in urban rail transit described in this invention has the following beneficial effects:

[0047] The present invention provides an optimization method for one-click station opening and closing self-inspection in urban rail transit. By encoding and sliding window feature engineering operations on historical self-inspection operation information of stations, the data is adapted to supervised learning and more features are obtained. The method realizes the preprocessing and feature engineering operations on the collected historical self-inspection operation records of one-click station opening and closing, and completes the prediction of the self-inspection information of one-click station opening and closing in the metro.

[0048] Meanwhile, residual block connections and self-attention mechanism modules are added to the traditional causal convolutional neural network to further improve prediction accuracy;

[0049] (2) The self-inspection optimization method for one-button station opening and closing in urban rail transit described in this invention further optimizes the self-inspection process of one-button station opening and closing in subway, reduces the workload of operators, shortens the one-button station opening and closing time, improves the efficiency of operation and management, and shortens the operation time by about 90 seconds compared with the conventional self-inspection method. Attached Figure Description

[0050] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0051] Figure 1 This is a flowchart of a self-inspection and optimization method for one-button switching stations in urban rail transit, as described in an embodiment of the present invention.

[0052] Figure 2 This is a schematic diagram of the causal convolutional neural network model described in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the residual block according to an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram of the main attention mechanism module according to an embodiment of the present invention. Detailed Implementation

[0055] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0056] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0057] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0058] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0059] It should be noted that the one-button switch station self-test operation records used in this embodiment are from 10 stations along Tianjin Metro Line 6. The self-test equipment and systems include roller shutters, escalators, elevators, wastewater pumps, drainage pumps, PSD systems, AFC systems, PIS systems, and PA systems. The alarm information types for roller shutter doors include remote / local status of entrance / exit roller shutter door opening / closing signals, entrance / exit roller shutter door opening / closing signals, and feedback on the descent of fire-resistant roller shutter doors. The alarm types for escalators include thermal overload relay protection alarms, handrail belt entrance fault alarms, step sagging alarms, step anti-jump alarms, skirt board fault alarms, handrail belt breakage alarms, comb plate fault alarms, emergency stop button triggered alarms, intermediate emergency stop button triggered alarms, maintenance cover plate fault alarms, and water level switch triggered alarms. Vertical elevators include safety circuit alarms. Wastewater pumps and drainage pumps include over-high and under-water level alarms. PSD systems include gap detection isolation mode alarms, door opening timeout alarms, and obstacle detection alarms. AFC systems include ticket machine self-test anomaly alarms. PIS and PA systems include communication anomaly alarms. This example also uses a convolutional neural network model built on the TensorFlow framework, with the final model obtained after 10 epochs of training. The training loss function is the cross-entropy function, and the average accuracy and mean squared error of the test set are used as the network evaluation criteria.

[0060] Please see Figure 1 As shown in the figure, this embodiment provides a self-checking and optimization method for one-button station opening and closing in urban rail transit. The method includes the following steps:

[0061] S1. Construct a time-series dataset; Encode historical self-check operation records, convert self-check operation categories from peer-level category information into numerical representations to form a time-series dataset;

[0062] The self-test operation categories include no alarm information displayed, alarm information displayed and confirmation selected, and alarm information displayed and ignore selected.

[0063] The encoding transforms the original feature variables in the historical self-check operation records into multi-dimensional variables for original feature classification, and uses 0 / 1 on each dimension to quantify yes / no, present / absent;

[0064] Convert any undisplayed alarm information in the self-test operation category to [0,0];

[0065] Change the display alarm information and confirmation option in the self-test operation category to [1,0];

[0066] Change the display alarm information and select ignore in the self-test operation category to [1,1].

[0067] In step S1, the time-series dataset consists of historical self-check operation records, arranged in chronological order by day, forming a time-series dataset. The time-series dataset is represented as follows: ;

[0068] Where S is the time series dataset, The historical self-inspection records are represented numerically and arranged in chronological order.

[0069] S2. Preprocessing operations on the time series dataset; Perform feature engineering on the data in the time series dataset to convert the time series dataset into a supervised learning dataset, which is divided into a training set, a test set, and a validation set;

[0070] Data statistics are performed on all time-series datasets in step S1, and the feature engineering includes ignoring the number of times within the previous week. Number of confirmations within the week prior to the date Number of times no warning was given in the week prior to the date Statistics on the most frequent self-test operations among all equipment categories within the week of the current day. ;

[0071] , , and Input characteristics that constitute a day ,by A supervised learning dataset is constructed using historical self-inspection records arranged chronologically in numerical representation.

[0072] The supervised learning dataset can be represented as follows: ,in For supervised learning datasets, The labeled samples;

[0073] The supervised learning dataset consists of a training set, a validation set, and a test set. The training set is used to train the network model; the validation set is used to verify the reliability of the model and to adjust the network parameters; and the test set is used to test the trained model.

[0074] S3. Build a causal convolutional neural network model, load training data containing time-series features into the causal convolutional neural network model, train the causal convolutional neural network model, and enable the causal convolutional neural network model to capture long-term time-series features.

[0075] like Figure 2 As shown, the causal convolutional neural network model includes a residual block, a self-attention mechanism module, and a fully connected layer connected in sequence.

[0076] This embodiment uses three residual blocks as an example. Each residual block includes a causal convolutional layer, a normalization layer, an activation function, and a Dropout layer. Additionally, skip connections are introduced into the residual blocks. The formula for each residual block is as follows:

[0077]

[0078] in, This is the output of the previous residual block. This is the computation layer function for the current residual block. This is the result of the current residual block.

[0079] The residual block construction method in step 3 is as follows: the residual block includes two residual modules with identical structures, which are connected in sequence. Each residual module includes:

[0080] (1) The first causal convolutional layer is divided into two layers with dilation coefficients of 2 and 4 respectively, and the kernel size of each layer is 3;

[0081] (2) The second normalization layer connects the input of the normalization layer to the output of the causal convolutional layer. The formula for this layer is:

[0082] , ;

[0083] in, and For the mean and variance of a batch of data, For minterms greater than 0, and These are the parameters for network learning.

[0084] (3) The third ReLU activation function layer, whose input is connected to the output of the normalization layer, has the following activation function formula:

[0085] ;

[0086] in, This is the output value of the previous layer. It is the ReLU activation function;

[0087] (4) The fourth layer is the Dropout layer, which discards half of the neuron connections. The input of the Dropout layer is connected to the output of the ReLU activation function layer, and the output of the Dropout layer is connected to the causal convolutional layer of the next residual module. The Dropout layer is used to reduce the model size, which is beneficial to accelerate training.

[0088] The skip connection is to add a channel between the start and end of the residual block, which is then added to the result of the main channel through a 1*1 convolutional layer.

[0089] The self-attention mechanism includes three channels, the first of which is formulated as follows:

[0090]

[0091] In the formula, RB represents the result after the last residual block. and These are the max pooling and average pooling layers, respectively. A 1x1 convolutional layer, For the sigmoid activation function, ";" indicates a concatenation operation on the channels;

[0092] The second channel passes through a 1x1 convolutional layer and a normalization layer, is multiplied by the result of the first channel, and then passed through a Softmax layer. This can be represented by the following formula:

[0093] ;

[0094] ;

[0095] in, This is the output of the first channel. This is a function for processing 1x1 size convolutional and normalization layers. This is the result of multiplying two channels. for Output after the SoftMax layer;

[0096] The third channel is passed through a 1x1 convolutional layer and multiplied with the results of the other two channels.

[0097] S4. Input the data from the validation set into the trained causal convolutional neural network model to verify the model performance. Input the statistical values ​​of the previous week, which were generated by feature engineering, into the trained neural network model to predict the self-check operation records for the day.

[0098] The self-test operation record contains all alarm information operation types of the self-test equipment and is provided to relevant subway personnel in one go.

[0099] Those skilled in the art will understand that the methods of the above embodiments can be implemented entirely or in parts by a software program instructing the relevant system, and the program can be stored in a computer memory.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for self-checking optimization of a one-key switch station of urban rail transit, characterized in that, The method comprises the following steps: S1, constructing a time series dataset; encoding the historical self-checking operation record, converting the self-checking operation category from a flat category information to a numerical representation, and constituting a time series dataset; In step S1, the self-checking operation category includes no display of alarm information, display of alarm information and selection of confirmation, and display of alarm information and selection of ignoring. The encoding converts the original feature variables in the historical self-checking operation record into multi-dimensional variables of the original feature category, and uses 0 / 1 to quantify yes / no and have / no on each dimension. The non-display of alarm information in the self-checking operation category is converted into [0, 0]; The display of alarm information and the selection of confirmation in the self-checking operation category are converted into [1, 0]; The display of alarm information and the selection of ignoring in the self-checking operation category are converted into [1, 1]; S2, preprocessing the time series dataset; performing feature engineering on the data in the time series dataset to convert the time series dataset into a supervised learning dataset, and dividing the supervised learning dataset into a training set, a test set and a validation set; S3, building a causal convolutional neural network model, loading the training set data containing time series features into the causal convolutional neural network model, and training the causal convolutional neural network model to capture long time series features; S4, inputting the data of the validation set into the trained causal convolutional neural network model for model performance verification, inputting the statistical values of the previous week into the trained neural network model, and predicting the self-checking operation record of the day; The causal convolutional neural network model comprises residual blocks, a self-attention mechanism module and a full connection layer connected in sequence; The number of residual blocks is multiple, each residual block comprises a causal convolutional layer, a normalization layer, an activation function and a Dropout layer, and a skip connection is introduced in the residual block, and the residual block formula is as follows: z i = z i-1 + f(z i-1 ); where z i-1 is the output of the previous residual block, f(z i-1 ) is the operation layer function of the current residual block, and z i is the result of the current residual block.

2. The self-checking optimization method for a one-key switching station of urban rail transit according to claim 1, characterized in that: In step S1, the numerical representation of the historical self-checking record is arranged in time sequence to constitute a time series dataset, and the time series dataset is represented as: S = {Y1, Y2, Y3,..., Y n}; In the formula, S is a time series data set, Y1, Y2, Y3,..., Y n is a numerical representation of a time-ordered history of self-test records.

3. The self-checking optimization method for a one-key switching station of urban rail transit according to claim 2, characterized in that: In step S2, data statistics are performed on the data in all the time-series data sets in step S1, and the feature engineering includes the number X of times ignored in the previous week of the day (1) , the number X of times confirmed in the previous week of the day (2) , the number X of times not warned in the previous week of the day (3) , and the number X of times of the most frequent category of the number of times of self-checking operations in each device in the week of the day (4) ; X (1) , X (2) , X (3) and X (4) constitute input features X i of a day, X i and the historical self-check records arranged in chronological order represented by numerical values to build a supervised learning data set, which is represented as: D = {(X1, Y1), (X2, Y2),..., (Xn, Yn)}; where n is the number of points in the set D. n n )};​ In the formula, D is a supervised learning data set, (X n ,Y n ) is a labeled sample.

4. The self-checking optimization method for a one-key switching station of urban rail transit according to claim 1, characterized in that, The residual block in step S3 is built by: the residual block comprises two residual modules with the same structure, and the two residual modules are connected in sequence, and each residual module comprises: (1) a first layer of causal convolutional layer, the causal convolutional layer is divided into two layers, the expansion coefficients are 2 and 4 respectively, and the convolution kernel size is 3; (2) a second layer of normalization layer, the input of the normalization layer is connected with the output of the causal convolutional layer, and the formula of the layer is: where μ B and are the mean and variance of a batch of data, ò is a small positive number, and γ and β are network learning parameters. (3) a third layer of ReLU activation function layer, the input of the ReLU activation function layer is connected with the output of the normalization layer, and the activation function formula of the layer is: g(y i ) = max(0, y i ); where y i is the output value of the planning layer, g(y i ) is a ReLU activation function; (4) a fourth layer of Dropout layer, half of the neurons are discarded, the input of the Dropout layer is connected with the output of the ReLU activation function layer, and the output of the Dropout layer is connected with the causal convolutional layer of the next residual module.

5. The method of claim 1, wherein the method further comprises: The skip connection is to add a channel between the start and end of the residual block, and the result is added through a 1*1 convolutional layer.

6. The self-checking optimization method for a one-key switching station of urban rail transit according to claim 1, characterized in that: The self-attention mechanism comprises three channels, wherein the formula of the first channel is as follows: F A (RB) = σ(f 1*1 [MaxPooling(RB); AvgPooling(RB)]; where RB is the result of the last residual block, MaxPooling and AvgPooling are the max and average pooling layers, respectively, f 1*1 is a 1 * 1 size convolution layer, and σ is a sigmoid activation function. The "concat" represents the concatenation operation on the channel. The second channel passes through a 1*1 size convolution layer and a normalization layer, and is multiplied by the result of the first channel and then passes through a Softmax layer, and the specific formula is as follows: C = f1(x) T • F A (RB); wherein F A (RB) is the output of the first channel, f1(x) is a 1*1 size convolution layer and normalization layer processing function, C is the multiplication result of the two channels, and Q is the output of C after the SoftMax layer. The third channel only passes through a 1*1 size convolution layer, and is multiplied by the results of the other two channels.

7. The method of claim 1, wherein the method further comprises: The self-checking operation record is all alarm information operation types of each self-checking device.

Images