LKJ braking distance calculation optimization method and system based on deep learning model

Abstract

The application discloses an LKJ braking distance calculation optimization method and system based on a deep learning model, relates to the technical field of rail transit, and optimizes the method by collecting braking parameters, braking initial speed and line data of a track where a target train is currently located of the target train, determining a target distance according to the line data, inputting the braking parameters and the target distance into a trained braking distance prediction model, determining a speed-distance curve of the target train, and determining a braking distance of the target train according to the braking initial speed and the speed-distance curve. The application calculates and optimizes the LKJ braking distance through the deep learning model, monitors and predicts the train operation state in real time, reduces the risk of accidents, reduces train delays and equipment maintenance time, and improves the overall efficiency of railway transportation.

Description

Technical Field

[0001] This invention relates to the field of rail transit technology, and in particular to an optimization method and system for calculating LKJ braking distance based on a deep learning model. Background Technology

[0002] Against the backdrop of continuous upgrades to the railway system, railway operation technology and management levels are improving in tandem, with increasingly stringent standards for railway operational efficiency and safety. To enhance railway transport capacity, it is necessary to continuously reduce train intervals while ensuring safety, making the optimization of braking distance a core element in improving operational efficiency. The Train Control and Monitoring System (LKJ) is a train control device independently developed by Chinese railway professionals, playing a crucial role in ensuring railway transport safety.

[0003] LKJ (Locomotive Braking Control) is a monitoring device installed on locomotives and EMUs (Electric Multiple Units) to prevent locomotives from overshooting and causing speeding accidents, and to assist the driver in operation. The train braking curve is a key component of the LKJ control mode. In practical engineering applications, due to the limited computing power of the onboard system, the calculation of the train braking curve in LKJ mode requires simplification of some complex theoretical calculation models. This results in a deviation between the calculated train braking distance and the actual train braking distance, leading to unnecessary braking output or overly conservative speed limits in the LKJ-generated braking curve.

[0004] Current methods for predicting train braking distance mainly employ calculations based on traditional physical models such as train dynamics models and train air braking system models. These methods have the following problems: (1) Insufficient handling of nonlinear factors: Traditional models are difficult to accurately describe the nonlinear relationships in the braking process, such as the change of wheel-rail friction coefficient and the dynamic characteristics of brake cylinder pressure.

[0005] (2) Poor adaptability to complex working conditions: The train operating environment is complex and varied, including different gradients, curves, loads and weather conditions, and existing models are difficult to fully cover these factors.

[0006] (3) The contradiction between real-time performance and accuracy: Although the calculation method based on analytical formulas has good real-time performance, it is not accurate enough under complex working conditions; while the calculation of high-precision physical models takes a long time and is difficult to meet the real-time control requirements.

[0007] (4) Difficulty in model parameter calibration: Key parameters in traditional models (such as the converted friction coefficient and the converted braking rate of the train) need to be determined through special tests. The calibration process is complex and difficult to adapt to dynamic changes.

[0008] (5) Low data utilization: A large amount of data is accumulated during the operation of the traction locomotive, but traditional methods have failed to make full use of this data for model optimization and parameter updates.

[0009] Therefore, given the crucial role trains play in the railway industry, promoting research on train braking distance, developing new optimization algorithms, and achieving the goals of shortening train braking distance, improving the accuracy of train braking distance calculation, and increasing railway transportation efficiency are the directions that railway industry researchers should strive for. This will further ensure transportation safety and efficiency and bring new development opportunities to the railway industry's train operation control system. Summary of the Invention

[0010] In view of this, the present invention provides an optimization method and system for LKJ braking distance calculation based on a deep learning model, focusing on the calculation of braking distance of locomotive-pulled trains. Under the existing algorithm framework, a more accurate optimization calculation model is studied, which can not only effectively reduce the braking distance of trains, but also provide important theoretical basis and technical support for improving train operation safety, implementing line reconstruction projects, and optimizing transportation efficiency.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: An optimization method for calculating LKJ braking distance based on a deep learning model includes: S1 Data Acquisition Steps: Acquire the braking parameters, initial braking speed, and track data of the target train's current location on the track. S2 Distance Determination Steps: Determine the target distance based on the track data; the target distance represents the distance between the target train's current position on the track and the braking target point; S3 curve determination steps: Input the braking parameters and target distance into the trained braking distance prediction model to determine the speed-distance curve of the target train; S4 Braking Determination Steps: Based on the initial braking speed, determine the braking distance of the target train using the speed-distance curve.

[0012] In the above method, the optional braking parameters of the target train in S1 include: the calculated gradient of the braking section in per mille, the train braking pressure reduction, the converted braking rate of the train, the braking calculation coefficient, the converted friction coefficient, the basic unit resistance of the train, and the number of vehicles.

[0013] Optionally, the training process of the braking distance prediction model in S3 includes the following steps: Processing steps: Acquire train data, preprocess the train data and divide it into sets according to a preset ratio to obtain training set, validation set and test set; Model initialization steps: Initialize the braking distance prediction model based on train data to obtain the current braking distance prediction model; Forward propagation steps: Input the training set into the current braking distance prediction model for training. After forward propagation calculation, the training prediction set is obtained. Backpropagation steps: Determine the loss function value of the current braking distance prediction model based on the training prediction set, and perform backpropagation calculation based on the loss function value of the current braking distance prediction model to obtain the adjustment gradient; Model adjustment steps: Based on the adjusted gradient and validation set, perform hyperparameter optimization on the current braking distance prediction model to obtain the adjusted braking distance prediction model; Model evaluation steps: Evaluate the performance of the adjusted braking distance prediction model based on the test set. If it meets the performance requirements, the adjusted braking distance prediction model is determined as the trained braking distance prediction model; if it does not meet the performance requirements, return to the forward propagation step.

[0014] In the above method, the train data may optionally include time-series characteristic data and static characteristic data.

[0015] In the above methods, the optional hyperparameter optimization of the model tuning step may include: grid search optimization, random search optimization, Bayesian optimization, and genetic algorithm optimization.

[0016] In the above methods, optionally, the performance evaluation of the model evaluation step specifically includes: regression performance indicators, braking safety indicators, and statistical significance test indicators.

[0017] Optionally, in the above method, the braking distance prediction model in S3 is a hybrid model based on the LSTM-Attention structure.

[0018] Among the above methods, the optional braking distance prediction models include: CNN-LSTM hybrid model, LSTM-Attention hybrid model and Transformer-LSTM hybrid model.

[0019] In the above method, optionally, in S3, the trained braking distance prediction model determines the speed-distance curve of the train based on the braking distance calculation formula of the train pulled by the locomotive, through the optimized braking distance calculation formula of the train pulled by the locomotive obtained by training optimization. The formula for calculating the braking distance of a train pulled by a locomotive is as follows: ; In the formula, S k Indicates the braking distance during idle driving. S e Indicates the effective braking distance. Indicates the initial braking speed of the train. Indicates the train's empty travel time. Indicates the final braking speed. Indicates the braking calculation coefficient. This indicates the converted friction coefficient of the brake shoe. Indicates the converted braking rate of the train. Indicates the commonly used braking coefficient. Indicates the basic resistance per unit of the train. The gradient percentage is added to the braking section; the train's idle travel time indicates the equipment's reaction time after the train driver issues a braking command.

[0020] The optimized formula for calculating the braking distance of a locomotive-hauled train is as follows:

[0021] In the formula, Indicates the distance traveled without moving. S k Divided into multiple intervals n After that, the i The initial braking velocity of each interval, Indicates the first n The initial braking velocity of each interval.

[0022] A deep learning model-based LKJ braking distance calculation optimization system is used to implement any of the above-described deep learning model-based LKJ braking distance calculation optimization methods, including a data acquisition module, a distance determination module, a curve determination module, and a braking determination module connected in sequence. The data acquisition module is used to collect the braking parameters, initial braking speed, and track data of the target train's current location. The distance determination module is used to determine the target distance based on the track data; the target distance represents the distance between the target train's current position on the track and the braking target point. The curve determination module is used to input braking parameters and target distance into a trained braking distance prediction model to determine the speed-distance curve of the target train. The braking determination module is used to determine the braking distance of the target train based on the initial braking speed and using a speed-distance curve. As can be seen from the above technical solution, compared with the prior art, the present invention provides an LKJ braking distance calculation optimization method and system based on a deep learning model, which has the following beneficial effects: (1) This invention uses a deep learning model to predict and calculate the braking distance of a train, thereby determining the braking distance of the train. Compared with the traditional physical model, the braking distance of the train determined by the deep learning model is shorter and more accurate. (2) This invention monitors and predicts the train's operating status in real time, promptly identifies and addresses potential safety hazards, reduces the risk of accidents, and improves the safety performance of the train braking process. (3) The commonly used braking distance of trains calculated by the optimization algorithm of this invention can shorten the train tracking interval, increase the traffic density, and improve the railway throughput capacity; at the same time, the optimized control strategy and fault diagnosis function reduce train delays and equipment maintenance time, further improving the overall efficiency of railway transportation. (4) The reduction in energy consumption and equipment maintenance costs has saved the railway company a lot of operating costs and enhanced the economic benefits of railway operation. Attached Figure Description

[0023] 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 only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the LKJ braking distance calculation optimization method based on a deep learning model disclosed in this invention; Figure 2 This is a schematic diagram of the training process of the braking distance prediction model disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the train braking speed-distance curve obtained based on the network model disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of the train braking speed-distance curve obtained by the LKJ braking distance calculation optimization method based on a deep learning model disclosed in this invention. Detailed Implementation

[0025] 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, and 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.

[0026] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0027] Reference Figure 1 This invention discloses an optimization method for calculating LKJ braking distance based on a deep learning model, comprising: S1 Data Acquisition Steps: Acquire the braking parameters, initial braking speed, and track data of the target train's current location.

[0028] S2 Distance Determination Steps: Determine the target distance based on the track data; the target distance represents the distance between the target train's current position on the track and the braking target point.

[0029] S3 curve determination steps: Input the braking parameters and target distance into the trained braking distance prediction model to determine the speed-distance curve of the target train.

[0030] S4 Braking Determination Steps: Based on the initial braking speed, determine the braking distance of the target train using the speed-distance curve.

[0031] Furthermore, the braking parameters of the target train in S1 include: the gradient of the braking section (in thousands), the train's braking pressure reduction, the train's converted braking rate, the service braking coefficient, the braking calculation coefficient, and the number of carriages.

[0032] Reference Figure 2 Optionally, the training process for the braking distance prediction model in S3 includes the following steps: Processing steps: Acquire train data, preprocess the train data and divide it into sets according to a preset ratio to obtain training set, validation set and test set.

[0033] Model initialization steps: Initialize the braking distance prediction model based on train data to obtain the current braking distance prediction model.

[0034] Forward propagation steps: Input the training set into the current braking distance prediction model for training. After forward propagation calculation, the training prediction set is obtained.

[0035] Backpropagation steps: Determine the loss function value of the current braking distance prediction model based on the training prediction set, and perform backpropagation calculation based on the loss function value of the current braking distance prediction model to obtain the adjustment gradient.

[0036] Model adjustment steps: Based on the adjusted gradient and validation set, perform hyperparameter optimization on the current braking distance prediction model to obtain the adjusted braking distance prediction model; Model evaluation steps: Evaluate the performance of the adjusted braking distance prediction model based on the test set. If it meets the performance requirements, the adjusted braking distance prediction model is determined as the trained braking distance prediction model; if it does not meet the performance requirements, return to the forward propagation step.

[0037] Furthermore, the train data acquired in the processing steps includes time-series feature data and static feature data.

[0038] Optionally, the hyperparameter optimization in the model tuning step may include: grid search optimization, random search optimization, Bayesian optimization, and genetic algorithm optimization.

[0039] Optionally, the performance evaluation of the model evaluation step may specifically include: regression performance indicators, braking safety indicators, and statistical significance test indicators.

[0040] Before the training of the braking distance prediction model begins, the present invention designs the train braking training scheme and the data required for training as follows, so that the braking distance prediction model can be trained smoothly.

[0041] Specifically, S1 also includes: developing training strategies based on the acquired train data.

[0042] This invention employs a batch training method for the braking distance prediction model, dividing all acquired train data into multiple batches, typically ranging from 32 to 128. Simultaneously, the acquired data is divided according to training function, specifically into a training set, a validation set, and a test set, with a typical ratio of 7:1:2. The training set is used for model parameter learning, the validation set for model hyperparameter adjustment and training process monitoring, and the test set for final model performance evaluation.

[0043] In addition, based on the training status of existing models, this invention analyzes the number of training rounds for the braking distance prediction model and initially determines that the number of training rounds is 100-500 rounds based on the model convergence. In the actual training process, a more refined number of rounds is preset according to the specific convergence status of the model.

[0044] Since the braking distance prediction model predicts the distance of a moving train, the acquired train data includes various forms of data, such as: time-series feature datasets, including time-series data on speed, braking pressure, etc., during the braking process; static feature datasets, including static data such as train load, track gradient, and environmental conditions; and label datasets, including actual measured braking distance data.

[0045] Furthermore, to ensure the balance of braking distance prediction data from the braking distance prediction model, the acquired train data must be rich and diverse, ensuring a relatively balanced distribution of data across different operating conditions (such as different loads and gradients), and preventing the braking distance prediction model from favoring a particular type of operating condition. Specifically, for operating conditions with limited data, oversampling, undersampling, or data augmentation methods can be used to achieve balance.

[0046] Furthermore, in S1, preprocessing refers to the standardization of train data.

[0047] Standardizing the input features to have a mean of 0 and a standard deviation of 1 helps improve model training speed and stability. Standardization parameters should be calculated based on the training set and then applied to the validation and test sets to avoid data leakage.

[0048] More specifically, in the train data, the time-series feature dataset includes data closely related to determining the braking distance, such as the speed change sequence and braking pressure change sequence during the train braking process. These are the main influencing factors, therefore the time-series features of the train are used as the primary input features. The static feature dataset, on the other hand, consists of secondary influencing factors such as train load, track gradient, and environmental conditions, and the static features of the train are used as auxiliary input features. Based on the typical duration of the braking process, the data input window size of the braking prediction model is set to 100-500 time steps, corresponding to a braking process of 10-50 seconds.

[0049] Optionally, the loss function in the backpropagation step includes mean squared error, mean absolute error, Huber loss, and a custom loss function.

[0050] Among them, Mean Squared Error (MSE), as a commonly used regression loss function, is sensitive to the difference between the predicted and the true values; Mean Absolute Error (MAE) is more robust to outliers than MSE and is more in line with practical application needs; Huber loss combines the advantages of MSE and MAE and has a certain degree of robustness to outliers; Custom loss functions can be designed to give different weights to small and large errors according to braking safety requirements, such as giving higher penalties to errors that exceed the safety threshold.

[0051] Furthermore, before determining the loss function value of the braking distance prediction model, the loss function needs to be optimized. Specifically, the loss function includes: quantile loss, weighted loss function, and safety-based loss function.

[0052] Quantile loss is a quantile that can be controlled in predicting outcomes and helps adjust the safety margin.

[0053] The weighted loss function can assign different weights to different braking conditions (such as high-speed braking and heavy-load braking), thereby improving the prediction accuracy for specific conditions.

[0054] A safety-based loss function imposes a higher penalty on predictions below the braking distance, thereby increasing the safety margin.

[0055] Methods for optimizing loss functions include: stochastic gradient descent, Adam optimizer, and RMSProp method.

[0056] Stochastic gradient descent (SGD): A basic optimization method that can be combined with momentum and learning rate decay.

[0057] Adam optimizer: An adaptive learning rate optimization method suitable for most deep learning models.

[0058] RMSProp: Another adaptive learning rate optimization method that performs well for non-stationary objective functions.

[0059] Optionally, in the model tuning step, hyperparameters include network structure parameters, training parameters, and temporal parameters. Network structure parameters include the number of layers, the number of neurons per layer, and the kernel size; training parameters include the learning rate, batch size, and regularization coefficient; and temporal parameters include the input window size and stride.

[0060] Among hyperparameter optimization methods, grid search is used to search for all possible combinations of hyperparameters within a specified range. It has a large computational cost but can find the global optimum. Random search is used to randomly sample combinations of hyperparameters within a specified range. It has higher computational efficiency than grid search. Bayesian optimization is a hyperparameter optimization method based on Bayesian theory. It can find a better solution in a smaller number of trials. Genetic algorithms are optimization methods that simulate the natural evolution process and are suitable for complex hyperparameter spaces.

[0061] Furthermore, in the model evaluation step, regression performance metrics include mean squared error, root mean squared error, mean absolute error, mean absolute percentage error, and coefficient of determination.

[0062] The mean squared error (MSE) reflects the average squared error between the true and predicted values; a smaller value is better. Its calculation formula is: ; In the formula, n Indicates sample size. i Indicates the sample order. Y i Indicates sample value, express n The average of the samples.

[0063] The root mean square error (RMSE) is similar to MSE, but its dimension is consistent with the prediction target (meters), making it more intuitive. Its calculation formula is as follows: ; In the formula, N Indicates the number of samples. Indicates the predicted value. x i Represents the actual value.

[0064] Mean Absolute Error (MAE) reflects the average absolute error between the true and predicted values, and is more robust to outliers than Mean Sequence (MSE). Its calculation formula is as follows: ; In the formula, Indicates the first i The true value of each sample Indicates the first i The predicted value for each sample.

[0065] The Mean Absolute Percentage Error (MAPE) reflects the relative magnitude of prediction error, facilitating comparisons of prediction performance across different magnitudes of data. Its calculation formula is as follows: ; In the formula, Indicates the first i The true value of each sample Indicates the first i The predicted value for each sample.

[0066] The coefficient of determination reflects the model's ability to explain data variation, and its value ranges from -∞ to 1. The closer the result is to 1, the better the model fits. Its calculation formula is as follows: 。

[0067] In the model evaluation step, braking safety-related indicators include maximum absolute error, safety margin compliance rate, and error distribution. Maximum absolute error represents the largest absolute error value among all predicted samples, reflecting the model's worst predictive performance; the safety margin compliance rate is the proportion of samples where the predicted braking distance is less than or equal to the actual braking distance, and should be as close to 100% as possible to ensure safety; the error distribution is the probability distribution of the predicted error, and ideally should approximate a normal distribution with a mean of 0.

[0068] In the model evaluation step, statistical significance tests include the t-test and the Wilcoxon signed-rank test. The t-test is used to test whether there is a significant difference between the prediction results of the deep learning model and the prediction results of the traditional model; the Wilcoxon signed-rank test is a non-parametric test method, which is suitable for data that do not meet the normal distribution assumption.

[0069] Optionally, the braking distance prediction model is a hybrid model based on an LSTM-Attention structure.

[0070] Furthermore, braking distance prediction models include: CNN-LSTM hybrid model, LSTM-Attention hybrid model and Transformer-LSTM hybrid model.

[0071] The core of the CNN-LSTM hybrid model is to use CNN (convolutional neural network) to extract local features of the braking signal and then use LSTM (long short-term memory network) to capture temporal dependencies (such as the order of video frames and changes in time-series data), which is suitable for tasks that contain both spatial and temporal information.

[0072] The LSTM-Attention hybrid model is a hybrid model that adds attention to the LSTM (Long Short-Term Memory) network. The core is to enable the model to dynamically focus on the parts of the input sequence that are more relevant to the current task when processing sequential data. This solves the problem that traditional LSTM cannot remember long sequence information and that key information is diluted. It can automatically focus on the features of key time nodes.

[0073] The Transformer-LSTM hybrid model combines the global modeling capabilities of Transformer and the temporal modeling capabilities of LSTM. Its core objective is to complement the advantages of the two models: use LSTM to efficiently capture local temporal coherence and use Transformer to capture long-distance global dependencies, while alleviating the high computational cost of pure Transformer and the limitations of pure LSTM in long sequence modeling.

[0074] Based on the braking distance prediction model obtained from the above algorithm model, given an initial operating speed for the train, a braking distance data table can be obtained, showing the braking distance of each algorithm model as the initial speed gradually decreases from 160 km / h to 0. As shown in Table 1, S DNN This represents the train braking distance predicted by the DNN deep neural network model. S T This represents the train braking distance predicted by the Transformer model. S ResNet This represents the train braking distance predicted by the ResNet recurrent neural network model. S CNN This represents the train braking distance predicted by the CNN convolutional neural network model. S RNN This represents the train braking distance predicted by the RNN recurrent neural network model, with units of meters (m).

[0075] Table 1. Braking Distance Data of Train Braking Intelligent Model

[0076] Figure 3 This data was plotted based on the data in Table 1, with the horizontal axis representing train speed and the vertical axis representing train braking distance. (From Table 1 and...) Figure 3 It can be seen that when the initial braking speed of the train is relatively large, the braking distance predicted by the DNN deep neural network model is the largest, while the braking distance predicted by the CNN convolutional neural network model is the smallest; when the initial braking speed is relatively small, the braking distances of the various algorithm models are similar.

[0077] After the braking distance model is trained in this invention, an initial speed is given to the train based on the trained braking distance prediction model, and an optimized braking distance data table is obtained. The braking distance of the optimized algorithm model can be seen as the initial speed gradually decreases from 160km / h to 0, as shown in Table 2.

[0078] Table 2. Optimized Train Braking Distance Data

[0079] As shown in Table 2, the distance corresponding to the optimized algorithm model decreases as the speed gradually decreases. Furthermore, based on Table 2, [the following data was generated / made / implemented]. Figure 4 .

[0080] Through Table 1 and Table 2, Figure 3 and Figure 4The comparison shows that, under the same initial braking speed, the braking distance prediction model trained in this invention is better at predicting braking distance, with DNN having the largest distance and CNN having the smallest distance.

[0081] Based on the data comparison between the aforementioned braking distance model and the trained braking distance model, this invention describes the braking distance optimization model constructed, which includes a CNN-LSTM hybrid model, an LSTM-Attention hybrid model, and a Transformer-LSTM hybrid model. In the braking distance optimization prediction process, the CNN first extracts spatial features, and then the LSTM learns the temporal dynamics. The Attention mechanism assigns dynamic weights to the LSTM to focus on key braking stages. The Transformer-LSTM balances global dependency and local coherence, significantly reducing computational overhead while maintaining performance. Finally, performance comparison analysis reveals that at an initial braking speed of 160 km / h, the CNN predicts the shortest braking distance, the DNN the longest, and the Transformer and ResNet fall in between. As the initial braking speed decreases, the differences in prediction results among the various models gradually narrow. Considering accuracy, real-time performance, and hardware constraints, CNN and GRU are more suitable for deployment in in-vehicle environments, while the Transformer and its hybrid models can be used for auxiliary optimization on ground-based supercomputing platforms, thus providing multi-scenario and multi-level technical support for intelligent braking decisions in LKJ vehicle control mode.

[0082] Optionally, in S3, the trained braking distance prediction model determines the train's speed-distance curve based on the train braking distance calculation formula obtained through training optimization of the train braking distance calculation formula pulled by the locomotive. The formula for calculating the braking distance of a train pulled by a locomotive is as follows: ; In the formula, S k Indicates the braking distance during idle driving. S e Indicates the effective braking distance. Indicates the initial braking speed of the train. Indicates the train's empty travel time. Indicates the final braking speed. Indicates the braking calculation coefficient. This indicates the converted coefficient of friction for the brake shoes (brake plates). Indicates the converted braking rate of the train. Indicates the commonly used braking coefficient. Indicates the basic resistance per unit of the train. The gradient percentage is added to the braking section; the train's idle travel time indicates the equipment's reaction time after the train driver issues a braking command.

[0083] The optimized formula for calculating the braking distance of a locomotive-hauled train is as follows:

[0084] In the formula, Indicates the distance traveled without moving. S k Divided into multiple intervals n After that, the i The initial braking velocity of each interval, Indicates the first n The initial braking velocity of each interval.

[0085] Table 3 Comparison of Experimental Results for Each Model

[0086] Referring to Table 3, the evaluation results of the optimized braking distance prediction model and the existing traditional model show that the trained braking distance prediction model in this invention is significantly superior to the traditional model: the root mean square error of the LSTM-Attention model is 1.64 meters lower than that of the traditional formula model, its mean absolute error is 4.17 meters lower, and its coefficient of determination is 13.4 higher. Safety margin achievement rate: The safety margin achievement rate of the deep learning model is 95.7%, significantly higher than the 82.3% of the traditional model, indicating that the deep learning model improves the safety performance during braking. Error distribution analysis: The error distribution range of the deep learning model is more concentrated, and the standard deviation is generally smaller, indicating that the prediction results are more stable and reliable.

[0087] The present invention also discloses an LKJ braking distance calculation optimization system based on a deep learning model, used to implement the LKJ braking distance calculation optimization method based on a deep learning model as described in any of the above embodiments, including a data acquisition module, a distance determination module, a curve determination module and a braking determination module connected in sequence.

[0088] The data acquisition module is used to collect the braking parameters, initial braking speed, and track data of the target train.

[0089] The distance determination module is used to determine the target distance based on the track data; the target distance represents the distance between the target train's current position on the track and the braking target point.

[0090] The curve determination module is used to input braking parameters and target distance into the trained braking distance prediction model to determine the speed-distance curve of the target train.

[0091] The braking determination module is used to determine the braking distance of the target train based on the initial braking speed and the speed-distance curve.

[0092] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing LKJ braking distance calculation based on a deep learning model, characterized in that, include: S1 Data Acquisition Steps: Acquire the braking parameters, initial braking speed, and track data of the target train's current location on the track. S2 Distance Determination Steps: Determine the target distance based on the track data; the target distance represents the distance between the target train's current position on the track and the braking target point; S3 curve determination steps: Input the braking parameters and target distance into the trained braking distance prediction model to determine the speed-distance curve of the target train; S4 Braking Determination Steps: Based on the initial braking speed, determine the braking distance of the target train using the speed-distance curve.

2. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 1, characterized in that, The braking parameters of the target train in S1 include: the gradient of the braking section (in thousands), the train's braking pressure reduction, the train's converted braking rate, the braking calculation coefficient, the converted friction coefficient, the train's basic unit resistance, and the number of vehicles.

3. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 1, characterized in that, The training process of the braking distance prediction model in S3 includes the following steps: Processing steps: Acquire train data, preprocess the train data and divide it into sets according to a preset ratio to obtain training set, validation set and test set; Model initialization steps: Initialize the braking distance prediction model based on train data to obtain the current braking distance prediction model; Forward propagation steps: Input the training set into the current braking distance prediction model for training. After forward propagation calculation, the training prediction set is obtained. Backpropagation steps: Determine the loss function value of the current braking distance prediction model based on the training prediction set, and perform backpropagation calculation based on the loss function value of the current braking distance prediction model to obtain the adjustment gradient; Model adjustment steps: Based on the adjusted gradient and validation set, perform hyperparameter optimization on the current braking distance prediction model to obtain the adjusted braking distance prediction model; Model evaluation steps: Evaluate the performance of the adjusted braking distance prediction model based on the test set. If it meets the performance requirements, the adjusted braking distance prediction model is determined as the trained braking distance prediction model; if it does not meet the performance requirements, return to the forward propagation step.

4. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 3, characterized in that, Train data includes time-series characteristic data and static characteristic data.

5. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 3, characterized in that, The hyperparameter optimization in the model tuning process specifically includes: grid search optimization, random search optimization, Bayesian optimization, and genetic algorithm optimization.

6. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 3, characterized in that, The performance evaluation of the model evaluation steps specifically includes: regression performance indicators, braking safety indicators, and statistical significance test indicators.

7. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 1, characterized in that, The braking distance prediction model in S3 is a hybrid model based on the LSTM-Attention structure.

8. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 7, characterized in that, Braking distance prediction models include: CNN-LSTM hybrid model, LSTM-Attention hybrid model and Transformer-LSTM hybrid model.

9. The method for optimizing LKJ braking distance calculation based on a deep learning model according to claim 1, characterized in that, In S3, the trained braking distance prediction model determines the speed-distance curve of the train based on the braking distance calculation formula of the train pulled by the locomotive and the optimized braking distance calculation formula of the train pulled by the locomotive obtained through training optimization. The formula for calculating the braking distance of a train pulled by a locomotive is as follows: ； In the formula, S k Indicates the braking distance during idle driving. S e Indicates the effective braking distance. Indicates the initial braking speed of the train. Indicates the train's empty travel time. Indicates the final braking speed. Indicates the braking calculation coefficient. This indicates the converted friction coefficient of the brake shoe. Indicates the converted braking rate of the train. Indicates the commonly used braking coefficient. Indicates the basic resistance per unit of the train. The gradient percentage is added to the braking section; the train's idle travel time indicates the equipment's reaction time after the train driver issues a braking command. The optimized formula for calculating the braking distance of a locomotive-hauled train is as follows: In the formula, Indicates the distance traveled without moving. S k Divided into multiple intervals n After that, the i The initial braking velocity of each interval, Indicates the first n The initial braking velocity of each interval.

10. A deep learning model-based optimization system for calculating LKJ braking distance, characterized in that, The method for optimizing LKJ braking distance calculation based on a deep learning model as described in any one of claims 1-9 includes a data acquisition module, a distance determination module, a curve determination module, and a braking determination module connected in sequence. The data acquisition module is used to collect the braking parameters, initial braking speed, and track data of the target train's current location. The distance determination module is used to determine the target distance based on the track data; the target distance represents the distance between the target train's current position on the track and the braking target point. The curve determination module is used to input braking parameters and target distance into a trained braking distance prediction model to determine the speed-distance curve of the target train. The braking determination module is used to determine the braking distance of the target train based on the initial braking speed and using the speed-distance curve.

Images