Pantograph current collection quality closed-loop optimization method and system

Abstract

The application discloses a pantograph current collection quality closed-loop optimization method and system, and the method comprises the following steps: according to a preset dilatation convolution chain structure network, multi-scale spatial feature extraction is performed on a multi-modal data set through a combination of convolution layers with different dilatation coefficients to obtain spatial features; the spatial features and a multi-modal time sequence set are input into a preset space-time convolution attention network, and a space-time fusion feature with space-time fusion is output through a spatial feature branch, a time sequence feature branch, a cross-modal attention mechanism and a dynamic memory mechanism in the space-time convolution attention network; a contact quality index of the pantograph is determined; the space-time fusion feature and the contact quality index are input into a preset digital twin model, and a real-time optimization control instruction is generated by using a gradient descent method and a PID closed-loop control algorithm. Real-time optimization of the contact quality index and dynamic generation of the control instruction are realized, and the system can quickly respond to external changes and maintain the stability and reliability of the current collection process.

Description

Technical Field

[0001] This invention belongs to the field of rail transit optimization technology, and particularly relates to a closed-loop optimization method and system for pantograph current collection quality. Background Technology

[0002] As a key energy transmission component between electric locomotives and the overhead contact line, the pantograph's current collection quality directly affects the safety and reliability of train operation. Therefore, there is a need to develop intelligent optimization systems and methods for pantograph current collection quality. While deep learning algorithms are widely used in industrial monitoring, their application in pantograph systems still faces some bottlenecks.

[0003] However, existing deep learning-based closed-loop optimization methods for pantograph current collection quality have significant technical limitations in practical applications. The pantograph current collection process exhibits a coupling characteristic of multi-scale spatial features and long-term dynamic evolution. However, existing methods mostly employ ordinary convolutional neural networks (CNNs) or recurrent neural networks (RNNs): ordinary CNNs have fixed receptive fields, making it difficult to simultaneously and accurately capture the multi-scale spatial differences between microscopic local and macroscopic overall dimensions; traditional RNNs (such as LSTM and GRU) are prone to gradient vanishing or temporal information loss when modeling long-term temporal correlations such as contact force abrupt changes and vibration transmission, resulting in incomplete and inaccurate extraction of the spatiotemporal features of the current collection process. Furthermore, pantograph current collection involves multi-modal data from mechanics, dynamics, and thermodynamics, and the contribution of different modes to current collection quality varies dynamically under different train operating speeds and catenary spans. However, existing cross-modal attention mechanisms often employ simple feature concatenation and fully connected layer fusion, failing to dynamically adjust the attention weights of each mode according to the real-time scenario. This results in fused features that cannot accurately reflect the key factors affecting current collection quality under different scenarios.

[0004] Furthermore, pantograph current collection is a dynamic and time-varying process. Factors such as changes in the catenary structure and fluctuations in train speed continuously alter the current collection state. However, existing dynamic memory mechanisms (such as ordinary gated loop units) struggle to efficiently retain long-term memories of historically optimal current collection states. Moreover, the logic for updating the interaction between the current current collection state and historical memories is poorly designed. This makes control command updates susceptible to transient interference, lacking the continuity and stability required for long-term optimization. Some methods rely on digital twin simulations to generate training data. However, the simulation models simplify the modeling of complex real-world factors such as catenary elastic nonlinearity, slip plate wear, and external wind disturbances. Furthermore, they lack a transfer learning mechanism from the simulation model to the actual pantograph-catenary system, ultimately resulting in poor generalization ability of deep learning models in practical engineering scenarios and significant deviations between the optimization results and the actual current collection quality requirements.

[0005] The aforementioned technical limitations make it difficult for existing methods to accurately characterize the multi-scale spatiotemporal coupling characteristics and multimodal dynamic correlation laws of pantograph current collection. The accuracy, real-time performance, and robustness of current collection quality optimization are insufficient, and it is impossible to continuously and stably ensure the good current collection status of the pantograph and the overhead contact line under complex and ever-changing operating scenarios. Summary of the Invention

[0006] This invention provides a closed-loop optimization method and system for pantograph current collection quality, which addresses the technical problems of existing deep learning-based pantograph current collection quality closed-loop optimization methods, which suffer from incomplete and inaccurate multi-scale spatiotemporal feature extraction, lack of specificity in cross-modal fusion, insufficient continuity and stability of control command updates, and poor generalization ability in real-world scenarios. These problems result in insufficient accuracy, real-time performance, and robustness in current collection quality optimization, making it impossible to continuously and stably ensure good current collection status of the pantograph and overhead contact line under complex and variable operating scenarios.

[0007] In a first aspect, the present invention provides a closed-loop optimization method for pantograph current collection quality, comprising:

[0008] A multimodal dataset and a multimodal time series set of pantograph data are obtained. The multimodal dataset contains pressure data, temperature data, and strain data of the pantograph data. The multimodal time series set contains time series of the pressure data, time series of the temperature data, and time series of the strain data.

[0009] Based on a pre-defined dilated convolutional chain structure network, spatial features are extracted from the multimodal dataset using a combination of convolutional layers with different dilation coefficients to obtain spatial features.

[0010] The spatial features and the multimodal time series set are input into a preset spatiotemporal convolutional attention network. The spatiotemporal fusion features that fuse space and time are output through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network.

[0011] The contact quality index of the pantograph is determined, including pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters.

[0012] The spatiotemporal fusion features and the contact quality index are input into a preset digital twin model. Gradient descent and PID closed-loop control algorithms are used to generate real-time optimization control commands, which are then sent to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph.

[0013] Secondly, the present invention provides a closed-loop optimization system for pantograph current collection quality, comprising:

[0014] The acquisition module is configured to acquire a multimodal dataset and a multimodal time series set of the pantograph. The multimodal dataset includes pressure data, temperature data, and strain data of the pantograph, and the multimodal time series set includes time series of the pressure data, time series of the temperature data, and time series of the strain data.

[0015] The first output module is configured to extract spatial features from the multimodal dataset by using a pre-defined dilated convolutional chain structure network with different dilation coefficients through a combination of convolutional layers, thereby obtaining spatial features.

[0016] The second output module is configured to input the spatial features and the multimodal time series set into a preset spatiotemporal convolutional attention network, and output spatiotemporal fusion features that fuse space and time through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network;

[0017] The determination module is configured to determine the contact quality index of the pantograph, wherein the contact quality index includes pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters;

[0018] The generation module is configured to input the spatiotemporal fusion features and the contact quality index into a preset digital twin model, generate real-time optimization control commands using gradient descent and PID closed-loop control algorithms, and send the real-time optimization control commands to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph.

[0019] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one 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 steps of the pantograph current collection quality closed-loop optimization method according to any embodiment of the present invention.

[0020] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the pantograph current collection quality closed-loop optimization method according to any embodiment of the present invention.

[0021] The pantograph current collection quality closed-loop optimization method and system of this application have the following beneficial effects:

[0022] 1. By constructing a dilated convolutional chain structure network and utilizing the combination of convolutional layers with different dilation coefficients, we can effectively capture multi-scale spatial features from microscopic deformation to macroscopic instability, thereby improving the ability to characterize the structural state of the pantograph under complex working conditions and providing richer feature information for subsequent optimization.

[0023] 2. A spatiotemporal convolutional attention network is introduced, which processes spatial and temporal information through spatial feature branches and temporal feature branches respectively. It also achieves deep fusion of spatial and temporal features with the help of cross-modal attention mechanism and dynamic memory mechanism, which significantly improves the comprehensive modeling ability of pantograph dynamic behavior and its evolution law.

[0024] 3. A high-fidelity pantograph-contact network system simulation environment was constructed based on a digital twin model. By combining the gradient descent method and PID control algorithm, the real-time optimization of the contact quality index and the dynamic generation of control commands were realized, ensuring that the system can quickly respond to external changes and maintain the stability and reliability of the current collection process.

[0025] 4. Through a hierarchical cross-modal attention mechanism and dynamic memory module, the system can adaptively focus on key features and historical states, achieving intelligent adaptation to different operating conditions, and has strong generalization ability and robustness.

[0026] 5. Through the joint optimization of multiple quality indices such as pressure uniformity, dynamic stability and thermal balance, the current collection quality of the pantograph has been effectively improved, reducing problems such as arc discharge, plate wear and contact wire vibration, extending equipment life and improving the safety and economy of train operation. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0028] Figure 1 A flowchart of a closed-loop optimization method for pantograph current collection quality provided in an embodiment of the present invention;

[0029] Figure 2 This is a structural block diagram of a pantograph current collection quality closed-loop optimization system provided in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0031] 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.

[0032] Please see Figure 1 The diagram shows a flowchart of a closed-loop optimization method for pantograph current collection quality according to this application.

[0033] like Figure 1 As shown, the closed-loop optimization method for pantograph current collection quality specifically includes the following steps:

[0034] Step S101: Obtain the pantograph multimodal dataset and multimodal time series set. The multimodal dataset includes pantograph pressure data, temperature data and strain data. The multimodal time series set includes time series of pressure data, time series of temperature data and time series of strain data.

[0035] In this step, a flexible multimodal sensor array with a flexible substrate, a multifunctional sensitive layer, and an arc-resistant protective layer is uniformly arranged on the surface of the pantograph sliding plate. The sensor array is precisely attached to the pantograph sliding plate with a matrix density of 5mm×5mm, and real-time synchronously collects three-dimensional data of pressure, temperature, and strain. The acquisition frequency reaches the sub-millisecond level, achieving strict clock synchronization of different modal data. The collected three-dimensional data of pressure, temperature, and strain directly constitute a multimodal dataset. The pressure value sequence, temperature value sequence, and strain value sequence arranged in order of acquisition time constitute the time series of pressure data, temperature data, and strain data, respectively. The three together form a multimodal time series set.

[0036] Step S102: Based on the preset dilated convolutional chain structure network, multi-scale spatial features are extracted from the multimodal dataset through a combination of convolutional layers with different dilation coefficients to obtain spatial features.

[0037] In this step, the dilation coefficient of the first-level dilated convolutional layer is d=1. Convolution calculation is performed on the multimodal dataset to extract deformation features at the microscale and output the first-level feature map. ;

[0038] The dilation coefficient of the second-level dilated convolutional layer is d=2, based on the first-level feature map. The input is used for convolution calculation to extract vibration features at the mesoscale, and the output is a second-level feature map. ;

[0039] The dilation coefficient of the third-level dilated convolutional layer is d=4, based on the second-level feature map. The input is used for convolution calculation to extract instability features at a macro scale, and the output is a third-level feature map. ;

[0040] For the first-level feature map Second-level feature map and the third-level feature map Spatial features are obtained by splicing or weighted fusion along the channel dimension;

[0041] The dilated convolutional layers at each level are connected by a residual connection structure, the specific expression of which is:

[0042] ,

[0043] In the formula, This is the output feature map of the nth level dilated convolutional layer. This is the input feature map for the nth level dilated convolutional layer. This indicates a convolution operation with an expansion factor of d, BN indicates a batch normalization operation, and LeakyReLU indicates a linear rectified function with a negative slope coefficient of 0.1.

[0044] Step S103: Input the spatial features and the multimodal time series set into a preset spatiotemporal convolutional attention network, and output the spatiotemporal fusion features that fuse space and time through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network.

[0045] In this step, the spatial features are input to the spatial feature branch, and the spatial feature branch outputs an initial spatial feature map. Spatial global average pooling and spatial global max pooling are then performed on the initial spatial feature map to obtain two features. Finally, the two pooled features are concatenated along the channel dimension and then input. The convolutional layer is then processed by the Sigmoid activation function to obtain the spatial attention weight matrix.

[0046] The initial spatial feature map is weighted according to the spatial attention weight matrix to output the target feature map after spatial attention enhancement;

[0047] The multimodal time series set is input into the time series feature branch. Based on the bidirectional gated loop unit in the time series feature branch, the dynamic evolution law of contact force over time is modeled, and the time series feature is output. The bidirectional gated loop unit consists of a forward gated loop unit (GRU) and a backward gated loop unit (GRU). The forward GRU processes the data in forward time order, and the backward GRU processes the data in reverse time order. Finally, the outputs of the forward GRU and the backward GRU at each time step are concatenated to obtain the time series feature containing bidirectional time series correlation.

[0048] Based on a hierarchical cross-modal attention mechanism, modal interaction fusion is first performed on the hierarchically segmented target feature map in the spatial modality and the hierarchically segmented temporal features in the temporal modality to obtain fused features. Then, the fused features at the current moment are concatenated with the historical memory state at the previous moment and input into the gating unit to generate gating weights, thereby updating the memory state, and finally obtaining the spatiotemporal fused features. This mechanism achieves cross-modal information fusion by learning pairwise associations between spatial hierarchical regions and temporal hierarchical windows, and its expression is:

[0049] ,

[0050] In the formula, For the pantograph The spatial local region and the multimodal time series of the first Local attention weights between time windows First, the initial spatial feature map is processed through spatial feature branching to obtain a spatially attention-enhanced target feature map. Then, the target feature map is split into multiple independent spatial local features according to a preset partitioning rule, where the first... The feature map of a local spatial region is used for extraction. The basic data, and then for the first Convolution operations are performed on the feature maps of local spatial regions to compress the feature dimensions and enhance key information. Finally, the convolved feature maps are globally averaged to output a vector of fixed dimensions. This vector is the... , To divide a multimodal time series set into time windows of fixed duration, the first... Multimodal time-series data within a given time window is input into a bidirectional gated recurrent unit (BiGRU). The BiGRU then processes the data in both forward and reverse time sequences to capture bidirectional temporal correlations and extract the first... The BiGRU output feature corresponding to each time window is... , For feature dimension, It is the transpose symbol;

[0051] The expression for updating the memory state is:

[0052] ,

[0053] ,

[0054] ,

[0055] ,

[0056] In the formula, This represents the updated memory state at the current moment, and also the spatiotemporal fusion feature ultimately output by the network. This refers to the state of historical memory from the previous moment. The cross-modal fusion features obtained at the current moment through a hierarchical cross-modal attention mechanism are... These are respectively resetting the gate output vector, updating the gate output vector, and the candidate memory vector. This is the weight matrix corresponding to the gating. For the bias term corresponding to the gating, Activated for Sigmoid. This is an element-wise multiplication, where λ is the residual coefficient.

[0057] Step S104: Determine the contact quality index of the pantograph, which includes pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters.

[0058] In this step, the pressure uniformity parameter Through formula The calculation yielded, where Let be the standard deviation of the contact pressure between the pantograph and the overhead contact line. This is the rated contact pressure of the pantograph;

[0059] Dynamic stability parameters Through formula The calculation yielded, where This represents the maximum fluctuation range of the contact force within a 1-second time window. This is the rated contact pressure of the pantograph;

[0060] Thermal balance parameters Through formula The calculation yielded, where The highest temperature of the pantograph slider. This is the maximum permissible temperature for the pantograph slider.

[0061] Step S105: Input the spatiotemporal fusion features and the contact quality index into a preset digital twin model, use the gradient descent method and PID closed-loop control algorithm to generate real-time optimization control commands, and send the real-time optimization control commands to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph.

[0062] In this step, a digital twin model is constructed based on the three-dimensional physical and thermodynamic multibody dynamics model of the pantograph and the overhead contact line. The state update equation expression of the digital twin model is as follows:

[0063] ,

[0064] In the formula, The state variables of the digital twin model at time t include the pantograph position, contact force, and temperature. Real-time control commands generated for digital twin models For contact quality index, For co-simulation functions;

[0065] An optimization loss function is constructed with the contact quality index as the objective, and the optimization control parameters are updated in real time using gradient descent. The expression of the optimization loss function is as follows:

[0066] ,

[0067] In the formula, These are the weighting coefficients for pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters, respectively. , , These are pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters, respectively. To optimize the loss function, To balance the optimization objective and smoothness, the regularization coefficient is set to a value ranging from 0.01 to 0.1. The L2 norm of the velocity change;

[0068] The expression for updating and optimizing the control parameters is:

[0069] ,

[0070] In the formula, These are the control parameters at time t+1. The control parameters at time t, The learning rate is 0.01. To optimize the partial derivative of the loss function with respect to the control parameters at time t;

[0071] Based on the optimized control parameters, a PID closed-loop control instruction is constructed, thus obtaining the real-time optimized control instruction. The expression of the real-time optimized control instruction is as follows:

[0072] ,

[0073] In the formula, This refers to the optimized control command at time t. This is the proportional gain coefficient for the contact force deviation. The proportional gain coefficient for velocity deviation. The deviation between the set value and the actual value of the contact force. The deviation between the speed setpoint and the actual value.

[0074] It should be noted that the magnetorheological actuator receives real-time optimized control commands output by the digital twin module. The control commands include the target contact force and dynamic weights, and generate a magnetorheological fluid excitation current control signal based on the target contact force and dynamic weights.

[0075] The magnetorheological actuator adjusts the viscosity of the magnetorheological fluid in real time according to the excitation current control signal to dynamically adjust the contact state between the pantograph slide and the contact wire, thereby maintaining the contact pressure uniformity index above 0.75 and controlling the fluctuation range of the dynamic stability index parameter within ±0.1.

[0076] Meanwhile, the multimodal flexible sensor array provides real-time feedback of the optimized contact quality index (CQI). When the feedback contact quality index is lower than the preset threshold for one consecutive second, the magnetorheological actuator automatically triggers the digital twin module to recalculate and update the control parameters and control signals, so as to achieve continuous optimization and closed-loop control of the pantograph's current collection quality.

[0077] In summary, the method of this application solves the defects of existing technologies and achieves significant technical effects through full-chain innovation: In the data acquisition stage, a multimodal sensor array with a flexible substrate, a multifunctional sensitive layer and an arc-resistant protective layer is adopted and attached to the pantograph sliding plate with a matrix density of 5mm×5mm. Pressure, temperature and strain data are collected synchronously in the sub-millisecond range, which solves the problems of asynchronous multimodal data, low accuracy and arc interference, and provides high-quality data with time alignment and accuracy for subsequent stages.

[0078] In the multi-scale spatial feature extraction stage, an innovative dilated convolutional chain structure network is used. With three levels of convolutional layers with dilation coefficients d=1, 2, and 4, combined with attention-guided residual connection structure (AGRC), the micro-deformation, meso-vibration, and macro-instability features of the skateboard are extracted. This accurately captures the correlation between the skateboard and the bow features, providing precise spatial feature support for spatiotemporal fusion.

[0079] In the spatiotemporal feature fusion stage, the bidirectional gated recurrent unit (BiGRU) of the spatiotemporal convolutional attention network solves the problem of temporal information loss in traditional recurrent networks, making the temporal feature characterization error less than 5%. The hierarchical cross-modal attention mechanism can dynamically adjust modal weights according to scenarios such as train speed and catenary span. The gated residual memory mechanism (GRM) can improve the continuity of control commands and reduce the impact of instantaneous interference through dual gating and residual terms.

[0080] In the process of quantifying the current collection quality and controlling the closed loop, the quantitative calculation of the contact quality index provides a clear target for control. The digital twin model fully considers actual factors such as the nonlinearity of the contact wire and the wear of the sliding plate, and generates optimization instructions by combining gradient descent and PID algorithms. The magnetorheological actuator adjusts the contact state accordingly, so that the pressure uniformity index is stabilized above 0.75, the fluctuation range of the dynamic stability parameter is controlled within ±0.1, and the deviation of the thermal balance parameter is reduced.

[0081] Please see Figure 2 The diagram shows a structural block diagram of a pantograph current collection quality closed-loop optimization system according to this application.

[0082] like Figure 2 As shown, the pantograph current collection quality closed-loop optimization system 200 includes an acquisition module 210, a first output module 220, a second output module 230, a determination module 240, and a generation module 250.

[0083] The acquisition module 210 is configured to acquire a multimodal dataset and a multimodal time series set of the pantograph. The multimodal dataset includes pressure data, temperature data, and three-dimensional strain data of the pantograph. The multimodal time series set includes time series of the pressure data, time series of the temperature data, and time series of the three-dimensional strain data. The first output module 220 is configured to extract multi-scale spatial features from the multimodal dataset using a pre-defined dilated convolutional chain structure network with convolutional layers of different dilation coefficients to obtain spatial features. The second output module 230 is configured to input the spatial features and the multimodal time series set into a pre-defined spatiotemporal convolutional attention network. In the process, the spatiotemporal fusion features are output through the spatial feature branch, temporal feature branch, cross-modal attention mechanism, and dynamic memory mechanism in the spatiotemporal convolutional attention network; the determination module 240 is configured to determine the contact quality index of the pantograph, which includes pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters; the generation module 250 is configured to input the spatiotemporal fusion features and the contact quality index into a preset digital twin model, generate real-time optimized control commands using gradient descent and PID closed-loop control algorithms, and send the control commands to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph, and finally output the optimized current collection quality result.

[0084] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0085] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the pantograph current collection quality closed-loop optimization method in any of the above method embodiments.

[0086] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0087] A multimodal dataset and a multimodal time series set of the pantograph are obtained. The multimodal dataset contains pressure data, temperature data, and three-dimensional strain data of the pantograph. The multimodal time series set contains time series of the pressure data, time series of the temperature data, and time series of the three-dimensional strain data.

[0088] Based on a pre-defined dilated convolutional chain structure network, spatial features are extracted from the multimodal dataset using a combination of convolutional layers with different dilation coefficients to obtain spatial features.

[0089] The spatial features and the multimodal time series set are input into a preset spatiotemporal convolutional attention network. The spatiotemporal fusion features that fuse space and time are output through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network.

[0090] The contact quality index of the pantograph is determined, including pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters.

[0091] The spatiotemporal fusion features and the contact quality index are input into a preset digital twin model. Gradient descent and PID closed-loop control algorithms are used to generate real-time optimized control commands, which are then sent to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph and finally output the optimized current collection quality results.

[0092] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the pantograph current collection quality closed-loop optimization system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, and these remote memories may be connected to the pantograph current collection quality closed-loop optimization system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the pantograph current collection quality closed-loop optimization method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the pantograph current collection quality closed-loop optimization system. The output device 340 may include a display screen or other display device.

[0094] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0095] In one implementation, the above-described electronic device is applied in a pantograph current collection quality closed-loop optimization system for a client, comprising: at least one processor; and a memory communicatively connected to the at least one 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:

[0096] A multimodal dataset and a multimodal time series set of the pantograph are obtained. The multimodal dataset contains pressure data, temperature data, and three-dimensional strain data of the pantograph. The multimodal time series set contains time series of the pressure data, time series of the temperature data, and time series of the three-dimensional strain data.

[0097] Based on a pre-defined dilated convolutional chain structure network, spatial features are extracted from the multimodal dataset using a combination of convolutional layers with different dilation coefficients to obtain spatial features.

[0098] The spatial features and the multimodal time series set are input into a preset spatiotemporal convolutional attention network. The spatiotemporal fusion features that fuse space and time are output through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network.

[0099] The contact quality index of the pantograph is determined, including pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters.

[0100] The spatiotemporal fusion features and the contact quality index are input into a preset digital twin model. Gradient descent and PID closed-loop control algorithms are used to generate real-time optimized control commands, which are then sent to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph and finally output the optimized current collection quality results.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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 closed-loop optimization method for pantograph current collection quality, characterized in that, include: A multimodal dataset and a multimodal time series set of pantograph data are obtained. The multimodal dataset contains pressure data, temperature data, and strain data of the pantograph data. The multimodal time series set contains time series of the pressure data, time series of the temperature data, and time series of the strain data. Based on a pre-defined dilated convolutional chain structure network, spatial features are extracted from the multimodal dataset using a combination of convolutional layers with different dilation coefficients to obtain spatial features. The spatial features and the multimodal time series set are input into a preset spatiotemporal convolutional attention network. The spatiotemporal fusion features that fuse space and time are output through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network. The process includes: inputting the spatial features into the spatial feature branch, the spatial feature branch outputting an initial spatial feature map, and using a spatial attention gating mechanism to dynamically enhance the feature response of key areas on the surface of the pantograph skateboard to obtain a spatial attention weight matrix. The initial spatial feature map is weighted according to the spatial attention weight matrix to output the target feature map after spatial attention enhancement; The multimodal time series set is input into the time series feature branch. Based on the bidirectional gated loop unit in the time series feature branch, the dynamic evolution law of contact force over time is modeled, and the time series features are output. The target feature map and the temporal features are fused based on a hierarchical cross-modal attention mechanism to obtain fused features. The fused features at the current time step are then concatenated with the historical memory state at the previous time step and input into a gating unit to generate gating weights to update the memory state, thus obtaining the spatiotemporal fused features. The expression for updating the memory state is as follows: ， ， ， ， In the formula, This represents the updated memory state at the current moment, and also the spatiotemporal fusion feature ultimately output by the network. This refers to the state of historical memory from the previous moment. The cross-modal fusion features obtained at the current moment through a hierarchical cross-modal attention mechanism are... These are respectively resetting the gate output vector, updating the gate output vector, and the candidate memory vector. This is the weight matrix corresponding to the gating. For the bias term corresponding to the gating, Activated for Sigmoid. For element-wise multiplication, The residual coefficient; The contact quality index of the pantograph is determined, including pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters. The spatiotemporal fusion features and the contact quality index are input into a preset digital twin model. A real-time optimization control command is generated using gradient descent and a PID closed-loop control algorithm. This real-time optimization control command is then sent to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph. Specifically, generating the real-time optimization control command includes: Based on the three-dimensional physical and thermodynamic multibody dynamics model of the pantograph and the overhead contact line, a digital twin model is constructed. The state update equation of the digital twin model is expressed as follows: ， In the formula, The state variables of the digital twin model at time t include the pantograph position, contact force, and temperature. Real-time control commands generated for digital twin models For contact quality index, For co-simulation functions; An optimization loss function with the contact quality index as the objective is constructed, and the optimization control parameters are updated in real time using the gradient descent method. Based on the optimized control parameters, a PID closed-loop control instruction is constructed, thus obtaining the real-time optimized control instruction. The expression of the real-time optimized control instruction is as follows: ， In the formula, This refers to the optimized control command at time t. This is the proportional gain coefficient for the contact force deviation. The proportional gain coefficient for velocity deviation. The deviation between the set value and the actual value of the contact force. The deviation between the speed setpoint and the actual value.

2. The pantograph current collection quality closed-loop optimization method according to claim 1, characterized in that, The dilated convolutional chain structure network includes a first-level dilated convolutional layer, a second-level dilated convolutional layer, a third-level dilated convolutional layer, and a residual connection structure. The multimodal dataset is subjected to multi-scale spatial feature extraction using a pre-defined dilated convolutional chain structure network with convolutional layers of different dilation coefficients. The resulting spatial features include: The first-level dilated convolutional layer has a dilation coefficient of d=1. It performs convolution calculations on the multimodal dataset to extract deformation features at the microscale and outputs the first-level feature map. ; The dilation coefficient of the second-level dilated convolutional layer is d=2, based on the first-level feature map. The input is used for convolution calculation to extract vibration features at the mesoscale, and the output is a second-level feature map. ; The dilation coefficient of the third-level dilated convolutional layer is d=4, based on the second-level feature map. The input is used for convolution calculation to extract instability features at a macro scale, and the output is a third-level feature map. ; For the first-level feature map Second-level feature map and the third-level feature map Spatial features are obtained by splicing or weighted fusion along the channel dimension; The dilated convolutional layers at each level are connected by a residual connection structure, the specific expression of which is: ， In the formula, This is the output feature map of the nth level dilated convolutional layer. This is the input feature map for the nth level dilated convolutional layer. This indicates a convolution operation with an expansion factor of d, BN indicates a batch normalization operation, and LeakyReLU indicates a linear rectified function with a negative slope coefficient of 0.

1.

3. The closed-loop optimization method for pantograph current collection quality according to claim 1, characterized in that, in, The expression for the hierarchical cross-modal attention mechanism is: ， In the formula, For the pantograph The spatial local region and the multimodal time series of the first Local attention weights between time windows To the pantograph sliding plate Spatial key feature vectors extracted from local spatial region feature maps through convolution and global average pooling. To obtain the first multimodal time series The key time feature vectors extracted from the data of each time window through a bidirectional gated loop unit. For feature dimension, This is the transpose symbol.

4. The closed-loop optimization method for pantograph current collection quality according to claim 1, characterized in that, The expression for the optimization loss function is: ， In the formula, These are the weighting coefficients for pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters, respectively. These are pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters, respectively. To optimize the loss function, To balance the optimization objective and smoothness, the regularization coefficient is set to a value ranging from 0.01 to 0.

1. The L2 norm of the velocity change; The expression for updating and optimizing the control parameters is: ， In the formula, These are the control parameters at time t+1. The control parameters at time t, The learning rate is 0.

01. To optimize the loss function pair The partial derivatives of the control parameters at time t.

5. A closed-loop optimization system for pantograph current collection quality, characterized in that, include: The acquisition module is configured to acquire a multimodal dataset and a multimodal time series set of the pantograph. The multimodal dataset includes pressure data, temperature data, and strain data of the pantograph, and the multimodal time series set includes time series of the pressure data, time series of the temperature data, and time series of the strain data. The first output module is configured to extract spatial features from the multimodal dataset by using a pre-defined dilated convolutional chain structure network with different dilation coefficients through a combination of convolutional layers, thereby obtaining spatial features. The second output module is configured to input the spatial features and the multimodal time series set into a preset spatiotemporal convolutional attention network, and output spatiotemporal fusion features that fuse space and time through the spatial feature branch, temporal feature branch, cross-modal attention mechanism and dynamic memory mechanism in the spatiotemporal convolutional attention network. The output module includes: inputting the spatial features into the spatial feature branch, the spatial feature branch outputting an initial spatial feature map, and using a spatial attention gating mechanism to dynamically enhance the feature response of key areas on the surface of the pantograph skateboard to obtain a spatial attention weight matrix. The initial spatial feature map is weighted according to the spatial attention weight matrix to output the target feature map after spatial attention enhancement; The multimodal time series set is input into the time series feature branch. Based on the bidirectional gated loop unit in the time series feature branch, the dynamic evolution law of contact force over time is modeled, and the time series features are output. The target feature map and the temporal features are fused based on a hierarchical cross-modal attention mechanism to obtain fused features. The fused features at the current time step are then concatenated with the historical memory state at the previous time step and input into a gating unit to generate gating weights to update the memory state, thus obtaining the spatiotemporal fused features. The expression for updating the memory state is as follows: ， ， ， ， In the formula, This represents the updated memory state at the current moment, and also the spatiotemporal fusion feature ultimately output by the network. This refers to the state of historical memory from the previous moment. The cross-modal fusion features obtained at the current moment through a hierarchical cross-modal attention mechanism are... These are respectively resetting the gate output vector, updating the gate output vector, and the candidate memory vector. This is the weight matrix corresponding to the gating. For the bias term corresponding to the gating, Activated for Sigmoid. For element-wise multiplication, The residual coefficient; The determination module is configured to determine the contact quality index of the pantograph, wherein the contact quality index includes pressure uniformity parameters, dynamic stability parameters, and thermal balance parameters; The generation module is configured to input the spatiotemporal fusion features and the contact quality index into a preset digital twin model, generate real-time optimization control commands using gradient descent and a PID closed-loop control algorithm, and issue the real-time optimization control commands to the magnetorheological actuator to dynamically optimize the current collection state of the pantograph. Specifically, generating the real-time optimization control commands includes: Based on the three-dimensional physical and thermodynamic multibody dynamics model of the pantograph and the overhead contact line, a digital twin model is constructed. The state update equation of the digital twin model is expressed as follows: ， In the formula, The state variables of the digital twin model at time t include the pantograph position, contact force, and temperature. Real-time control commands generated for digital twin models For contact quality index, For co-simulation functions; An optimization loss function with the contact quality index as the objective is constructed, and the optimization control parameters are updated in real time using the gradient descent method. Based on the optimized control parameters, a PID closed-loop control instruction is constructed, thus obtaining the real-time optimized control instruction. The expression of the real-time optimized control instruction is as follows: ， In the formula, This refers to the optimized control command at time t. This is the proportional gain coefficient for the contact force deviation. The proportional gain coefficient for velocity deviation. The deviation between the set value and the actual value of the contact force. The deviation between the speed setpoint and the actual value.

6. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 4.

Images