A rail vehicle parameter identification processing method, device and equipment

Abstract

The application provides a configuration and verification method and device for autonomously identifying train parameters and electronic equipment. The method comprises: identifying a target parameter phrase from original text recording train information by a Bert algorithm identification method in Python, writing the identified target parameter phrase into a text file, and forming an ATO parameter configuration file; automatically generating an ATO automatic driving dynamic library file by calling Microsoft Visual Studio through CMake; calling the ATO automatic driving dynamic library file by a traction calculation simulation platform, and generating a train simulation running curve to obtain configuration information of the train; and comparing and verifying the configuration information with the target parameter phrase in the original text. The vehicle interface text file can be automatically analyzed, that is, the rail vehicle parameters in the text file can be automatically identified, and then the ATO parameter configuration file can be easily generated.

Description

Technical Field

[0001] This invention belongs to the field of rail vehicle control technology, and particularly relates to a train parameter identification and processing method, device and equipment. Background Technology

[0002] In the existing technology, each time a rail vehicle company, such as a subway operating company, provides relevant documents such as vehicle operation, engineering technicians need to analyze hundreds of key information items such as traction, braking, level, and protection section in the vehicle documents one by one, manually modify the parameter configuration file in the source code of the ATO software, then compile and generate a dynamic library file and apply it to the traction calculation tool. Finally, they also need to modify some parameters in the traction calculation tool interface before they can generate the ATO vehicle control simulation curve of the vehicle running on the actual line.

[0003] Potential shortcomings in this process:

[0004] Increased time and manpower investment: After receiving the vehicle interface files, product department personnel first need to have engineering technicians in the R&D department locate the source code corresponding to the parameters, modify the parameters within the source code, and then compile the program. The compiled dynamic library then replaces the original dynamic library of the traction calculation tool. Each parameter modification takes approximately four hours. If further modifications are needed after the initial modification, it will significantly increase the time and manpower investment.

[0005] Accuracy of manual parameter modification: The vehicle interface file contains hundreds of parameters with inconsistent units. Manually modifying parameters at the source code level may result in incomplete or incorrect modifications, leading to ATO (Automatic Train Control) anomalies in the traction calculation tool's simulation. Checking parameters when errors occur requires comparing them one by one, which is time-consuming and labor-intensive. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure provides a method, apparatus, and device for identifying rail vehicle parameters. This device can automatically analyze vehicle interface text files, allowing engineers to identify rail vehicle parameters such as traction, braking, and delay in the text files without modifying the source code, thereby easily generating ATO parameter configuration files.

[0007] This disclosure provides a method for configuring and verifying autonomous train parameters, the method comprising,

[0008] The target parameter phrases are identified from the original text recording train information using the Bert algorithm in Python software. The identified target parameter phrases are written to a text file to form an ATO parameter configuration file.

[0009] The ATO autonomous driving dynamic library file is automatically generated by calling Microsoft Visual Studio via CMake;

[0010] The traction calculation simulation platform calls the ATO automatic driving dynamic library file and generates train simulation operation curves to obtain the train configuration information;

[0011] The configuration information is compared and verified with the target parameter phrases in the original text.

[0012] Optionally, the target parameter phrases can be identified from the original text recording train information using the BERT algorithm in Python software, including:

[0013] Extract target parameter phrases for rail vehicles from the original text and map the extracted target parameter phrases to a predefined vocabulary;

[0014] Based on the identifiers mapped to the target parameter phrases in the predefined vocabulary, they are transformed into a dense parameter word vector through the Embedding layer of the BERT model; the position index of the parameter phrases in the original text is encoded into a text position vector;

[0015] The parameter word vector and text position vector are logically combined to form a dual information vector;

[0016] The dual information vector is input into the encoder of the Transformer model.

[0017] Optionally, the method further includes a multi-head self-attention step:

[0018] The parameter word vectors and other word vectors are correlated using Scaled Dot-Product Attention to obtain the final correlation weight between the parameter word and each other word, thereby acquiring different semantic information.

[0019] Optionally,

[0020] Construct a key matrix for multiple other word vectors;

[0021] Determine the attn_scores, the correlation metric between the dot product of the parameter word vectors and the Key matrix;

[0022] Different attention weights are assigned to each word in the Context based on the magnitude of the dot product value in the dot product association metric.

[0023] Optionally, the parameter word vectors can be non-linearly transformed using two fully connected layers to obtain semantic feature representations.

[0024] Optionally,

[0025] The semantic feature representations are passed through residual connections and aggregated to a shallow layer to form a shallow semantic vector.

[0026] The shallow semantic vectors are semantically abstracted through an intermediate layer to form semantic abstract vectors;

[0027] The semantic abstract vector is subjected to deep semantic analysis to form a deep semantic vector;

[0028] The deep semantic vectors are then fed back to the shallow layer through residual connections to form complex semantic vectors.

[0029] Optionally, the complex semantic vector can be normalized using a layer normalization method.

[0030] This disclosure also provides a configuration and verification device for autonomously identifying train parameters. The device includes an identification unit, a call generation unit, a control unit, and a verification unit.

[0031] The recognition unit is configured to identify target parameter phrases from the original text recording train information using the Bert algorithm recognition method in Python software, write the identified target parameter phrases to a text file, and form an ATO parameter configuration file.

[0032] The invocation unit is configured to automatically generate the ATO autonomous driving dynamic library file by calling Microsoft Visual Studio via CMake;

[0033] The control unit is configured to control the traction calculation simulation platform to call the ATO automatic driving dynamic library file, generate train simulation operation curves, and obtain train configuration information.

[0034] The verification unit is configured to compare and verify the configuration information with the target parameter phrases in the original text.

[0035] Optionally, the recognition unit performs the recognition of target parameter phrases from the original text recording train information using the BERT algorithm recognition method in Python software, including:

[0036] Extract target parameter phrases for rail vehicles from the original text and map the extracted target parameter phrases to a predefined vocabulary;

[0037] Based on the identifiers mapped to the target parameter phrases in the predefined vocabulary, they are transformed into a dense parameter word vector through the Embedding layer of the BERT model; the position index of the parameter phrases in the original text is encoded into a text position vector;

[0038] The parameter word vector and text position vector are logically combined to form a dual information vector;

[0039] The dual information vector is input into the encoder of the Transformer model.

[0040] This disclosure also provides an electronic device comprising at least one processor and at least one memory, wherein the memory is data-connected to the processor, wherein...

[0041] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0042] Compared with the prior art, this disclosure has the following advantages:

[0043] In this embodiment, engineers do not need to modify the source code. The Bert algorithm in Python software can automatically identify train configuration parameters such as traction, braking, and delay in text files and generate an ATO parameter configuration file. Then, CMake calls Microsoft Visual Studio software to automatically update and compile this configuration file and generate a dynamic library file. Finally, the traction calculation tool automatically calls this dynamic library file, which allows for a direct view of the impact of changes in project parameters on the vehicle motion curve. This greatly reduces the input of manpower and time costs, making simulation analysis and control very convenient.

[0044] 1. Resolved the issue that configuring vehicle interface file parameters in the traction calculation tool required manual modification of the source code.

[0045] 2. Resolved the issue of ATO vehicle control malfunction caused by incorrect manual parameter modification.

[0046] 3. It solves the problem of time-consuming and laborious verification and checking of parameter errors.

[0047] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 An overall system block diagram of an implementation of the identification processing method according to an embodiment of the present disclosure is shown;

[0050] Figure 2 This illustration shows a flowchart of a method for autonomously identifying and verifying train parameter configurations according to an embodiment of the present disclosure.

[0051] Figure 3 A schematic diagram of a vehicle interface file page is shown;

[0052] Figure 4 A structural block diagram of a configuration and verification device for autonomously identifying train parameters according to an embodiment of the present disclosure is shown;

[0053] Figure 5 A block diagram of an electronic device structure according to an embodiment of the present disclosure is shown. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Figure 1 An overall system block diagram of an implementation of an identification processing method according to an embodiment of this disclosure is shown. Figure 1 As shown, the system includes a computing device that executes a rail vehicle parameter identification method. This computing device can obtain the aforementioned vehicle interface file from the train or a related server through its interface. The computing device can be a computer, processor, etc.

[0056] Figure 2 A schematic flowchart illustrating a method for autonomously identifying and verifying train parameter configurations according to an embodiment of this disclosure is shown. Figure 2As shown, the computing device can obtain the project vehicle interface file from the subway operating company's server. Based on the project text file provided by the subway operating company or subway vehicle department's server, engineers do not need to modify the source code. In this embodiment, the BERT software in Python can autonomously identify parameters such as traction, braking, and delay in the text file and generate an ATO parameter configuration file config.c. This configuration file is then automatically compiled in Microsoft Visual Studio to generate a dynamic library file. Finally, the traction calculation simulation platform automatically calls this dynamic library file to visually present the train operation curve as a time-speed and time-level graph. This allows for a direct view of the impact of changes in project parameters on the train operation curve, and enables the verification and diagnosis of parameter correctness based on the vehicle operation curve. The generated train simulation operation curve can also identify key parameters, such as average mileage speed, maximum operating speed, common braking deceleration, or departure speed, which can then be compared with the obtained project interface file for verification.

[0057] In this embodiment of the disclosure, the identification of keywords using BERT will be further explained.

[0058] The target parameter phrases for rail vehicles are extracted from the original text and mapped to a predefined vocabulary. Based on the identifiers mapped to the predefined vocabulary, the target parameter phrases are converted into a 256-dimensional dense parameter word vector through the embedding layer of the BERT model. The position index of the parameter phrases in the original text is encoded into a text position vector. The parameter word vector and the text position vector are logically combined to form a dual information vector. The dual information vector is then input into the encoder of the Transformer model.

[0059] This disclosure is based on project documents provided by the subway operation and subway vehicle departments. The original documents may have mentioned terms such as "Automatic Train Operation System" and "ATO signal control mode". However, it should be noted that this disclosure is not limited to subways and can be applied to the entire rail transit field.

[0060] For example, such as Figure 2 The vehicle interface file within the project text file shown can contain the signaling system's basic vehicle parameters, traction, and braking performance parameters. During the early design liaison phase of the project, the vehicle supplier provides design simulation data as input for the initial design of the Automatic Traffic Control (ATC) system. After the first train completes type testing, the vehicle supplier provides actual data as input for the actual vehicle control of the ATC system.

[0061] The vehicle interface file may also include the following parameters: train speed (vehicle design speed, maximum operating speed, speed threshold during emergency braking, speed threshold during traction cut-off), train weight, train dimensions, parameters related to the installation of the speed pulse sensor, parameters related to the installation of the BTM antenna, door operation process duration, traction and braking performance parameters, train delay index, traction cut-off index, slippage index, emergency braking index, service braking index, mixed braking index, and parameters related to train resistance calculation.

[0062] This disclosure uses the vehicle parameter "maximum operating speed" as an example to explain how it is processed internally within the BERT model:

[0063] 1. Lexical mapping

[0064] The parameter phrase “maximum operating speed” is mapped to a unique ID in a predefined vocabulary, such as 957.

[0065] 2. Convert to vector

[0066] ID957 is transformed into a 256-dimensional dense word vector through the BERT model's embedding layer, representing the initial semantics of the parameter phrase.

[0067] In this embodiment of the disclosure, 256-dimensional dense word vectors are a vectorization method for providing a detailed and continuous representation of word semantics.

[0068] The parameter phrase "maximum operating speed" was converted to word ID: 957

[0069] ID957 enters the BERT embedding layer and is mapped to a 256-dimensional word vector:

[0070] [0.21,0.35,0.62,1.32,...,0.17,-2.03,1.19,-0.37].

[0071] The word vector contains 256 real values, representing 256 semantic features of the current word. For example, the first dimension, 0.23, indicates the relevance of the word to the meaning of "speed," while the 124th dimension, 1.32, indicates the match between the word and the meaning of "operation." Each dimension reflects a specific semantic attribute of the word. These 256 dimensions provide a multi-dimensional and comprehensive description of the semantic components of a word.

[0072] In this embodiment, the continuous and dense representation is more powerful and better suited to the neural network structure. Through model training, the 256-dimensional uniform distribution of word vectors describes rich semantic information.

[0073] 3. Add position encoding

[0074] The position index of the parameter phrase in the original text, for example 35, is encoded into a vector representing its position information in the sequence.

[0075] For example, the position index of the phrase "maximum operating speed" in the input sequence is 35.

[0076] 1) This position index 35 will be encoded into a position vector of the same length as the word vector dimension, which is 256 dimensions in this case.

[0077] 2) The value of each dimension in the position vector is a function of the position index 35.

[0078] For example:

[0079] pos_enc

[35] =[sin(35 / 10000^2 / i),cos(35 / 10000^2 / i)];

[0080] Where i is the dimension index, ranging from 0 to 255.

[0081] 3) A specific example of a positional encoding vector is as follows:

[0082] [sin(35 / 10000^2 / 0),cos(35 / 10000^2 / 0),...,sin(35 / 10000^2 / 255),cos(35 / 10000^2 / 255)]

[0083] 4) Summing this positional encoding vector with the word vector of the parameter phrase:

[0084] Word vectors plus position vectors become word vector inputs for sequence information enhancement.

[0085] Therefore, the position index is encoded as a fixed-length symbolic vector, which incorporates sequence order information and is beneficial for the model to understand the context of words.

[0086] 4. Enter Transformer

[0087] The word vector representing the "maximum operating speed" is combined with the position vector and input into BERT's TransformerEncoder.

[0088] The specific process can be as follows:

[0089] 1) Convert the parameter phrases into 256-dimensional word vectors to represent their semantics:

[0090] word_vec = [0.21, 0.35, 0.62, ...] #256 dimensions

[0091] 2) The phrase's position in the sequence, index = 35, is encoded as a position vector:

[0092] pos_vec = [sin(35 / 1000^(2*i / d)); i ranges from 0 to 255]

[0093] 3) Word vectors and position vectors are added element-wise to combine them:

[0094] input_vec = word_vec + pos_vec

[0095] 4) input_vec then becomes a vector containing both semantic and positional information, such as:

[0096] [-0.02, 1.41, 0.87, ...]#256 dimensions

[0097] 5) The effect of this additive combination is:

[0098] -The original meaning is preserved.

[0099] - Added semantic components related to sequential position.

[0100] 6) Input_vec passes more information into the Transformer Encoder.

[0101] - Enables parameter semantic representation to contain rich context-dependent information

[0102] The simple additive combination of word / position vectors, which mixes semantic and sequential information, makes BERT's understanding of the parameter text more comprehensive.

[0103] 5. Bullish Self-Attention

[0104] The parameter word vectors are correlated with other word vectors (context, descriptive words) through Scaled Dot-Product Attention to obtain the final correlation weight between the parameter word and each other word, thus obtaining different semantic information.

[0105] For example, the calculation process using Dot-Product Attention is as follows:

[0106] 1) The word vector of "maximum operating speed" is used as the query, and it has 256 dimensions:

[0107] query_vec=[0.21,0.35,0.62,...,0.06];

[0108] 2) CONTEXT_WORDS represents other words in the text. A key matrix of these word vectors is constructed, with dimensions 64x256.

[0109] key_mat=[[...],[...],...,[...]]#64x256;

[0110] 3) Calculate the dot product correlation metric between the query vector and the key matrix:

[0111] attn_scores=key_mat·query_vec

[0112] attn_scores = [0.02, 0.03, 0.10, ...] #64 dimensions; where 64 represents the number of keywords. In this embodiment, 64 context words related to the current "maximum operating speed" are selected, and these 64 words are represented by word vectors. The vectors are concatenated together to form the Key matrix.

[0113] 4) In attn_scores, each dimension corresponds to the dot product of a Context word and a Query word. The Query word is the term to be analyzed, such as "maximum operating speed"; the Context words are the 64 words in the text that are related to the Query word. The first dimension of attn_scores is the dot product of the Query word and Context word 1, the second dimension is the dot product of the Query word and Context word 2, and so on.

[0114] 5) A larger dot product value indicates a higher semantic correlation between the two words;

[0115] 6) Finally, assign different attention weights to each word in the Context based on the magnitude of the dot product.

[0116] Dot product calculation can evaluate the directional similarity between word vectors, which can be used to determine the degree of semantic dependence and association between words and guide the allocation of attention weights.

[0117] 6. Feedforward Fully Connected Network

[0118] The parameter word vectors undergo non-linear transformation (matrix multiplication) through two fully connected layers to obtain a higher-level feature representation.

[0119] Examples are as follows:

[0120] 1) "Maximum operating speed" is mapped to word vectors, 256 dimensions:

[0121] word_vec=[0.21,0.35,0.62,...]

[0122] 2) The first fully connected matrix W1, with a size of 256x1024, maps word vectors to 1024 dimensions.

[0123] 3) First-level calculation:

[0124] h1 = word_vec * W1

[0125] h1 is a projection of a higher-dimensional space with 1024 dimensions.

[0126] 4) Nonlinear transformation using the ReLU activation function:

[0127] h2 = ReLU(h1)

[0128] 5) The second layer fully connected matrix W2, with a size of 1024x256, has compressed dimensions.

[0129] 6) Second-level calculation:

[0130] z = h2 * W2 ultimately outputs a 256-dimensional z, obtaining a richer and more abstract semantic feature expression to more comprehensively represent the semantic connotation of the parameter "maximum operating speed".

[0131] In this way, through two layers of nonlinear transformation, the parameter word vectors are transformed into new high-level semantic feature representations that integrate more contextual dependencies.

[0132] 7. Residual Connection

[0133] Deeper semantic information is passed and aggregated to the shallower layers via residual connections. In this embodiment, the shallow layer refers to the first few Transformer Block layers near the input layer in the BERT model, such as the first or first two Block layers. These layers are responsible for extracting relatively intuitive and simple semantic information from the input words, with a lower level of abstraction. The deep layer refers to the later Transformer Block layers near the output layer in the BERT model, such as the last one or two Block layers. These layers contain highly abstract and complex semantic information representations of words after multiple rounds of bottom-up semantic abstraction and aggregation, and have a stronger ability to understand context. The basic process is as follows:

[0134] 1) Word vectors are first mapped to vectors representing shallow semantics;

[0135] 2) Semantic abstraction is performed step by step through intermediate layers;

[0136] 3) Generate vectors representing deep semantics;

[0137] 4) Deep semantic vectors are fed back to shallow vectors via residual connections;

[0138] 5) Shallow vectors acquire complex semantics, generating fused semantic vectors.

[0139] 8. Layer Normalization

[0140] Optimizing the numerical distribution of parameter word vectors reduces random fluctuations during model training and accelerates the convergence of the neural network.

[0141] Layer normalization optimizes the parameter word vectors by adjusting the numerical distribution of the input vectors, thus accelerating the convergence of the BERT model. The main methods include:

[0142] 1) Shifting the mean: Calculate the mean μ along the dimension of the word vector and subtract μ from each element.

[0143] Original word vectors: x = [0.21, 0.35, 0.62, ...], μ = 0.39

[0144] 2) Adjusting the variance: Calculate the variance σ of each dimension of the word vector and divide each element by σ.

[0145] Variance: σ = 0.52

[0146] 3) Scaling and shifting: Learnable scaling factors γ and bias β are introduced again on the normalized word vectors.

[0147] Normalization: x^=(x-μ) / σx^=[-0.34,-0.07,0.43,...]

[0148] Scaling and displacement: y = γx^+βy = [0.21, 0.15, 0.83, ...]

[0149] 4) More normalized output distribution: After the above transformation, the value distribution of word vectors is adjusted to have a lower variance and a mean close to 0.

[0150] 5) Accelerate model convergence: More stable numerical values ​​are beneficial for model optimization, alleviate internal random fluctuations, and reduce computational difficulty.

[0151] Through the above processing, the semantics of the parameter words are refined layer by layer and fused with the semantics of the context to output the probability of them being keywords. The optimized vector calculated through the previous steps represents the probability of relevance.

[0152] This disclosure provides a method for autonomously identifying and verifying train parameters. Based on project text files provided by the subway operation and subway vehicle departments, engineering technicians do not need to modify the source code. Using BERT software in Python, the method can autonomously identify parameters such as traction, braking, and delay in the text file and generate an ATO parameter configuration file. This configuration file is then automatically compiled in Microsoft Visual Studio software to generate a dynamic library file. Finally, the traction calculation tool automatically calls this dynamic library file to visually present the train operation curve as a time-speed and time-level graph. This allows for a clear view of the impact of changes in project parameters on the train operation curve and enables the verification and diagnosis of parameter correctness based on the vehicle operation curve.

[0153] This embodiment employs the Transformer model to identify and process train parameters in rail vehicles. Its main advantages are:

[0154] 1. More powerful semantic understanding capabilities

[0155] The Transformer model learns the semantic information of text through the Multi-Head Attention mechanism, which can capture more complex semantic relationships, rather than just word frequency statistics.

[0156] 2. Consider the global dependencies between words.

[0157] Self-Attention in Transformer can model dependencies between words at arbitrary distances, while Python only considers local statistics.

[0158] 3. Utilize contextual information

[0159] Transformer Encoder can make full use of contextual information to infer the meaning of words, while Python has a weaker understanding of context.

[0160] 4. Knowledge of pre-trained language models

[0161] Existing Transformer models can load pre-trained parameters and utilize knowledge trained on large corpora, making them more accurate than Python Stats.

[0162] 5. More suitable for long texts

[0163] Python keyword extraction performs poorly on long texts, while Transformer can handle texts of any length.

[0164] 6. Higher accuracy

[0165] In multiple keyword extraction tasks, the Transformer model significantly outperformed the keyword extraction using only Python statistical methods.

[0166] Based on the above method, this disclosure also provides a configuration and verification device for autonomously identifying train parameters, such as... Figure 4 The device includes an identification unit, a call generation unit, a control unit, and a verification unit. The identification unit is configured to identify target parameter phrases from the original text recording train information using the BERT identification method, write the identified target parameter phrases to a text file, and form an ATO parameter configuration file. The call generation unit is configured to automatically generate an ATO automatic driving dynamic library file by calling Microsoft Visual Studio via CMake. The control unit is configured to control the traction calculation simulation platform to call the ATO automatic driving dynamic library file, generate train simulation operation curves, and obtain train configuration information. The verification unit is configured to compare and verify the configuration information with the target parameter phrases in the original text. The recognition unit performs the following steps to identify target parameter phrases from the original text recording train information using the BERT recognition method: extracting target parameter phrases for rail vehicles from the original text; mapping the extracted target parameter phrases to a predefined vocabulary; converting the target parameter phrases into a dense parameter word vector through the Embedding layer of the BERT model based on the identifiers mapped to the predefined vocabulary; encoding the position index of the parameter phrases in the original text into a text position vector; logically combining the parameter word vector and the text position vector to form a dual information vector; and inputting the dual information vector into the encoder of the Transformer model.

[0167] Based on the above disclosure, this disclosure also provides an electronic device. For example... Figure 5 As shown, the electronic device of this disclosure includes at least one processor electrically connected to the present invention and at least one memory electrically connected to the processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method steps as executed by the controller above.

[0168] In this embodiment of the disclosure:

[0169] 1. Engineering technicians do not need to modify the source code. The Bert software in Python can automatically identify parameters such as traction, braking, and delay in the vehicle interface file and generate the ATO parameter configuration file config.c. config.c is a C file in the source code.

[0170] 2. By calling Microsoft Visual Studio software through CMake, this configuration file is automatically updated, compiled, and a dynamic library file is generated. Finally, the traction calculation tool automatically calls this dynamic library file to present the train operation curve in a visually intuitive time-speed and time-level graph. This allows for a clear view of the impact of changes in project parameters on the train operation curve.

[0171] 3. By calling BERT in Python software through CMake, the key information in the train operation curve is compared with the parameters in the vehicle interface file to automatically check the correctness of the curve.

[0172] 4. The entire process does not require engineers to adjust parameters. It enables one-click conversion and verification of vehicle interface files to simulated train operation curves, greatly improving work efficiency and quality.

[0173] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for configuring and verifying autonomous train parameters, the method comprising: The target parameter phrases are identified from the original text recording train information using the BERT algorithm in Python software. The identified target parameter phrases are written to a text file to form an ATO parameter configuration file, which is specifically the config.c file in the source code. By calling Microsoft Visual Studio through CMake, the config.c file is automatically updated and compiled, automatically generating the ATO autonomous driving dynamic library file; The traction calculation simulation platform calls the ATO automatic driving dynamic library file and generates train simulation operation curves to obtain train configuration information. The train simulation operation curves specifically include time-speed curves and time-level curves. The configuration information includes key parameters identified based on the train simulation operation curves. The key parameters include at least one of the following: average mileage speed, maximum operating speed, common braking deceleration, or station departure speed. The configuration information is compared and verified with the target parameter phrases in the original text to reverse-engineer and diagnose the correctness of the parameters.

2. The method according to claim 1, wherein, The BERT algorithm in Python software is used to identify target parameter phrases from raw text recording train information, including: Extract target parameter phrases for rail vehicles from the original text and map the extracted target parameter phrases to a predefined vocabulary; Based on the identifiers mapped to the target parameter phrases in the predefined vocabulary, they are transformed into a dense parameter word vector through the Embedding layer of the BERT model; the position index of the parameter phrases in the original text is encoded into a text position vector; The parameter word vector and text position vector are logically combined to form a dual information vector; The dual information vector is input into the encoder of the Transformer model.

3. The method according to claim 2, wherein, The method also includes a multi-head self-attention step: The parameter word vectors and other word vectors are correlated using Scaled Dot-Product Attention to obtain the final correlation weight between the parameter word and each other word, thereby acquiring different semantic information.

4. The method according to claim 3, wherein, Construct a key matrix for multiple other word vectors; Determine the attn_scores, the correlation metric between the dot product of the parameter word vectors and the Key matrix; Different attention weights are assigned to each word in the Context based on the magnitude of the dot product value in the dot product association metric.

5. The method according to claim 2, wherein, The semantic feature representation is obtained by performing a nonlinear transformation on the parameter word vectors through two fully connected layers.

6. The method according to claim 5, wherein, The semantic feature representations are passed through residual connections and aggregated to a shallow layer to form a shallow semantic vector. The shallow semantic vectors are semantically abstracted through an intermediate layer to form semantic abstract vectors; The semantic abstract vector is subjected to deep semantic analysis to form a deep semantic vector; The deep semantic vectors are then fed back to the shallow layer through residual connections to form complex semantic vectors.

7. The method according to claim 6, wherein, The complex semantic vector is normalized using a layer normalization method.

8. A configuration and verification device for autonomously identifying train parameters, the device comprising an identification unit, a call generation unit, a control unit, and a verification unit, wherein... The recognition unit is configured to identify target parameter phrases from the original text recording train information using the Bert algorithm recognition method in Python software, write the identified target parameter phrases to a text file, and form an ATO parameter configuration file, specifically the config.c file in the source code. The call generation unit is configured to call Microsoft Visual Studio via CMake to automatically update and compile the config.c file and automatically generate the ATO autonomous driving dynamic library file; The control unit is configured to control the traction calculation simulation platform to call the ATO automatic driving dynamic library file, generate train simulation operation curves, and obtain train configuration information. The train simulation operation curves specifically include time-speed curves and time-level curves. The configuration information includes key parameters identified based on the train simulation operation curves. The key parameters include at least one of the following: average mileage speed, maximum operating speed, common braking deceleration, or station departure speed. The verification unit is configured to compare the configuration information with the target parameter phrases in the original text to verify and diagnose the correctness of the parameters.

9. The apparatus according to claim 8, wherein, The recognition unit executes the BERT algorithm in Python software to identify target parameter phrases from the raw text recording train information, including: Extract target parameter phrases for rail vehicles from the original text and map the extracted target parameter phrases to a predefined vocabulary; Based on the identifiers mapped to the target parameter phrases in the predefined vocabulary, they are transformed into a dense parameter word vector through the Embedding layer of the BERT model; the position index of the parameter phrases in the original text is encoded into a text position vector; The parameter word vector and text position vector are logically combined to form a dual information vector; The dual information vector is input into the encoder of the Transformer model.

10. An electronic device comprising at least one processor and at least one memory, the memory being data-connected to the processor, wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

Images