Locomotive wheel rail adhesion identification and traction motor cooperative control method and system

Abstract

The invention relates to the field of traffic locomotives, in particular to a locomotive wheel rail adhesion identification and traction motor cooperative control method and system. The method comprises the following steps: acquiring wheel rail adhesion data; preprocessing the wheel rail adhesion data to generate an input data sequence; building and training an adhesion state prediction module based on a long and short term memory neural network, inputting the input data sequence into the trained adhesion state prediction module, and outputting an adhesion coefficient prediction value; constructing a torque optimization calculation module, inputting the adhesion coefficient predicted value into the torque optimization calculation module, and calculating to obtain an optimized traction motor torque; and the optimized traction motor torque is transmitted to a locomotive motor control module, and the locomotive motor control module drives a motor to adjust the output motor shaft torque based on the optimized traction motor torque, so that the wheel track is in the optimal adhesion state. Accurate control over the locomotive is achieved, the phenomena of wheel slipping and idling are effectively reduced, and energy consumption and wheel track abrasion are reduced.

Description

Technical Field

[0001] The present invention belongs to the field of transportation locomotives, and in particular relates to a locomotive wheel-rail adhesion identification and traction motor coordinated control method and system. Background Art

[0002] Locomotives are widely used for heavy-load transportation and play a key role in mining, industry, and freight logistics. Locomotives provide traction and braking force through the adhesion between their wheels and rails. Adhesion directly determines the locomotive's operating efficiency and safety. Insufficient adhesion between the wheels and rails can lead to wheel slippage, resulting in energy waste, accelerated wear of wheel-rail components, and operational safety risks. Therefore, accurately identifying the real-time adhesion status between the wheels and rails is crucial.

[0003] Currently, common locomotive adhesion control technologies include the following: anti-slip control based on a fixed threshold; control based on a proportional-integral (PI) controller; and control based on a state observer. Although these methods have been applied in practical engineering, they all have significant limitations and are difficult to adapt to complex and changing actual operating conditions, limiting further improvements in locomotive control performance.

[0004] Existing threshold anti-skid control technology identifies the adhesion state by setting fixed threshold parameters such as the speed difference between the wheel and the vehicle or the wheelset acceleration. When the wheel speed difference or wheelset acceleration exceeds the predetermined threshold, the control strategy is triggered to adjust the traction or braking force to achieve vehicle re-adhesion. This method is relatively simple, has a small amount of calculation, and is easy to implement in an embedded system. However, the use of fixed threshold strategies has significant limitations: (1) Poor robustness and insufficient adaptability to sudden loads or changes in road conditions. (2) Excessive sensitivity can easily cause false operations, repeatedly triggering re-adhesion control, and reducing locomotive operating stability and traction performance. (3) The threshold requires a lot of on-site debugging and determination for fixed lines, and the experimental process is time-consuming.

[0005] Existing PI control strategies use a proportional-integral approach to control the locomotive wheel set adhesion coefficient to maintain optimal adhesion. This approach typically adjusts the traction or braking force based on the error to stabilize the slip ratio and adhesion. However, fixed gains are difficult to adapt to the significant nonlinearity and dynamic changes in the wheel-rail interface. In particular, adaptive performance is weak when the vehicle operating conditions change, resulting in reduced control performance.

[0006] Model approaches based on full-state observers use state-space observers to capture state variables that are difficult to measure directly. These approaches are widely used in urban rail transit to find and maintain optimal adhesion. However, these methods rely on accurate mechanical parameter models. Once actual parameters or environmental conditions change, estimation errors increase, resulting in a decrease in robustness.

[0007] In addition to the long short-term memory network, other neural networks (such as feedforward neural networks) also have certain predictive capabilities, but they perform poorly in capturing the dynamic correlation between time series data. Therefore, it is difficult to efficiently achieve the various technical purposes of the present invention. At present, the long short-term memory network is the best solution. The purpose of the present invention is to overcome the problems of insufficient robustness of the existing adhesion state estimation algorithm and poor adaptability to changing working conditions, and to propose a locomotive wheel-rail adhesion identification and traction motor coordinated control method and system. The present invention is based on the LSTM (long short-term memory) neural network to achieve more accurate adhesion state prediction, and further guide the locomotive multi-motor coordinated optimization of traction control to improve the locomotive operation stability and efficiency. Summary of the Invention

[0008] To address the aforementioned technical issues of existing adhesion control methods, such as insufficient robustness, poor adaptability to operating conditions, and reliance on fixed parameters or precise models, which leads to reduced control performance, a method and system for identifying locomotive wheel-rail adhesion and coordinating traction motor control is provided. This invention primarily leverages the dynamic feature extraction capabilities of a long short-term memory (LSTM) neural network for multi-source time-series sensor data, combined with a multi-objective rolling optimization algorithm and an embedded real-time collaborative control architecture. This achieves the technical benefits of dynamic, high-precision estimation of the adhesion coefficient, effectively reducing wheel slip and idling, and lowering energy consumption and wheel-rail wear.

[0009] The technical means adopted in the present invention are as follows:

[0010] A method for identifying locomotive wheel-rail adhesion and coordinating control of a traction motor comprises the following steps:

[0011] Collecting wheel-rail adhesion data, wherein the wheel-rail adhesion data includes wheelset speed, locomotive running speed, wheelset slip data, locomotive torque, locomotive acceleration and environmental parameters;

[0012] preprocessing the wheel-rail adhesion data to generate an input data sequence;

[0013] Building and training an adhesion state prediction module based on a long short-term memory neural network, inputting the input data sequence into the trained adhesion state prediction module, and outputting a predicted value of the adhesion coefficient;

[0014] Constructing a torque optimization calculation module, inputting the adhesion coefficient prediction value into the torque optimization calculation module, calculating an optimized traction motor torque, and generating an optimized torque command based on the optimized traction motor torque;

[0015] The optimized torque instruction is transmitted to the locomotive motor control module, and the locomotive motor control module drives the motor to adjust the output motor shaft torque based on the optimized traction motor torque to achieve the best wheel-rail adhesion state.

[0016] Furthermore, the pre-processing of the wheel-rail adhesion data to generate an input data sequence includes:

[0017] performing filtering processing on the adhesion data, and normalizing the filtered adhesion data to generate normalized wheel-rail adhesion data;

[0018] The normalized wheel-rail adhesion data are combined to generate an input data sequence.

[0019] Furthermore, the adhesion state prediction module includes an input layer, a long short-term memory network unit layer, a random inactivation layer and an output layer connected in sequence.

[0020] The workflow of the adhesion state prediction module is as follows:

[0021] The input layer receives the input data sequence;

[0022] Inputting the input data sequence into a long short-term memory network unit layer, wherein the long short-term memory network unit layer captures hidden state features in the input data sequence based on the input data series;

[0023] Inputting the hidden state features into a random deactivation layer to generate hidden state features processed by the random deactivation layer;

[0024] The hidden state features processed by the random inactivation layer are input to the output layer, and the adhesion coefficient prediction value is output.

[0025] Furthermore, the workflow of the torque optimization calculation module is as follows:

[0026] evaluating the working condition of the wheel-rail adhesion state according to the predicted value of the adhesion coefficient;

[0027] Establishing a target optimization function with the objectives of maximizing the adhesion coefficient, minimizing the slip difference, maximizing the vehicle traction force, and minimizing the locomotive torque change, inputting the locomotive operating speed, the predicted adhesion coefficient value, and the locomotive torque into the target optimization function, and calculating and generating the optimized traction motor torque;

[0028] An optimized torque command is generated according to the optimized traction motor torque.

[0029] Furthermore, the method further comprises:

[0030] The online diagnosis module automatically diagnoses and analyzes the current wheel-rail operating state and prediction error based on the wheel-rail adhesion data, and regularly evaluates the performance of the adhesion state prediction module based on the current wheel-rail operating state and prediction error;

[0031] If performance degradation is found, the adaptive parameter update mechanism is triggered to train the adhesion state prediction module using the wheel-rail adhesion data and update the network parameters to optimize long-term performance.

[0032] A locomotive wheel-rail adhesion identification and traction motor coordinated control system, used to implement the locomotive wheel-rail adhesion identification and traction motor coordinated control method, comprising:

[0033] A data acquisition module is used to collect wheel-rail adhesion data, wherein the wheel-rail adhesion data includes wheelset speed, locomotive running speed, wheelset slip data, locomotive torque, locomotive acceleration and environmental parameters;

[0034] a data preprocessing module, connected to the data acquisition module, for filtering and normalizing the acquired wheel-rail adhesion data and combining them to generate an input data sequence;

[0035] an adhesion state prediction module, connected to the data preprocessing module, comprising an input layer, a long short-term memory network unit layer, a random inactivation layer, and an output layer connected in sequence, for obtaining a current adhesion coefficient prediction value based on the input data sequence;

[0036] a torque optimization calculation module connected to the adhesion state prediction module, configured to calculate an optimized traction motor torque based on the adhesion coefficient prediction value, with the optimization objectives of maximizing the adhesion coefficient, minimizing the slip difference, maximizing the vehicle traction force, and minimizing the torque variation, and generate an optimized torque command based on the optimized traction motor torque;

[0037] The motor control module is connected to the torque optimization calculation module and is used to drive the motor to adjust the output motor shaft torque based on the optimized torque instruction.

[0038] Furthermore, the system further comprises:

[0039] The online diagnosis module is connected to the adhesion state prediction module and the data acquisition module, and is used to monitor the prediction error and trigger the update of model parameters when the performance is degraded.

[0040] Furthermore, the system adopts a real-time operating system embedded platform to deploy the adhesion coefficient prediction module and the torque optimization module.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1. The present invention combines a real-time prediction method with the time series modeling capability of a long short-term memory neural network (LSTM) to achieve dynamic and high-precision estimation of the adhesion coefficient. The real-time prediction accuracy and robustness of this method are superior to those of the traditional threshold method and proportional-integral control method.

[0043] 2. The traction motor coordinated control strategy provided by the present invention, by combining with a multi-objective optimization algorithm (such as minimizing slip rate and maximizing traction), can achieve precise control of the locomotive, effectively reduce wheel slip and idling, and reduce energy consumption and wheel-rail wear.

[0044] 3. The dynamic adaptive control system provided by this invention, combined with online diagnostics and a self-updating model parameter mechanism, achieves stable control in complex operating environments. Dynamic control can significantly improve the stability and safety of actual operating locomotives, allowing them to better adapt to complex and ever-changing operating environments.

[0045] Based on the above reasons, the present invention can be widely promoted in the fields of transportation locomotives and the like. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0047] Figure 1 This is a structural block diagram of the locomotive wheel-rail adhesion identification and traction motor coordinated control method of the present invention.

[0048] Figure 2 This is a structural diagram of the adhesion state prediction module based on the long short-term memory (LSTM) neural network of the present invention.

[0049] Figure 3 This is a calculation flow chart of the dynamic collaborative torque optimization control strategy of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0051] like Figure 1-3 As shown, the present invention provides a method for identifying locomotive wheel-rail adhesion and coordinating control of a traction motor, the specific steps of which are as follows:

[0052] S1. Collect wheel-rail adhesion data, which includes wheelset speed, locomotive running speed, wheelset slip data, locomotive torque, locomotive acceleration and environmental parameters.

[0053] Among them, wheelset speed refers to the speed signal of each driving wheelset (i.e., the real-time speed of wheel rotation), locomotive running speed refers to the longitudinal driving speed, wheelset slip data refers to the slip between the wheel and the rail, locomotive torque refers to the traction or braking torque currently applied by the locomotive, locomotive acceleration refers to the longitudinal acceleration signal of the vehicle body, and environmental parameters refer to sensor signals related to the weather environment (dry, humid, etc.) (which can be collected and supplemented according to specific conditions).

[0054] Specifically, the locomotive in this embodiment has a total of six drive wheel pairs, each driven by an independent traction motor. A speed sensor is installed on each drive wheel set to collect the angular velocity of the wheel rotation in real time. A speed radar system is installed at the locomotive chassis to collect the actual longitudinal driving speed of the locomotive as a whole in real time and obtain real-time slip data for each wheel axle. In addition, each traction motor and its converter unit can detect the current output torque in real time, and a three-dimensional inertial measurement sensor is installed near each axle device to measure the locomotive's longitudinal acceleration in real time. The real-time acquisition frequency is 100Hz (i.e., data is recorded every 0.01 seconds), and the real-time data is transmitted to the on-board control unit (TCU) via the CAN (Controller Area Network) bus or a dedicated data bus.

[0055] S2. Preprocess the wheel-rail adhesion data to generate an input data sequence.

[0056] Specifically, the adhesion data is filtered to remove noise interference from the sensor signal. The filtered wheel-rail adhesion data is normalized and converted to values ​​within a standard range, generating normalized wheel-rail adhesion data. Multidimensional input data sequences with a fixed time step (e.g., 0.1 seconds, i.e., 10 Hz sampling) are continuously combined into a multidimensional input data sequence with a fixed step length (e.g., 10 consecutive sampling points, i.e., 1 second of data) to generate optimized torque commands suitable for long-short-term memory neural networks.

[0057] All collected real-time data is filtered by a real-time on-board controller unit to eliminate system noise that occurs in high-speed collected data. A Min-Max (minimum-maximum) normalization algorithm is used to normalize each input variable to the range [0,1] to improve the accuracy of long-short-term memory network model training and on-site operations.

[0058] The formula involved in the specific normalization method is:

[0059] X_norm=(X-X_min) / (X_max-X_min)

[0060] Among them, X_min and X_max are the upper and lower limits of the wheel-rail adhesion data after filtering, which are used to standardize the actual working condition data. X_norm is the normalized wheel-rail adhesion data, and X is the filtered wheel-rail adhesion data.

[0061] S3. Build and train an adhesion state prediction module based on a long short-term memory neural network, input the input data sequence into the trained adhesion state prediction module, and output the adhesion coefficient prediction value.

[0062] Specifically, the workflow of the adhesion state prediction module is as follows:

[0063] First, the input layer receives the input data sequence.

[0064] Secondly, the input data sequence is input into the long short-term memory network unit layer. The long short-term memory network unit layer captures the hidden state features in the input data sequence based on the input data series. The hidden state features include dynamic (time series) relationships and nonlinear features.

[0065] Subsequently, the hidden state features are input into the dropout layer to generate hidden state features processed by the dropout layer, which is used to prevent overfitting of the neural network during training and improve the generalization performance of the network.

[0066] Finally, the hidden state features processed by the random inactivation layer are input into the output layer. The output layer uses a fully connected linear layer to output the predicted value of the adhesion coefficient.

[0067] Based on the historical on-site operation data of the locomotive, a training sample library is established, including a time series set of continuous operation data under different working conditions (different loads, different climatic conditions, and different rail surface conditions). A long short-term memory network layer structure is built with a random inactivation ratio of 0.2 to prevent overfitting during training. Finally, the wheel-rail adhesion coefficient estimation result is obtained through a fully connected output layer. During the training process, the historical collected data set is divided into a training set, a validation set, and a test set with a ratio of 7:2:1. The optimizer uses the Adam algorithm, and the loss function is the mean square error loss function (MSE). The system optimizes the optimal network structure and network weight parameters through offline training to ensure that the network can achieve good online prediction accuracy.

[0068] The trained LSTM network model parameters are imported into the locomotive's on-site real-time control unit (CCU). The network model outputs a predicted value for the current wheel-rail adhesion coefficient at a refresh rate of 0.1 seconds (10Hz). This predicted value is displayed in real time on the on-site control unit's data display interface, allowing locomotive drivers and monitoring personnel to monitor the locomotive's wheel-rail adhesion status in real time.

[0069] S4. Construct a torque optimization calculation module, input the adhesion coefficient prediction value into the torque optimization calculation module, and calculate the optimized traction motor torque.

[0070] Specifically, the workflow of the torque optimization calculation module is as follows:

[0071] The real-time control unit uses the predicted adhesion coefficient as a feedback parameter based on the predicted adhesion coefficient. It evaluates the working conditions of the wheel-rail adhesion state (e.g., normal, slight slip, or severe slip). It establishes a target optimization function with the goals of maximizing the adhesion coefficient, minimizing the slip difference, maximizing the vehicle traction, and minimizing the locomotive torque change. Based on the states of each axle, the locomotive operating speed, the predicted adhesion coefficient, and the locomotive torque are input into the target optimization function to calculate and generate the optimized traction motor torque. The states of each axle specifically refer to whether the motor is in traction, braking, or driving state. Based on the optimized traction motor torque, an optimized torque command is generated.

[0072] The real-time control unit determines whether the current operating state is normal, safe, or slipping based on the predicted value of the adhesion coefficient, and constructs the following multi-objective real-time optimization control model:

[0073] The optimization objective function uses a weighted combination strategy of maximizing adhesion, minimizing slip, maximizing torque, and minimizing torque variation. The weighting factors can be adjusted offline or online based on the characteristics of different operating conditions. Typical weighting factors include: 0.3 for adhesion, 0.2 for slip, 0.4 for traction torque, and 0.1 for torque variation.

[0074] The weighted optimization objective function is constructed as follows:

[0075] J=,w -1 .·μ-w -2 .·Δs+,w -3 .·T-,w-4.ΔT

[0076] Among them, w -1 is the adhesion coefficient weight factor, w -2 is the slip difference weight factor, w -3 is the traction torque weight factor, w -4is the torque change weight factor, μ is the wheel-rail adhesion coefficient, Δs is the slip difference, T is the traction torque, and ΔT is the torque change difference.

[0077] Real-time optimization uses a sliding time window method to determine the optimal torque distribution scheme between traction motors every 0.1 seconds and output it to the control unit of each axle traction motor in real time.

[0078] For example, when the network identifies a slight slipping trend on the 5th and 6th axes in real time, the control algorithm will actively reduce the torque output of the 5th and 6th axis motors and increase the output torque of other non-slipping axes (such as the 1st, 2nd, 3rd, and 4th axes) to achieve rapid recovery of the slipping axles and maintain the traction of the entire vehicle.

[0079] To ensure real-time performance and control response, real-time rolling optimization technology is used to continuously update and optimize the torque output distribution plan of each motor in real time. The output plan is recalculated and optimized every certain sampling period (such as 0.1 second) to adapt to different adhesion states and load conditions.

[0080] S5. The optimized traction motor torque is transmitted to the locomotive motor control module. The locomotive motor control module drives the motor to adjust the output motor shaft torque based on the optimized traction motor torque. The drive motors work in coordination to achieve optimal wheel-rail adhesion.

[0081] Specifically, the optimized torque command obtained in step S4 is transmitted via the MVB bus. Each traction motor control unit directly receives and executes the command, achieving coordinated torque distribution among the different axles. Closed-loop motor vector control ensures stable locomotive operation, reduces the number and intensity of slips, reduces wheel wear, and improves overall operating efficiency. Field tests have shown that this method maintains overall stable wheel-rail friction, with no significant repetitive slipping or oscillations, and significantly reduces wheel-rail wear.

[0082] S6. The online diagnosis module predicts the error and triggers the adaptive parameter update mechanism to optimize the long-term performance.

[0083] Specifically, real-time data from the field is continuously collected into the backend system. The online diagnostic module automatically analyzes the current wheel-rail operating status and prediction error based on the wheel-rail adhesion data, and periodically evaluates model performance. Based on the current wheel-rail operating status and prediction error, the adhesion prediction module's performance is regularly evaluated. If significant performance degradation is detected, an adaptive parameter update mechanism is triggered. Recent real-time wheel-rail adhesion data is used to train the adhesion prediction module and update network parameters. This enables online self-learning and automatically improves long-term prediction performance and accuracy to adapt to operating conditions caused by long-term changes in the track surface.

[0084] When the locomotive operating conditions change significantly (such as changes in seasonal characteristics of the track surface and changes in the long-term operating parameters of the locomotive), the operating condition data will be automatically uploaded regularly through the background, and the existing network training library and model parameters will be updated online, which can realize the online adaptive update of the network model and maintain the long-term accuracy and adaptability of the system.

[0085] The present invention also includes a locomotive wheel-rail adhesion identification and traction motor coordinated control system, which is used to implement the above-mentioned locomotive wheel-rail adhesion identification and traction motor coordinated control method. The system specifically includes:

[0086] The data acquisition module is used to collect wheel-rail adhesion data, which includes wheelset speed, locomotive running speed, wheelset slip data, locomotive torque, locomotive acceleration and environmental parameters.

[0087] The data preprocessing module is connected to the data acquisition module and is used to filter and normalize the collected wheel-rail adhesion data and combine them to generate an input data sequence.

[0088] The adhesion state prediction module is connected to the data preprocessing module and includes an input layer, a long short-term memory network unit layer, a random inactivation layer and an output layer connected in sequence, and is used to predict the current adhesion coefficient based on the input data sequence.

[0089] The torque optimization calculation module is connected to the adhesion state prediction module and is used to calculate the optimized traction motor torque based on the predicted value of the adhesion coefficient, with the optimization goals of maximizing the adhesion coefficient, minimizing the slip difference, maximizing the vehicle traction force and minimizing the torque change, and generate an optimized torque command based on the optimized traction motor torque.

[0090] The motor control module is connected to the torque optimization calculation module and is used to drive the motor to adjust the output motor shaft torque based on the optimized torque instruction.

[0091] The online diagnosis module is connected to the adhesion state prediction module and the data acquisition module to monitor the prediction error and trigger the model parameter update when the performance degrades.

[0092] The software system utilizes an embedded real-time operating system (RTOS) platform. Within the structured software flow, a separate module for real-time estimation of the adhesion coefficient using a long-short-term memory network is incorporated. The adhesion coefficient estimation module rapidly identifies real-time data collected and processed on-site, and the results are then transmitted to the torque optimization calculation module. After calculations are completed, control instructions are directly passed to the motor control (traction converter) unit. All steps are performed in real time, achieving a closed-loop coordinated control effect with real-time identification of the adhesion coefficient, real-time updating of the motor control strategy, and real-time dynamic response.

[0093] This specific embodiment, through real-time wheel-rail adhesion coefficient prediction based on a long short-term memory network and combined with real-time collaborative optimization control of multiple motors, effectively ensures the safe and efficient operation of the locomotive, improves the traction stability of the locomotive, and reduces power consumption and equipment wear, with obvious practical engineering application value.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A locomotive wheel-rail adhesion identification and traction motor coordinated control method, characterized in that: The following steps are involved: Collecting wheel-rail adhesion data, wherein the wheel-rail adhesion data includes wheelset speed, locomotive running speed, wheelset slip data, locomotive torque, locomotive acceleration and environmental parameters; preprocessing the wheel-rail adhesion data to generate an input data sequence; Building and training an adhesion state prediction module based on a long short-term memory neural network, inputting the input data sequence into the trained adhesion state prediction module, and outputting a predicted value of the adhesion coefficient; Constructing a torque optimization calculation module, inputting the adhesion coefficient prediction value into the torque optimization calculation module, calculating an optimized traction motor torque, and generating an optimized torque command based on the optimized traction motor torque; The optimized torque instruction is transmitted to the locomotive motor control module, and the locomotive motor control module drives the motor to adjust the output motor shaft torque based on the optimized traction motor torque to achieve the best wheel-rail adhesion state.

2. The locomotive wheel-rail adhesion identification and traction motor coordinated control method according to claim 1, characterized in that: The pre-processing of the wheel-rail adhesion data to generate an input data sequence comprises: performing filtering processing on the adhesion data, and normalizing the filtered adhesion data to generate normalized wheel-rail adhesion data; The normalized wheel-rail adhesion data are combined to generate an input data sequence.

3. The locomotive wheel-rail adhesion identification and traction motor coordinated control method according to claim 1, characterized in that: The adhesion state prediction module includes an input layer, a long short-term memory network unit layer, a random inactivation layer and an output layer connected in sequence. The workflow of the adhesion state prediction module is as follows: The input layer receives the input data sequence; Inputting the input data sequence into a long short-term memory network unit layer, wherein the long short-term memory network unit layer captures hidden state features in the input data sequence based on the input data series; Inputting the hidden state features into a random deactivation layer to generate hidden state features processed by the random deactivation layer; The hidden state features processed by the random inactivation layer are input to the output layer, and the adhesion coefficient prediction value is output.

4. The locomotive wheel-rail adhesion identification and traction motor coordinated control method according to claim 1, characterized in that: The workflow of the torque optimization calculation module is as follows: evaluating the working condition of the wheel-rail adhesion state according to the predicted value of the adhesion coefficient; Establishing a target optimization function with the objectives of maximizing the adhesion coefficient, minimizing the slip difference, maximizing the vehicle traction force, and minimizing the locomotive torque change, inputting the locomotive operating speed, the predicted adhesion coefficient value, and the locomotive torque into the target optimization function, and calculating and generating the optimized traction motor torque; An optimized torque command is generated according to the optimized traction motor torque.

5. The locomotive wheel-rail adhesion identification and traction motor coordinated control method according to claim 1, characterized in that: The method further comprises: The online diagnosis module automatically diagnoses and analyzes the current wheel-rail operating state and prediction error based on the wheel-rail adhesion data, and regularly evaluates the performance of the adhesion state prediction module based on the current wheel-rail operating state and prediction error; If performance degradation is found, the adaptive parameter update mechanism is triggered to train the adhesion state prediction module using the wheel-rail adhesion data and update the network parameters to optimize long-term performance.

6. A locomotive wheel-rail adhesion identification and traction motor coordinated control system, used to implement the locomotive wheel-rail adhesion identification and traction motor coordinated control method according to any one of claims 1 to 5, characterized in that: include: A data acquisition module is used to collect wheel-rail adhesion data, wherein the wheel-rail adhesion data includes wheelset speed, locomotive running speed, wheelset slip data, locomotive torque, locomotive acceleration and environmental parameters; a data preprocessing module, connected to the data acquisition module, for filtering and normalizing the acquired wheel-rail adhesion data and combining them to generate an input data sequence; an adhesion state prediction module, connected to the data preprocessing module, comprising an input layer, a long short-term memory network unit layer, a random inactivation layer, and an output layer connected in sequence, for obtaining a current adhesion coefficient prediction value based on the input data sequence; a torque optimization calculation module connected to the adhesion state prediction module, configured to calculate an optimized traction motor torque based on the adhesion coefficient prediction value, with the optimization objectives of maximizing the adhesion coefficient, minimizing the slip difference, maximizing the vehicle traction force, and minimizing the torque variation, and generate an optimized torque command based on the optimized traction motor torque; The motor control module is connected to the torque optimization calculation module and is used to drive the motor to adjust the output motor shaft torque based on the optimized torque instruction.

7. The locomotive wheel-rail adhesion identification and traction motor coordinated control system according to claim 6, characterized in that: The system further comprises: The online diagnosis module is connected to the adhesion state prediction module and the data acquisition module, and is used to monitor the prediction error and trigger the update of model parameters when the performance is degraded.

8. The locomotive wheel-rail adhesion identification and traction motor coordinated control system according to claim 6, characterized in that: The system adopts a real-time operating system embedded platform to deploy an adhesion coefficient prediction module and a torque optimization module.

Images