High-speed train internal and external light environment prediction method and system based on multi-source information fusion

Abstract

The invention provides a high-speed train internal and external light environment prediction method and system based on multi-source information fusion. The method comprises the following steps: installing a sensor and an HDR image acquisition unit; acquiring data, preprocessing and fusing the data; performing light field inversion and prediction based on the fused data; and executing a regulation strategy based on a prediction result. According to the invention, the sensor array and the HDR image acquisition unit work cooperatively, high-frequency discrete monitoring of the light environment outside the vehicle is realized, and the system has the advantages of comprehensive space coverage, high data fidelity and fast dynamic response; natural illumination parameters, line GIS information and tunnel optical characteristics are incorporated into a unified coordinate system and considered, the prediction precision is improved, and the scene adaptability is high; the light transmittance of the electrochromic glass and the light supplementing intensity of the tunnel LED are dynamically adjusted based on the light intensity gradient prediction result, vehicle-tunnel collaborative millisecond regulation and control are formed, the visual load of a driver can be remarkably reduced, and meanwhile energy consumption is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of high-speed train light regulation, and in particular to a method and system for predicting the internal and external light environment of a high-speed train based on multi-source information fusion. Background Art

[0002] The transient light environment inside and outside high-speed train tunnels has become a key factor affecting driving safety and passenger comfort. When a train passes through a tunnel at high speed, the light environment in the driver's cab undergoes a dramatic alternation from natural sunlight to tunnel darkness within seconds, causing a surge in the driver's visual load and even temporary visual blind spots. Existing technologies for monitoring and predicting the external light environment mainly focus on a single dimension, failing to effectively address the needs for high-precision perception and real-time control of dynamic light environments. Specifically, there are the following limitations:

[0003] Existing high-speed train light environment monitoring mostly relies on fixed sensors, which cannot dynamically capture transient light field characteristics (such as sudden changes in brightness gradients inside and outside tunnels);

[0004] Existing prediction models ignore the multi-physics coupling effect of vehicle-tunnel-environment, resulting in low accuracy in predicting the spatial distribution of the light environment.

[0005] The existing buffer control strategy lacks real-time integration of natural light, line topology and tunnel structure parameters, resulting in obvious regulation lag. Summary of the Invention

[0006] The purpose of the present invention is to address the deficiencies in the above-mentioned background technology and provide a high-speed train internal and external light environment prediction and control solution based on three major directions: hardware architecture innovation, data fusion algorithm optimization, and prediction model accuracy improvement.

[0007] To achieve the above objectives, the present invention provides a method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion, comprising the following steps:

[0008] S1, install sensors and HDR image acquisition units at preset locations on the high-speed train. The sensors include front sensors and side window sensors to obtain light intensity data and HDR images;

[0009] S2, noise is filtered out of sensor data, HDR images, and imported data, and then the data is fused to complete spatiotemporal alignment;

[0010] S3, based on the fused data, performs light field inversion through the U-Net light field inversion network, and predicts the light intensity gradient in the future period through the LSTM-Kalman hybrid prediction model;

[0011] S4, executing a control strategy based on the prediction results, changing the light transmittance of the electrochromic glass of the vehicle window, and regulating the fill light process at the same time.

[0012] Furthermore, the front sensor in S1 is a multispectral sensor; the side window sensor is a polarized light sensor; the HDR image acquisition unit is electrically connected to the edge computing unit, which can obtain the bidirectional reflectance distribution function parameters of the tunnel wall through multi-frame synthesis, and at the same time extract the solar incidence angle and cloud transmission attenuation coefficient based on the HDR image.

[0013] Furthermore, the data imported into S2 include environmental parameters, line GIS data, and tunnel optical parameters; the environmental parameters include solar altitude angle, weather type, and surface reflectivity; the line GIS data includes tunnel position, curvature radius, and slope; and the tunnel optical parameters include buffer structure transmittance and inner wall reflection coefficient.

[0014] Furthermore, the sensor data in S2 uses Daubechies4 wavelet transform to filter out high-frequency noise above 50Hz and retain the effective signal frequency band of 0-50Hz; HDR images are synthesized through multi-frame weighted averaging, and the images are matched based on SIFT feature points to automatically remove blurred images. The trigger mechanism is the frame difference standard deviation σ>15%.

[0015] Furthermore, data fusion includes time alignment and spatial alignment. In spatial alignment, sensor data, HDR images, and line GIS data are unified into the same coordinate system. In temporal alignment, GPS and PPS signals are used as the benchmark to align light environment data and train position information.

[0016] During data fusion, a dynamic weight allocation model based on Kalman filtering is used, and the environmental parameter weights, line data weights, and optical parameter weights are reset to preset values.

[0017] Furthermore, the input of the U-Net light field inversion network in S3 is discrete sensor data + tunnel point cloud data; the network structure is set to a 4-layer encoder-decoder architecture, each layer contains 2 convolution blocks, and the jump connection introduces the SE attention module to weightedly fuse low-level details and high-level semantic features; the output is set to a 512×512 pixel light field distribution map with a resolution of 0.25m / pixel, corresponding to a real scene of 128m×128m; the network training data set includes: 100,000 sets of tunnel crossing scene data, 70% training set, 15% validation set, and 15% test set, the initial learning rate is set to 0.001, the batch size is set to 32, the loss function is weighted MSE, the illumination weight is set to 0.6, and the color temperature weight is set to 0.4.

[0018] Furthermore, the input of the LSTM-Kalman hybrid prediction model in S3 is set as the historical 10-second light intensity sequence, vehicle speed, tunnel curvature, and solar altitude angle; the LSTM parameters are set as 256 neurons in the hidden layer and 10 time steps; the output is set as the light intensity gradient ΔL inside and outside the tunnel within a preset time in the future;

[0019] After Kalman filter correction, the state equation is:

[0020] x k =Ax k-1 +Bu k +w k

[0021] Among them, A is the state transfer matrix, B is the control matrix, u k is the vehicle speed / light intensity change rate, w k To control noise;

[0022] The observation equation is:

[0023] z k =Hx k +v k ,v k ~N(0,R),R=0.1

[0024] Among them, H is the observation matrix, v k is the sensor noise.

[0025] Furthermore, when the transmittance of the electrochromic glass is regulated in S4, the control signal is based on the light intensity gradient ΔL output by the prediction model:

[0026] T target =80%-0.7×ΔL(ΔL∈[0,1000lx / m])

[0027] Among them, T target is the target transmittance;

[0028] The car windows adopt a zoning control strategy, with the transmittance of each zone independently controlled, and the difference in transmittance between adjacent zones ≤15%.

[0029] Furthermore, in S4, the LED light array is deployed at a preset distance in front of the tunnel entrance. The formula for the fill light intensity of the LED array is:

[0030] I 补 =max(0,0.8·ΔL-I 环境 )

[0031] Among them, I 补 is the fill light intensity of the LED array, I 环境 is the ambient light intensity; the fill light intensity is dynamically adjusted according to the output of the prediction model.

[0032] The present invention also provides a high-speed train internal and external light environment prediction system based on multi-source information fusion, which adopts the above-mentioned method and is characterized by comprising an acquisition module, a data processing module, a light field inversion prediction module and a control execution module;

[0033] The acquisition module includes a vehicle head sensor, a side window sensor and an HDR image acquisition unit, and is used to realize high-frequency discrete monitoring of the light environment outside the vehicle;

[0034] The data processing module is used to filter noise from sensor data, HDR images and imported data, align the data in time and space, and perform dynamic weight allocation based on Kalman filtering;

[0035] The light field inversion prediction module uses the U-Net light field inversion network and the LSTM-Kalman hybrid prediction model to invert and predict the light intensity gradient inside and outside the tunnel within a certain period of time in the future;

[0036] The control execution module executes the control strategy based on the light intensity gradient prediction result, changes the light transmittance of the electrochromic glass of the vehicle window, and controls the fill light process at the same time.

[0037] The above solution of the present invention has the following beneficial effects:

[0038] The method and system for predicting the interior and exterior light environment of high-speed trains based on multi-source information fusion, based on high-density dynamic perception and precise inversion, are specifically embodied in the following aspects: A sensor array and an HDR image acquisition unit work together to achieve high-frequency discrete monitoring of the exterior light environment, with the advantages of comprehensive spatial coverage, high data fidelity, and fast dynamic response; natural lighting parameters, line GIS information, and tunnel optical characteristics are incorporated into a unified coordinate system and considered, improving prediction accuracy and providing strong scenario adaptability; and dynamically adjusting the transmittance of electrochromic glass and the intensity of tunnel LED fill light based on the light intensity gradient prediction results, achieving millisecond-level vehicle-tunnel coordinated control, which can significantly reduce the driver's visual burden while optimizing energy consumption.

[0039] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of the method of the present invention;

[0041] Figure 2 Schematic diagram of the arrangement of the sensor and HDR image acquisition unit in S1 of the present invention;

[0042] Figure 3 Schematic diagram of the U-Net light field inversion network structure in S3 of the present invention. DETAILED DESCRIPTION

[0043] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0044] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement an apparatus and / or practice a method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this apparatus and / or practice this method.

[0045] It should also be noted that the diagrams provided in the following embodiments are merely schematic illustrations of the basic concepts of the present disclosure. The diagrams only show components relevant to the present disclosure and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the configuration, quantity, and proportion of each component may be varied at will, and the component layout may be more complex. Furthermore, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will appreciate that the described aspects may be practiced without these specific details.

[0046] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the internal and external light environment of a high-speed train based on multi-source information fusion, comprising the following steps:

[0047] S1: Install the sensor and HDR image acquisition unit. This step specifically includes:

[0048] Arrangement of vehicle head sensors: Considering that the curvature radius of the vehicle head surface is ≥2m, to maintain field of view uniformity, three groups of multispectral sensors (model OMS-200) are arranged equidistantly along the central curved surface, with a spacing of preferably 0.5m. By adjusting the bracket of the vehicle head sensor, the lens of the vehicle head sensor is tilted downward 15° horizontally, enabling wide-angle coverage of the vehicle head, for example, vertical ±30° and horizontal ±60°.

[0049] Among them, the multispectral sensor uses a combination of a spectroscopic prism and a filter, with the coverage wavelength set to 380~780nm (visible light) and the resolution set to 0.1nm.

[0050] Among them, the bracket of the front sensor can be made of 3D printed carbon fiber composite material with a thickness of preferably 2mm. The mounting holes are fixed with hexagonal countersunk bolts. The seismic resistance level complies with the EN 12663-1 standard, thus meeting the installation requirements of high-speed trains.

[0051] Among them, the diameter of the CAN-FD bus cable of the front sensor is preferably set to 6mm, the surface is covered with a fluororubber high-temperature sheath, the wiring is laid along the front keel, and the interface number is CAN-ID-0x30~0x32.

[0052] Arrangement of side window sensors: For an 8-car train, an average of 1.5 sets of polarization sensors (model PL-300) are arranged in each car, for a total of 12 sets. The side window sensors are located 10 cm above the upper edge of each car's side window. An adjustable universal joint is installed at the bottom of the side window sensor bracket. After installation, a laser instrument is used for horizontal calibration to ensure that the horizontal error does not exceed ±0.5°. The polarization sensor can eliminate reflection interference from the glass surface.

[0053] Among them, the housing of the side window sensor has an IP68 rating, and the front of the housing is made of tempered glass and coated with an anti-glare AR film to effectively reduce interference from sunlight mirror reflections.

[0054] Arrangement of the HDR image acquisition unit: A global shutter CMOS (model XIEMEI XiQ-4096, resolution 4096×2160, dynamic range 120dB) with a fisheye lens (FOV 180°, focal length 2.8mm, field of view 180°) is used. It is placed on the top of the vehicle body and fixed by a three-axis Maxwell elastic vibration reduction component to avoid image blur during high-speed operation.

[0055] The HDR image acquisition unit is connected to the edge computing unit via Gigabit Ethernet. The image format is RAW 12-bit and supports HDR synthesis. The edge computing unit can obtain the BRDF (bidirectional reflectance distribution function) parameters of the tunnel wall through multi-frame synthesis, and extract the solar incidence angle (accuracy of ±0.1°) and cloud transmission attenuation coefficient based on the HDR image.

[0056] Among them, the specific layout positions of the front sensor, side window sensor, and HDR image acquisition unit are as follows: Figure 2 As shown, other sensor parameters and communication configurations are shown in Table 1:

[0057] Table 1 Sensor parameters and communication configuration

[0058] unit Parameter configuration description Multispectral sensors Frequency 100Hz, error ≤±1.5% Polarization sensor Polarization angle adjustable (0°~180°), light intensity resolution 0.01lx HDR image acquisition unit Aperture f / 2.8, exposure range 1 / 1000s to 1 / 30s, frame rate 3fps, output RAW image Time synchronization mechanism GPS and PPS signals are updated every second, and with CAN clock synchronization, the time error is ≤1ms

[0059] S2, acquires data, preprocesses and fuses data.

[0060] In this step, in addition to sensor data and HDR images, the data also includes imported data, specifically environmental parameters, route GIS data, and tunnel optical parameters. Environmental parameters include solar altitude, weather type (sunny / rainy / foggy), and surface reflectivity (snowy / mountainous). Route GIS data includes tunnel location, curvature radius, and slope. Tunnel optical parameters include buffer structure transmittance (adjustable from 50% to 90%) and inner wall reflectivity (0.2 to 0.8).

[0061] This data is filtered for noise. Specifically, the Daubechies4 wavelet transform is applied to the sensor data to remove high-frequency noise above 50 Hz, retaining the valid signal frequency range of 0-50 Hz. HDR images are synthesized using a weighted average of multiple frames (for example, with weights of 0.6:0.3:0.1). Images are matched based on SIFT feature points, with an error control of within 0.5 pixels. Blurred images are automatically removed, triggered by a frame difference standard deviation (σ) greater than 15%.

[0062] Data fusion involves the temporal and spatial alignment of data, including both temporal and spatial alignment. Spatial alignment unifies sensor data, HDR imagery, and route GIS data into the UTM Zone 49N coordinate system. Temporal alignment uses GPS and PPS signals as a benchmark to align light environment data and train position information, with an error of ≤1ms.

[0063] At the same time, during data fusion, a dynamic weight allocation model based on Kalman filtering is used, with the weight of environmental parameters set to 0.4, the weight of line data set to 0.3, and the weight of optical parameters set to 0.3.

[0064] S3, performing light field inversion and prediction based on the fused data. This step specifically includes the following sub-steps:

[0065] S31, at the same time Figure 3As shown in the figure, a U-Net light field inversion network is set, whose input is discrete sensor data (256-dimensional vector) + tunnel point cloud data (XYZ coordinates + reflectivity); the network structure is set to a 4-layer encoder-decoder architecture, each layer contains 2 convolution blocks (3×3 convolution + BN + ReLU), and the jump connection introduces the SE attention module to weightedly fuse low-level details and high-level semantic features; the output is set to a 512×512 pixel light field distribution map (resolution 0.25m / pixel), with a resolution of 0.25m / pixel, corresponding to a real scene of 128m×128m; the network training data set includes: 100,000 sets of tunnel crossing scene data (experimental data has been collected), 70% training set, 15% validation set, and 15% test set, the initial learning rate is set to 0.001, the batch size is set to 32, the loss function is weighted MSE, the illumination weight is set to 0.6, and the color temperature weight is set to 0.4.

[0066] S32, build an LSTM-Kalman hybrid prediction model, with the input set as the historical 10-second light intensity sequence (sampling interval 0.1s), vehicle speed, tunnel curvature, and solar altitude angle; the LSTM parameters are set to 256 neurons in the hidden layer and a time step of 10 (corresponding to the prediction of the next 5 seconds); the output is set to the light intensity gradient ΔL inside and outside the tunnel in the next 5 seconds (unit: lx / m).

[0067] Through Kalman filter correction, its state equation is:

[0068] x k =Ax k-1 +Bu k +w k

[0069] Among them, A is the state transfer matrix, B is the control matrix, u k is the vehicle speed / light intensity change rate, w k To control noise, w k ~N(0,Q),Q=0.01.

[0070] The observation equation is:

[0071] z k =Hx k +v k ,v k ~N(0,R),R=0.1

[0072] Among them, H is the observation matrix, v k is the sensor noise, Q and R are covariances.

[0073] In this embodiment, the output of the Kalman filter-corrected prediction model is the light intensity gradient ΔL (unit: lx / m) inside and outside the tunnel within the next 5 seconds, and the prediction error is ≤50lx (strong light) or ≤5lx (weak light).

[0074] The prediction results are spatially smoothed using cubic spline interpolation to eliminate mutation noise. The specific constraint conditions are: the gradient change rate of light intensity in adjacent areas is ≤ 200lx / m 2 At the same time, the prediction results can be mapped into a pseudo-color heat map (color scale 0 ~ 10^5lx) and superimposed on the high-speed train HUD display interface.

[0075] S4: Execute the control strategy based on the prediction results of S3. This step specifically includes:

[0076] The transmittance of the electrochromic glass of the driver's cab (or other compartment) window is regulated, and the control signal is based on the light intensity gradient ΔL output by the prediction model:

[0077] T target =80%-0.7×ΔL(ΔL∈[0,1000lx / m])

[0078] Among them, T target The target transmittance is 10% to 80%, and the response time is ≤ 100ms.

[0079] In this embodiment, the vehicle window adopts a zoning control strategy. For example, the vehicle window is divided into 5 zones longitudinally, and the width of each zone is set to 0.5m. The transmittance of each zone is independently controlled, and the difference in transmittance between adjacent zones is ≤15%.

[0080] Improve the light environment through supplementary lighting linkage. Specifically, the supplementary lighting device can be deployed 50m before the tunnel entrance, and a group of LED lights can be arranged every 5m, with each group having a power of 15W and an adjustable color temperature range of 3000K to 6500K. The supplementary lighting intensity formula of the LED array formed is:

[0081] I 补 =max(0,0.8·ΔL-I 环境 )

[0082] Among them, I 补 is the fill light intensity of the LED array, I 环境 is the ambient light intensity. The fill light intensity is dynamically adjusted according to the output of the prediction model, with the target gradient ΔL ≤ 150 lx / m.

[0083] As described above, the method for predicting the internal and external light environment of a high-speed train based on multi-source information fusion provided in this embodiment is based on high-density dynamic perception and precise inversion, and is specifically embodied in: using a sensor array and an HDR image acquisition unit to work together to achieve high-frequency discrete monitoring of the external light environment of the vehicle, which has the advantages of comprehensive spatial coverage, high data fidelity, and fast dynamic response; incorporating natural lighting parameters, line GIS information, and tunnel optical characteristics into a unified coordinate system and taking them into consideration improves the prediction accuracy and has strong scene adaptability; dynamically adjusting the transmittance of the electrochromic glass and the tunnel LED fill light intensity based on the light intensity gradient prediction results, forming vehicle-tunnel collaborative millisecond-level regulation, which can significantly reduce the driver's visual burden while optimizing energy consumption.

[0084] Based on the same inventive concept, this embodiment also provides a high-speed train interior and exterior light environment prediction system based on multi-source information fusion, including an acquisition module, a data processing module, a light field inversion prediction module, and a control execution module. The acquisition module includes a head sensor, a side window sensor, and an HDR image acquisition unit, which are used to achieve high-frequency discrete monitoring of the exterior light environment; the data processing module is used to filter noise from sensor data, HDR images, and other imported data, align the data in time and space, and perform dynamic weight allocation based on Kalman filtering; the light field inversion prediction module predicts the light intensity gradient inside and outside the tunnel within a certain period of time in the future through a U-Net light field inversion network and an LSTM-Kalman hybrid prediction model; the control execution module executes a control strategy based on the light intensity gradient prediction results, changes the transmittance of the electrochromic glass of the window, and controls the light filling process.

[0085] The high-speed train interior and exterior light environment prediction system based on multi-source information fusion provided by the present invention has the same inventive concept and beneficial effects as the aforementioned solution, and will not be described in detail here.

[0086] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] The above embodiments merely illustrate several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion, characterized in that: The steps include: S1, install sensors and HDR image acquisition units at preset locations on the high-speed train. The sensors include front sensors and side window sensors to obtain light intensity data and HDR images; S2, noise is filtered out of sensor data, HDR images, and imported data, and then the data is fused to complete spatiotemporal alignment; S3, based on the fused data, performs light field inversion through the U-Net light field inversion network, and predicts the light intensity gradient in the future period through the LSTM-Kalman hybrid prediction model; S4, executing a control strategy based on the prediction results, changing the light transmittance of the electrochromic glass of the vehicle window, and regulating the fill light process at the same time.

2. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: The front sensor in S1 is a multispectral sensor; the side window sensor is a polarized light sensor; the HDR image acquisition unit is electrically connected to the edge computing unit, which can obtain the bidirectional reflectance distribution function parameters of the tunnel wall through multi-frame synthesis, and at the same time extract the solar incidence angle and cloud transmission attenuation coefficient based on the HDR image.

3. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: The data imported into S2 include environmental parameters, line GIS data, and tunnel optical parameters; the environmental parameters include solar altitude angle, weather type, and surface reflectivity; the line GIS data includes tunnel location, curvature radius, and slope; and the tunnel optical parameters include buffer structure transmittance and inner wall reflection coefficient.

4. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: The sensor data in S2 uses Daubechies4 wavelet transform to filter out high-frequency noise above 50Hz and retain the effective signal frequency band of 0-50Hz; HDR images are synthesized by weighted averaging of multiple frames, and the images are matched based on SIFT feature points to automatically remove blurred images. The trigger mechanism is that the frame difference standard deviation σ>15%.

5. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: Data fusion includes time alignment and spatial alignment. Spatial alignment unifies sensor data, HDR images, and line GIS data into the same coordinate system. Time alignment uses GPS and PPS signals as a benchmark to align light environment data and train location information. During data fusion, a dynamic weight allocation model based on Kalman filtering is used, and the environmental parameter weights, line data weights, and optical parameter weights are reset to preset values.

6. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: The input of the U-Net light field inversion network in S3 is discrete sensor data + tunnel point cloud data; the network structure is set to a 4-layer encoder-decoder architecture, with each layer containing 2 convolutional blocks. The SE attention module is introduced in the jump connection to weightedly fuse low-level details and high-level semantic features; The output is set to 512×512 pixel light field distribution map with a resolution of 0.25m / pixel, corresponding to a real scene of 128m×128m; The dataset for network training includes: 100,000 sets of tunnel crossing scene data, 70% training set, 15% validation set, and 15% test set. The initial learning rate is set to 0.001, the batch size is set to 32, the loss function is weighted MSE, the illumination weight is set to 0.6, and the color temperature weight is set to 0.

4.

7. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: The input of the LSTM-Kalman hybrid prediction model in S3 is set as the historical 10-second light intensity sequence, vehicle speed, tunnel curvature, and solar altitude angle; the LSTM parameters are set to 256 neurons in the hidden layer and 10 time steps; the output is set to the light intensity gradient ΔL inside and outside the tunnel within a preset time in the future; After Kalman filter correction, the state equation is: x k =Ax k-1 +Bu k +w k Among them, A is the state transfer matrix, B is the control matrix, u k is the vehicle speed / light intensity change rate, w k To control noise; The observation equation is: z k =Hx k +v k ,v k ~N(0,R),R=0.1 Among them, H is the observation matrix, v k is the sensor noise.

8. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1 is characterized in that: When the transmittance of the electrochromic glass is regulated in S4, the control signal is based on the light intensity gradient ΔL output by the prediction model: T target =80%-0.7×ΔL(ΔL∈[0,1000lx / m]) Among them, T target is the target transmittance; The car windows adopt a zoning control strategy, with the transmittance of each zone independently controlled, and the difference in transmittance between adjacent zones ≤15%.

9. The method for predicting the interior and exterior light environment of a high-speed train based on multi-source information fusion according to claim 1, characterized in that: In S4, the LED light array is deployed at a preset distance in front of the tunnel entrance. The formula for the fill light intensity of the LED array is: I 补 =max(0,0.8·ΔL-I 环境 ) Among them, I 补 is the fill light intensity of the LED array, I 环境 is the ambient light intensity; the fill light intensity is dynamically adjusted according to the output of the prediction model.

10. A high-speed train interior and exterior light environment prediction system based on multi-source information fusion, using the method according to any one of claims 1 to 9, characterized in that: It includes acquisition module, data processing module, light field inversion prediction module and control execution module; The acquisition module includes a vehicle head sensor, a side window sensor and an HDR image acquisition unit, and is used to realize high-frequency discrete monitoring of the light environment outside the vehicle; The data processing module is used to filter noise from sensor data, HDR images and imported data, align the data in time and space, and perform dynamic weight allocation based on Kalman filtering; The light field inversion prediction module uses the U-Net light field inversion network and the LSTM-Kalman hybrid prediction model to invert and predict the light intensity gradient inside and outside the tunnel within a certain period of time in the future; The control execution module executes the control strategy based on the light intensity gradient prediction result, changes the light transmittance of the electrochromic glass of the vehicle window, and controls the fill light process at the same time.

Images