A vehicle-mounted ETC field strength dynamic detection method and system for mountainous environments

Abstract

The application discloses a vehicle-mounted ETC field strength dynamic detection method and system for mountainous areas, relates to the technical field of artificial intelligence, and comprises the following steps: collecting multi-source original data and marking; extracting multi-dimensional heterogeneous features and normalizing, and performing environment coupling feature enhancement; constructing a gated feature fusion network based on a gated recurrent unit, combining dynamic multi-path interference features and space-time continuity constraints, and completing field strength dynamic detection model training; and realizing vehicle-mounted real-time field strength detection based on the trained model. The application explicitly captures the geometric correlation between the vehicle and the antenna through composite space relation coding, introduces a dynamic multi-path interference function to simulate signal fading, and designs a composite loss function that fuses environment weighted error and space-time smooth regularization, thereby significantly improving the precision and robustness of field strength prediction in complex mountainous environments.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for dynamic detection of field strength of vehicle-mounted ETC (Electronic Toll Collection) devices in mountainous environments. Background Technology

[0002] This invention relates to the field of artificial intelligence technology, and in particular to a method for dynamic detection of field strength in vehicle-mounted ETC systems suitable for complex mountainous environments.

[0003] As my country's expressway network continues to extend into mountainous areas with complex terrain and geological conditions, the Electronic Toll Collection (ETC) system, as a core infrastructure for improving traffic efficiency, faces severe challenges in ensuring reliability and stability in these unique environments. Mountainous environments are characterized by dramatic terrain undulations, numerous ravines, and variable weather conditions, leading to severe shielding, reflection, and multipath interference on the ETC signal propagation path. This results in a highly non-uniform, nonlinear, and rapidly changing spatiotemporally signal strength distribution. Traditional static field strength testing methods rely on fixed-point measurements and manual analysis, which are not only inefficient and costly but also fail to capture the dynamic fading characteristics of signals during high-speed vehicle movement. These methods cannot meet the engineering requirements for comprehensive, real-time assessment and diagnosis of ETC coverage quality in complex road sections.

[0004] The main shortcomings of existing technologies are as follows: Conventional normalization methods process features independently for each dimension. While this unifies the units of measurement, it destroys the inherent spatial geometric relationships between positional parameters, making it difficult for the model to effectively learn the complex impact of the relative position between the vehicle and the ETC antenna on signal propagation. Conventional models typically use terrain occlusion angle and weather level as independent feature inputs, relying on the network's hidden layers to passively learn their interaction relationships. This is inefficient in mountainous scenarios with highly coupled environments, lacks guidance from physical mechanisms, resulting in poor model interpretability and insufficient generalization ability in new environments. Conventional prediction methods often treat multipath effects and time-varying fading as noise, or rely solely on implicit learning from large amounts of data. This leads to poor generalization ability in dynamic mountainous scenarios with limited data, and fails to explicitly characterize signal interference and rapid fluctuations caused by terrain reflection and vehicle movement. Conventional mean squared error loss functions only focus on point-to-point accuracy. In mountainous areas with abrupt environmental changes and drastic signal fluctuations, this can easily lead to unsmooth predicted trajectories, excessive sensitivity to outliers, and a lack of constraints on the spatial feature distribution learned by the model, easily resulting in overfitting to the geographical patterns of the training set. Summary of the Invention

[0005] To address the technical problems in the prior art, this invention provides a method and system for dynamic detection of field strength in vehicle-mounted ETC systems used in mountainous environments.

[0006] This invention is achieved through the following technical solution:

[0007] A method for dynamic detection of field strength in vehicle-mounted ETC systems used in mountainous environments, comprising:

[0008] Data acquisition and annotation; including synchronously acquiring and recording multi-source raw data at a fixed frequency. The acquisition process involves an experimental vehicle using an equipment-integrated navigation system, a wireless signal acquisition card, an on-board meteorological sensor, and an on-board computing unit pre-stored with high-precision digital elevation model data to travel through the ETC gantry area to collect data.

[0009] Data processing and feature extraction and enhancement; including feature extraction and normalization of multidimensional heterogeneous data, and enhancement of environmental coupling features;

[0010] The model for dynamic field strength detection is constructed and trained. The model adopts a gated feature fusion network with gated recurrent units, combined with dynamic multipath interference features, and is subject to spatiotemporal continuity constraints.

[0011] Vehicle-mounted ETC field strength dynamic detection module; including real-time detection based on a trained field strength dynamic detection model.

[0012] Furthermore, the multidimensional heterogeneous data feature extraction and normalization includes feature extraction and global normalization as well as spatial relationship encoding;

[0013] The feature extraction and global normalization yield a preliminary normalized feature vector;

[0014] The spatial relationship encoding specifically involves extracting a subset of location-related features from the preliminary normalized feature vector and inputting it into a composite encoder composed of three parallel encoding functions.

[0015] Furthermore, the three encoding functions are the horizontal plane geometric encoding function, the vertical profile geometric encoding function, and the line-of-sight condition encoding function, respectively.

[0016] Furthermore, the environmental coupling feature enhancement combines the terrain attenuation factor, the meteorological modulation coefficient, and the normalized horizontal distance to generate a comprehensive environmental coupling attenuation, expressed as:

[0017] ;

[0018] In the formula, Indicates the amount of environmental coupling attenuation; Represents the normalized horizontal distance; The scale parameter representing the linear decay term; The scaling parameter representing the exponentially decaying term; Indicates the terrain attenuation factor; This represents the weather modulation coefficient.

[0019] Furthermore, the dynamic multipath interference feature construction is based on the vehicle's dynamic position, terrain shading angle, and meteorological rain / fog level, and is expressed as follows:

[0020] ;

[0021] In the formula, Represents the dynamic multipath interference eigenvalues; A scale factor representing the overall intensity of multipath effects; Represents the spatial frequency parameters related to the carrier wavelength; Represents the normalized horizontal distance; Modulation coefficients representing the influence of meteorology and topography on multipath difference; Indicates the terrain attenuation factor; Indicates the meteorological rain and fog level; This represents the damping coefficient that decreases with distance; Indicates the normalized elevation difference; This represents the suppression coefficient of multipath signal strength due to terrain shielding angle; This represents the normalized terrain shading angle.

[0022] Furthermore, the prediction results output by the field strength dynamic detection model are obtained based on the smoothing term of the final prediction result at the previous moment, the basic prediction value at the current moment, and the environmental coupling attenuation.

[0023] Furthermore, the basic predicted value is obtained by concatenating the environmental enhancement feature vector and the dynamic multipath interference feature value, and inputting it into a network composed of a gated recurrent unit and a fully connected layer. The network uses the gated recurrent unit to capture the short-term dependence of the signal strength and adaptively fuses the current spatial environment features.

[0024] Furthermore, the loss function for training the field strength dynamic detection model is the sum of the main loss function and the spatial coding distribution consistency regularization loss term. The main loss function consists of the sample-level weighted absolute error and the first-order difference penalty term of the predicted values ​​at adjacent times. The spatial coding distribution consistency regularization loss term is obtained based on the variance of the set of spatial relationship coding vectors at all times within a training batch.

[0025] This invention also provides a vehicle-mounted ETC field strength dynamic detection system for mountainous environments, based on the vehicle-mounted ETC field strength dynamic detection method for mountainous environments described above, comprising:

[0026] The data acquisition and annotation module is used to synchronously acquire and record raw data from multiple sources at a fixed frequency to form an initial data stream.

[0027] The data processing and feature extraction and enhancement module is used for multidimensional heterogeneous data feature extraction and normalization, and environmental coupling feature enhancement.

[0028] The field strength dynamic detection model construction and training module is used to construct dynamic multipath interference features. It adopts a gated feature fusion network with gated recurrent units to model the signal time-series dependence for dynamic field strength detection.

[0029] The vehicle-mounted ETC field strength dynamic detection module is used for real-time detection based on a trained field strength dynamic detection model.

[0030] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing program instructions for a vehicle-mounted ETC field strength dynamic detection method for mountainous environments. The program instructions for the vehicle-mounted ETC field strength dynamic detection method for mountainous environments can be executed by one or more processors to implement the steps of the vehicle-mounted ETC field strength dynamic detection method for mountainous environments as described above.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] This invention employs a composite spatial relationship encoder, mapping basic latitude, longitude, altitude, and other location features into high-dimensional relationship codes through Gaussian radial basis functions, affine transformations, and periodic functions. This explicitly captures the complex geometric relationship between the vehicle and the ETC antenna, overcoming the limitation of traditional normalization methods that only eliminate dimensions and lose spatial relationships. Combined with an environmental coupling feature enhancement mechanism, it analytically calculates terrain attenuation factors and meteorological modulation coefficients, fusing them to generate physically meaningful dynamic attenuation features. This explicitly models the nonlinear coupling effect of terrain and meteorology, replacing the inefficient traditional method of treating them as independent features and relying on implicit learning by a network. Furthermore, it utilizes dynamic multipath interference features and integrates a gated recurrent network, simulating the interference model of direct and reflected signals through a sine function. Combined with a GRU to capture temporal dependencies, and finally applying spatiotemporal continuity constraints, it outputs smooth predictions, achieving dynamic modeling of complex multipath effects and rapid signal fading caused by vehicle motion. A composite loss function that integrates environmental weighted error, spatiotemporal smoothing regularization, and spatial coding distribution consistency is adopted. The sample weights are dynamically adjusted according to environmental decay, the predicted trajectory is forced to be smooth, and the coding feature distribution is constrained to improve the generalization ability. This overcomes the shortcomings of traditional mean square error in mountainous scenarios, which is sensitive to outliers and easily leads to prediction jumps. Attached Figure Description

[0033] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0034] Figure 1This is a schematic flowchart of a vehicle-mounted ETC field strength dynamic detection method for mountainous environments according to an embodiment of this application;

[0035] Figure 2 This is a schematic diagram of the multidimensional heterogeneous data normalization process according to an embodiment of this application;

[0036] Figure 3 This is a graph showing the robustness performance of various prediction models under different meteorological conditions;

[0037] Figure 4 This is a graph showing the relationship between the prediction error and the horizontal distance between the vehicle and the antenna of the electronic non-stop toll collection system according to an embodiment of this application;

[0038] Figure 5 This is a box plot of interval distance versus prediction error according to an embodiment of this application. Detailed Implementation

[0039] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0040] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the shape, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0042] See Figure 1 A method for dynamic detection of field strength in vehicle-mounted ETC systems used in mountainous environments includes the following steps:

[0043] S1. Data collection and annotation;

[0044] Data collection was conducted on selected sections of the complex mountainous highway ETC gantry system. Experimental vehicles equipped with a high-precision integrated navigation system (providing latitude, longitude, and altitude), a professional wireless signal acquisition card (for receiving and recording raw ETC signal strength data), onboard meteorological sensors, and an onboard computing unit pre-stored with high-precision digital elevation model data were used. As the vehicles repeatedly traveled along a predetermined route through the ETC gantry area, the system synchronously collected and recorded multi-source raw data at a fixed frequency, forming an initial data stream.

[0045] In one embodiment, the properties of the raw data stream include:

[0046] 1-Longitude, specifically referring to the longitude coordinates of the vehicle's location, measured in degrees;

[0047] 2-Latitude, specifically refers to the latitudinal coordinates of the vehicle's location, measured in degrees;

[0048] 3-Vehicle altitude, specifically refers to the altitude of the vehicle's location, measured in length and in meters;

[0049] 4-Horizontal distance, specifically refers to the horizontal straight-line distance between the vehicle and the ETC antenna, measured in length and in meters;

[0050] 5-Elevation difference, specifically refers to the altitude difference between the vehicle and the ETC antenna, measured in length and in meters;

[0051] 6-Terrain shielding angle, specifically refers to the angle that characterizes the degree of obstruction of the direct path of a signal by the terrain. The dimension is angle and the unit is degree. It is calculated by the digital elevation model.

[0052] 7-Meteorological rain and fog level, specifically refers to a discrete level characterizing the current meteorological conditions. It is a dimensionless integer with a value range of [value missing]. ;

[0053] 8 - Received signal strength indicator mean, specifically the average value of the received signal strength sequence within a sliding time window, measured in logarithms of power, with units of... ;

[0054] 9- Received signal strength indication variance, specifically the variance of the received signal strength sequence within the sliding time window, measured in units of the logarithm of power squared. ;

[0055] 10 - Received signal strength indication slope, specifically the slope of the linear fit of the received signal strength sequence within the sliding time window, measured in units of the logarithmic rate of change of power. .

[0056] In this embodiment, the longitude and latitude of the vehicle location are obtained by the GPS / BeiDou module, the vehicle altitude is calculated by the barometric altimeter with the assistance of GNSS, the horizontal straight-line distance and altitude difference are calculated in real time based on the fixed position of the vehicle and the ETC antenna, the terrain shading angle is calculated by querying the digital elevation model based on the real-time position of the vehicle, the weather rain and fog level is obtained and quantified by the vehicle-mounted meteorological sensor, and the mean, variance and linear fitting slope are obtained by performing sliding window statistical calculation on the ETC signal reception strength indication.

[0057] The data annotation process is carried out simultaneously with the data acquisition process, belonging to the regression task annotation in supervised learning. The target value (i.e., the label) is the true value of the received signal strength indication actually measured and calibrated by the on-board signal acquisition equipment at the same moment, and its dimension is... Therefore, each data sample at any given time consists of a feature vector containing the aforementioned ten attributes, and a corresponding, continuously numerical label representing the actual signal field strength.

[0058] The labeled dataset is divided into training, validation, and test sets for subsequent model building, tuning, and performance evaluation.

[0059] S2, Data Processing and Feature Extraction and Enhancement;

[0060] S201, Normalization of multidimensional heterogeneous data;

[0061] The raw data for predicting the signal field strength of vehicle-mounted ETC has multidimensional and heterogeneous characteristics, covering vehicle dynamic position, real-time environmental parameters and signal timing characteristics. These data have significant differences in physical dimensions and numerical ranges, complex spatial geometric relationships, nonlinear coupling of environmental influencing factors, and violent signal fluctuations that are strongly correlated with spatial position. Conventional normalization methods such as Min-Max or Z-Score usually process each dimension of features independently. Although they can unify the dimensions, they are difficult to preserve the spatial geometric relationship between position parameters.

[0062] This invention eliminates dimensional differences through global normalization and constructs a spatial relationship encoding mechanism to map basic location features to a high-dimensional space, thereby enhancing the model's ability to model spatial geometric relationships. Specific steps are as follows: Figure 2 As shown:

[0063] 1) Feature extraction and global normalization

[0064] Extract all numerical features from the original data stream and perform Min-Max normalization on them, linearly mapping all feature values ​​to... The interval is used to obtain the preliminary normalized feature vector, and the definition is given. This represents the initial normalized feature vector, with dimension . ,Right now ,in, This represents the total number of features in the original data stream. Indicates the first A preliminary normalized eigenvalue, , Indicates the feature index.

[0065] 2) Spatial Relationship Coding

[0066] A subset of location-related features is extracted from the initial normalized feature vector and input into a composite encoder consisting of three parallel encoding functions. The composite encoder utilizes a set of learnable spatial context anchors to transform the basic location features into a fused encoding feature vector that more effectively represents spatial relationships, as shown below:

[0067] ;

[0068] In the formula, The encoded fusion feature vector represents the comprehensive spatial relationship between the vehicle position and the ETC antenna, including horizontal geometric similarity, vertical profile nonlinear transformation, and line-of-sight condition periodic pattern.

[0069] This represents the vector concatenation operator;

[0070] This represents a subvector of location features, which includes the normalized horizontal distance and elevation difference;

[0071] Indicates the index of the encoding function;

[0072] Indicates the first The weight coefficients of each encoding function, used to adaptively balance the importance of different spatial relationship representations, are trainable parameters.

[0073] Indicates the first Each of the following encoding functions corresponds to a horizontal plane geometric encoding function. Vertical profile geometric coding function Line-of-sight conditional coding function ;

[0074] Indicates input to the first Encoding functions The geometric feature vectors, respectively, correspond to the first geometric feature vector. Second geometric feature vector Third geometric eigenvector ;

[0075] This represents the type identifier of the j-th geometric feature vector. Corresponding to the geometry of the horizontal plane, Corresponding vertical section geometry Corresponding line-of-sight condition geometry;

[0076] Represents the spatial context anchor matrix, with dimension . These are trainable parameters designed to learn typical spatial environment patterns from the overall data; the spatial context anchor matrix. The OK Corresponding to the Encoding functions The set of anchor point parameters used;

[0077] This represents the output dimension of each encoding function, i.e., the number of anchor points; the example value is 32.

[0078] In practical implementation, the location-related feature sub-vectors included in the location-related feature subset are defined as follows:

[0079] This represents a subvector of location features, which includes the normalized horizontal distance and elevation difference;

[0080] This represents the first geometric feature vector, which includes normalized longitude and latitude;

[0081] This represents the second geometric feature vector, which contains the normalized vehicle altitude and elevation difference;

[0082] This represents the third geometric feature vector, which includes the normalized horizontal distance and the terrain occlusion angle.

[0083] In one implementation, the horizontal plane geometric coding function uses a Gaussian radial basis function, and the output vector represents the similarity between the input horizontal coordinates and a set of anchor points, expressed as:

[0084]

[0085] In the formula, This represents a hyperparameter that controls the width of the radial basis function, affecting the local sensitivity of the feature; an example value is 10.0.

[0086] Represents the natural exponential function;

[0087] The width hyperparameter represents the Gaussian radial basis function, which controls the local sensitivity of the basis function. The larger the value, the narrower the function and the more sensitive it is to changes in distance.

[0088] Representation spatial context anchor matrix The first row corresponds to the first Anchor vectors, with dimensions and The same parameter is a trainable parameter;

[0089] This represents the L2 norm, i.e., the Euclidean distance;

[0090] Indicates by A vector consisting of scalar elements.

[0091] In this embodiment, the vertical profile geometric coding function captures the nonlinear effects of altitude and elevation difference through affine transformation and nonlinear activation, and is expressed as follows:

[0092]

[0093] In the formula, This represents the scaling factor, used to control the range of values ​​after transformation; an example value is 2.0.

[0094] Representation spatial context anchor matrix The second line corresponds to the first A weight vector, express The transpose of is a trainable parameter;

[0095] Indicates the relationship with the first The bias parameters corresponding to each output unit are trainable parameters;

[0096] This means taking the larger of the two parameters.

[0097] In this embodiment, the line-of-sight conditional coding function combines horizontal distance and terrain obstruction angle to simulate the influence of terrain obstruction on signal line-of-sight propagation, as expressed as:

[0098]

[0099] In the formula, This represents a sine function, used to generate periodic features that are strongly correlated with angles;

[0100] Representation spatial context anchor matrix The third row in the middle corresponds to the first Anchor vectors, express The transpose of is a trainable parameter.

[0101] It should be noted that, Items are used for calculation and The cosine similarity reflects the degree of proximity in direction between two objects, mapping directional similarity to... The interval is then converted into periodic features using a sine function, thereby capturing the angle-related patterns in line-of-sight propagation and enhancing the model's ability to model the impact of terrain obstruction.

[0102] S202, Enhanced environmental coupling characteristics;

[0103] In complex mountainous environments, the attenuation of ETC signal strength is affected by the nonlinear coupling of terrain shading and meteorological conditions. Conventional techniques usually use terrain shading angle and meteorological level as two independent features to input into the neural network, relying on the hidden layer of the network to implicitly learn their complex interaction relationship. However, this is inefficient when the environmental coupling is extremely strong, and it is difficult to accurately capture the physical mechanism, resulting in insufficient generalization ability and interpretability.

[0104] This invention constructs dynamic terrain attenuation characteristics and analytically calculates the additional signal attenuation component generated by the coupling effect between terrain and meteorology, providing strong physical priors for the model. The specific steps are as follows:

[0105] 1) Calculation of terrain attenuation factor

[0106] The terrain attenuation factor, based on the calculation of the normalized terrain shading angle, is used to quantify the attenuation intensity of terrain shading on signal propagation, and is expressed as:

[0107]

[0108] In the formula, The terrain attenuation factor is a dimensionless scalar with a range of [value range missing]. Within the range, the smaller the value, the stronger the signal attenuation caused by terrain.

[0109] Represents the normalized terrain shading angle, with values ​​ranging from The interval represents the degree to which the connection between the vehicle and the ETC antenna is obstructed by the terrain;

[0110] This represents the arctangent function, used to map the terrain shading angle to a smooth attenuation factor;

[0111] This represents the reference angle hyperparameter, used to adjust the sensitivity of the attenuation curve to changes in angle. An example value is 0.167.

[0112] It should be noted that the normalized terrain shielding angle Corresponding to actual angle arrive ,Pick Can make Item in It has moderate sensitivity to changes, avoiding an excessively steep or gradual decay factor, thereby balancing the model's response to terrain changes.

[0113] 2) Calculation of meteorological modulation coefficient

[0114] The meteorological modulation coefficient, calculated based on meteorological rain and fog levels, characterizes the aggravating effect of meteorological conditions on signal attenuation and is expressed as follows:

[0115]

[0116] In the formula, This represents the meteorological modulation coefficient, which is a dimensionless scalar with a value greater than or equal to 1. The larger the value, the stronger the modulation effect of meteorological conditions on attenuation.

[0117] This represents the meteorological rain / fog level, and is a discrete integer value. The example value range is... ;

[0118] Represent the natural logarithm function;

[0119] This represents a hyperparameter that controls the intensity of meteorological impacts. It is used to adjust the increase in the meteorological modulation coefficient. An example value is 0.1.

[0120] 3) Generation of dynamic decay characteristics

[0121] By combining the terrain attenuation factor, meteorological modulation coefficient, and normalized horizontal distance, and through a function fusing linear and exponential attenuation forms, a comprehensive environmental coupling attenuation is generated, expressed as:

[0122]

[0123] In the formula, This represents the environmental coupling attenuation, in units of... This characterizes the additional signal attenuation caused by the nonlinear coupling effect of terrain shading and meteorological conditions;

[0124] Represents the normalized horizontal distance, used as a dependency variable for attenuation, to simulate the phenomenon that near-field attenuation decreases rapidly with distance;

[0125] The scale parameter representing the linear decay term is a trainable parameter used to control the contribution of terrain and meteorological coupling to the distance-related decay.

[0126] The scaling parameter, representing the exponential decay term, is a trainable parameter used to control the intensity of near-field diffraction or reflection signal attenuation.

[0127] It should be noted that, The simulation shows an approximately linear additional decay with increasing distance due to the combined effects of terrain and weather. The simulation considers possible diffraction or reflection signals in the near-field region, whose attenuation decreases exponentially with distance and is modulated by the terrain attenuation factor.

[0128] 4) Feature splicing and fusion

[0129] The encoded fusion feature vector is concatenated with the environmental coupling attenuation factor to form the environmental enhancement feature vector, represented as:

[0130]

[0131] In the formula, The environmental enhancement feature vector represents the comprehensive features that integrate spatial geometric relationships and environmental attenuation information. It introduces environmental coupling effects to improve the model's generalization ability and interpretability in complex mountainous environments.

[0132] This represents the vector concatenation operator.

[0133] S3, Construction and Training of Vehicle-Mounted ETC Field Strength Dynamic Detection Model

[0134] In complex mountainous dynamic scenarios, in addition to being affected by the factors mentioned above, the ETC signal strength is also significantly affected by the multipath effect caused by reflection and diffraction from complex terrain, as well as the rapid signal fading caused by high-speed vehicle movement. Conventional prediction models usually regard multipath and time-varying fading as noise or rely on implicit learning based on large-scale data, resulting in poor generalization ability when data is limited or the scene changes abruptly.

[0135] This invention employs a vehicle-mounted ETC field strength prediction method based on modeling dynamic multipath interference and spatiotemporal continuity constraints. It constructs dynamic multipath interference features to characterize the multipath superposition effect of the signal and uses a gated feature fusion network with gated recurrent units to model the signal's temporal dependence. The specific steps are as follows:

[0136] S301, Dynamic Multipath Interference Feature Construction

[0137] Based on the vehicle's dynamic position (including horizontal distance and elevation difference), terrain shading angle, and meteorological rain / fog level, a composite feature is constructed to simulate signal intensity fluctuations caused by dynamic multipath interference. This feature simulates the interference model between a direct signal and a main reflected signal, and its intensity attenuation and phase difference dynamically change with distance, terrain, and weather conditions, as expressed below:

[0138]

[0139] In the formula, Represents the dynamic multipath interference eigenvalue, in units of It can be positive or negative, and represents the fluctuation of signal strength relative to the free space propagation model caused by multipath effects;

[0140] The scaling factor, which represents the overall intensity of the multipath effect, is a trainable parameter used to control the overall amplitude of the dynamic multipath interference feature so that it adapts to the actual signal fluctuation range.

[0141] This represents a sine function used to simulate periodic interference phenomena caused by path differences;

[0142] It represents the spatial frequency parameter related to the carrier wavelength, which is a trainable parameter that controls the density of the interference fringes;

[0143] The modulation coefficients, representing the influence of weather and topography on multipath path difference, are trainable parameters used to adjust how weather and fog levels and topographic attenuation factors affect path difference, thereby changing the density and phase of interference fringes.

[0144] The damping coefficient, which represents the attenuation with distance, is a trainable parameter that simulates the attenuation of the reflected signal intensity with the total propagation distance.

[0145] This represents the normalized elevation difference, and ;

[0146] Represent the natural logarithm function;

[0147] It represents the suppression coefficient of the terrain occlusion angle on the multipath signal strength (especially the reflection path). It is a trainable parameter. The more severe the terrain occlusion, the weaker the reflection path signal.

[0148] It should be noted that, when simulating interference using a sine function, the path difference between the direct and reflected signals leads to a phase difference, which in turn causes the signal intensities to periodically superimpose or cancel each other out. The item reflects a poor basic path. Modulating the effects of weather and topography on path difference allows the interferometric model to change dynamically with the environment.

[0149] S302, Gated Feature Fusion Network Prediction

[0150] The environmental enhancement feature vector and the dynamic multipath interference feature value are concatenated and input into a network composed of a gated recurrent unit and a fully connected layer to predict the basic value of the ETC signal field strength at the current moment. This network uses the gated recurrent unit to capture the short-term time dependence of the signal strength (such as the trend caused by vehicle motion) and adaptively fuses the current spatial environment features, as shown below:

[0151]

[0152] In the formula, This represents the baseline predicted value of the ETC received signal strength indicator at time t, output by the network, in units of... This is the initial predicted signal strength without smoothing.

[0153] This represents the hidden state vector of the gated recurrent unit at time t, used to memorize temporal information. Its calculation method is expressed as follows: ;

[0154] This represents the weight matrix of the output fully connected layer, which are trainable parameters;

[0155] This represents the bias vector of the output fully connected layer, which are trainable parameters;

[0156] t represents the time index;

[0157] This represents the environmental enhancement feature vector at time t;

[0158] The dynamic multipath interference eigenvalue represents the value at time t.

[0159] This represents a gated recurrent unit, a recurrent neural network structure used to capture temporal dependencies;

[0160] Indicates the time of the gated loop unit. The hidden state vector.

[0161] S303, Spatiotemporal Continuity Constraints and Final Prediction

[0162] Considering the continuity of the vehicle's trajectory, the vehicle's position and signal strength should not change drastically between adjacent moments. A smoothing term based on the final prediction result from the previous moment is used to constrain the base prediction value at the current moment, generating a more physically meaningful final prediction value, expressed as:

[0163]

[0164] In the formula, This represents the final predicted value of the ETC signal strength at time t, in units of... It is a predicted signal strength with a more reasonable physical meaning after smoothing constraints;

[0165] The spatiotemporal continuity smoothing coefficient is a hyperparameter set based on the average vehicle speed. The faster the speed, Generally, the smaller the value, the better; for example, a value of 0.2.

[0166] Indicates time The final predicted value of the ETC signal field strength can be initialized with the actual value or the basic predicted value at the start of inference.

[0167] This represents the environmental coupling decay at time t, used to ensure that the predicted value has included the environmental decay effect, thus enhancing the interpretability of the model.

[0168] S304. Calculate the loss function

[0169] Conventional mean squared error loss only focuses on point-to-point accuracy, which can easily lead to unsmooth predicted trajectories or sensitivity to outliers in complex mountainous scenarios.

[0170] This invention employs a loss function that integrates weighted absolute error, spatiotemporal smoothing regularization, and spatial coding distribution consistency. It dynamically weights the error importance of different samples through environmental attenuation, forces spatiotemporal smoothness by penalizing abrupt changes in adjacent predicted values, and improves the model's generalization ability in unseen geographical areas by constraining the distribution of spatial coding features. The specific steps are as follows:

[0171] 1) Calculate the main loss function

[0172] The main loss function consists of sample-level weighted absolute error and a first-order difference penalty term for adjacent time-time predicted values. The weighted absolute error term ensures prediction accuracy, while the first-order difference penalty term for adjacent time-time predicted values ​​enforces time smoothness, preventing drastic jumps in predicted values ​​due to environmental changes or noise, thereby improving trajectory continuity. It is expressed as:

[0173]

[0174] In the formula, This represents the main loss function, used to measure the error between the predicted and actual values, and includes a spatiotemporal smoothing regularization term to balance accuracy and smoothness.

[0175] This represents the total number of time steps in the training batch;

[0176] The actual ETC signal field strength value at time t is obtained through actual measurement by the on-board equipment and is used as a training label.

[0177] The sample error weight at time t is used to dynamically adjust the importance of the sample in the loss based on the severity of environmental degradation. The calculation method is expressed as follows: ;

[0178] This represents the hyperbolic tangent activation function, used to introduce nonlinearity;

[0179] This represents the weighting factor, used to control the sensitivity of sample error weights to environmental decay. The larger the value, the higher the weight of samples with severe environmental degradation in the loss; an example value is 0.5.

[0180] This represents the temperature coefficient hyperparameter, used to adjust the sensitivity of the weights to the decay amount; an example value is 5.0.

[0181] This represents the weighting coefficient of the spatiotemporal smoothing regularization term, used to balance prediction accuracy and smoothness. An example value is 0.01.

[0182] 2) Calculate the spatial coding distribution consistency regularization loss term

[0183] To improve the model's ability to generalize to unknown spatial locations and avoid overfitting to the geographical patterns of the training set, this loss term encourages the model to learn spatial relationship encodings. Maintaining a certain degree of distributional consistency within batches (e.g., the variance should not be too large) allows for the learning of more general rather than specific spatial mappings, represented as:

[0184]

[0185] In the formula, The spatial encoding distribution consistency regularization loss term is used to encourage the spatial relationship encodings learned by the model to be compactly distributed within a batch, avoiding overfitting to the geographical patterns of the training set, thereby improving the model's generalization ability in unseen spatial locations.

[0186] This represents the loss weight coefficient, which is the weight coefficient of the spatial coding distribution consistency regularization loss term. An example value is 0.1.

[0187] This indicates the calculation of the variance of the set;

[0188] This represents the spatial relation encoding vector across all time points within a training batch. The set that constitutes;

[0189] This represents the upper bound threshold of the expected coding distribution variance. It is a hyperparameter used to control the compactness of the feature distribution, with an example value of 0.1.

[0190] 3) Calculate the total loss function

[0191] Total loss function of model training The sum of the principal loss function and the spatial coding distribution consistency regularization loss term, i.e. .

[0192] In one embodiment, such as Figure 3 As shown, this paper analyzes the robustness of various prediction models under different meteorological conditions, especially the stability of model performance under severe weather conditions. This experiment simulates five typical mountainous meteorological conditions: sunny, light rain, moderate rain, heavy rain, and fog, comparing the proposed technique with three conventional techniques: linear regression, random forest, and neural networks. The bar chart visually displays the mean absolute error of each model under different meteorological conditions. All models perform best under sunny conditions, and the error gradually increases as weather conditions worsen. The proposed technique maintains the lowest prediction error under all meteorological conditions, and the red bars of the proposed technique are at their lowest positions under all meteorological conditions, proving that the meteorological modulation coefficient and dynamic attenuation feature construction module introduced can effectively capture the influence mechanism of meteorological factors on signal propagation.

[0193] S305, vehicle-mounted ETC field strength dynamic detection model training

[0194] The training of the vehicle-mounted ETC field strength dynamic detection model is based on an iterative optimization process of gradient descent. The training uses the labeled dataset prepared in the S1 stage, calculates the predicted value through forward propagation, and then updates all trainable parameters in the model through the backpropagation algorithm.

[0195] Specifically, in each training batch, a batch of sample data arranged in chronological order is input into the model and processed sequentially through three core modules: multidimensional heterogeneous data normalization, environmental coupling feature enhancement, and vehicle-mounted ETC field strength prediction. Finally, the signal field strength prediction sequence at the corresponding time is output.

[0196] Then, the total loss function value between the predicted sequence and the real label sequence of the batch is calculated. This total loss function integrates the sample-level weighted absolute error, the spatiotemporal smoothing regularization term, and the spatial coding distribution consistency regularization term.

[0197] The optimizer calculates the gradient of the model parameters based on the total loss function and performs parameter updates so that the model predictions continuously approximate the actual signal intensity variation.

[0198] The training process is usually monitored on the validation set. The criteria for stopping iteration are mainly based on the loss function value or evaluation metric (such as mean absolute error) on the validation set no longer decreasing, i.e., reaching the early stopping point, or the preset maximum number of training iterations has been reached, so as to ensure that the model learns fully and avoids overfitting.

[0199] In one embodiment, such as Figure 4 As shown, this paper analyzes the relationship between prediction error and the horizontal distance between the vehicle and the antenna of the electronic toll collection system, revealing the distribution characteristics of the error within different distance ranges. A combination of scatter plots and box plots is used to display the detailed characteristics of the error distribution. The horizontal axis of the scatter plot represents the horizontal distance between the vehicle and the antenna in meters, and the vertical axis represents the prediction error in decibels and milliwatts. The scatter plot distribution shows that error points are relatively dense and small in the near-distance region, but gradually disperse and increase in value as the distance increases, consistent with the physical law that signal attenuation and increased uncertainty with distance.

[0200] like Figure 5 As shown in this embodiment, the box plot divides the distance into four intervals, displaying the statistical distribution characteristics of the error within each interval, including the median, interquartile range, and outlier range. The technology of this invention exhibits a compact error distribution across all distance intervals, verifying the effectiveness of the horizontal distance normalization and spatial relationship encoding modules in this invention, and enabling better modeling of complex patterns of signal strength variation with distance.

[0201] S4, Vehicle-mounted ETC Field Strength Dynamic Detection

[0202] Once the vehicle-mounted ETC field strength dynamic detection model has been trained and meets the performance requirements, it can be deployed on the actual vehicle-mounted computing unit for real-time dynamic detection in complex mountainous environments.

[0203] During the detection phase, the vehicle-mounted system continuously collects multi-source raw data streams in real time while driving. These real-time data first undergo a preprocessing process that is completely consistent with the training phase, and then are input into the pre-trained solidified model. The model then sequentially performs steps such as multi-dimensional heterogeneous data normalization, environmental coupling feature enhancement, and vehicle-mounted ETC field strength prediction.

[0204] Specifically, the model first performs global normalization and spatial relation encoding on the real-time input features, then calculates the coupling attenuation in the current environment and constructs an environment enhancement feature vector. At the same time, it constructs dynamic multipath interference features by combining real-time location and environmental information. Finally, it combines historical hidden states through a gated feature fusion network and applies spatiotemporal continuity constraints to output the final predicted value of the ETC signal strength at the current moment. This predicted value serves as a dynamic assessment of the signal coverage quality at the current location and can be displayed, recorded, or used to trigger relevant early warning and optimization mechanisms, thereby realizing the vehicle-mounted, highly dynamic ETC signal strength detection function.

[0205] In this embodiment, a composite spatial relationship encoder is used to map the basic latitude, longitude, altitude and other location features into a high-dimensional relationship code through Gaussian radial basis, affine transformation and periodic function mapping. This explicitly captures the complex geometric relationship between the vehicle and the ETC antenna, overcoming the limitation of traditional normalization methods that only eliminate dimensions and lose spatial relationships.

[0206] An environmental coupling feature enhancement mechanism is adopted. The terrain attenuation factor and meteorological modulation coefficient are calculated analytically and fused to generate a dynamically attenuation feature with physical meaning. This explicitly models the nonlinear coupling effect between terrain and meteorology, replacing the inefficient traditional method of treating them as independent features and relying on implicit learning by the network.

[0207] By employing dynamic multipath interference features and integrating a gated recurrent network, a sinusoidal function is used to simulate the interference model of direct and reflected signals. Combined with GRU to capture temporal dependencies, and finally applying spatiotemporal continuity constraints to output smooth predictions, dynamic modeling of complex multipath effects and rapid signal fading caused by vehicle motion is achieved.

[0208] A composite loss function that integrates environmental weighted error, spatiotemporal smoothing regularization, and spatial coding distribution consistency is adopted. The sample weights are dynamically adjusted according to environmental decay, the predicted trajectory is forced to be smooth, and the coding feature distribution is constrained to improve the generalization ability. This overcomes the shortcomings of traditional mean square error in mountainous scenarios, which is sensitive to outliers and easily leads to prediction jumps.

[0209] This invention also proposes a vehicle-mounted ETC field strength dynamic detection system for mountainous environments, based on the method described above, including:

[0210] The data acquisition and annotation module is used to synchronously acquire and record raw data from multiple sources at a fixed frequency to form an initial data stream.

[0211] The data processing and feature extraction and enhancement module is used for multidimensional heterogeneous data feature extraction and normalization, and environmental coupling feature enhancement.

[0212] The field strength dynamic detection model construction and training module is used to construct dynamic multipath interference features. It adopts a gated feature fusion network with gated recurrent units to model the signal time-series dependence for dynamic field strength detection.

[0213] The vehicle-mounted ETC field strength dynamic detection module is used for real-time detection based on a trained field strength dynamic detection model.

[0214] Furthermore, this embodiment of the invention also proposes a computer-readable storage medium storing program instructions for a vehicle-mounted ETC field strength dynamic detection method for mountainous environments. The program instructions for the vehicle-mounted ETC field strength dynamic detection method for mountainous environments can be executed by one or more processors to implement the steps of the vehicle-mounted ETC field strength dynamic detection method for mountainous environments as described above.

[0215] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for dynamic detection of field strength of vehicle-mounted ETC (Electronic Toll Collection) devices in mountainous environments, characterized in that, include: Data collection and annotation; The experiment involved synchronously collecting and recording multi-source raw data at a fixed frequency. The data collection process included an experimental vehicle that used an equipment-integrated navigation system, a wireless signal acquisition card, an onboard meteorological sensor, and an onboard computing unit that pre-stored high-precision digital elevation model data to collect data as it drove through the ETC gantry area. Data processing, feature extraction, and enhancement; This includes feature extraction and normalization of multidimensional heterogeneous data, and enhancement of environmental coupling features; Construction and training of dynamic field strength detection model; The model employs a gated feature fusion network with gated recurrent units, combined with dynamic multipath interference features, and subject to spatiotemporal continuity constraints. Vehicle-mounted ETC field strength dynamic detection; including real-time detection based on a trained field strength dynamic detection model; The environmental coupling feature enhancement combines the terrain attenuation factor, the meteorological modulation coefficient, and the normalized horizontal distance to generate a comprehensive environmental coupling attenuation, expressed as: ； In the formula, Indicates the amount of environmental coupling attenuation; Represents the normalized horizontal distance; The scale parameter representing the linear decay term; The scaling parameter representing the exponentially decaying term; Indicates the terrain attenuation factor; Indicates the weather modulation coefficient; The dynamic multipath interference feature is constructed based on the vehicle's dynamic position, terrain shading angle, and meteorological rain / fog level, and is expressed as follows: ； In the formula, Represents the dynamic multipath interference eigenvalues; A scale factor representing the overall intensity of multipath effects; Represents the spatial frequency parameters related to the carrier wavelength; Represents the normalized horizontal distance; Modulation coefficients representing the influence of meteorology and topography on multipath difference; Indicates the terrain attenuation factor; Indicates the meteorological rain and fog level; This represents the damping coefficient that decreases with distance; Indicates the normalized elevation difference; This represents the suppression coefficient of multipath signal strength due to terrain shielding angle; Indicates the normalized terrain shielding angle; The prediction results output by the field strength dynamic detection model are obtained based on the smoothing term of the final prediction result at the previous moment, the basic prediction value at the current moment, and the environmental coupling attenuation. The basic predicted value is obtained by concatenating the environmental enhancement feature vector and the dynamic multipath interference feature value, and inputting it into the network composed of gated recurrent units and fully connected layers stacked together; The network utilizes gated cyclic units to capture short-term dependencies in signal strength and adaptively fuses current spatial environment features. The loss function for training the field strength dynamic detection model is the sum of the main loss function and the spatial coding distribution consistency regularization loss term. The main loss function consists of the sample-level weighted absolute error and the first-order difference penalty term of the predicted values ​​at adjacent time steps. The spatial coding distribution consistency regularization loss term is obtained based on the variance of the set of spatial relationship coding vectors at all time steps within a training batch.

2. The method for dynamic detection of field strength of vehicle-mounted ETC in mountainous environments according to claim 1, characterized in that, The multidimensional heterogeneous data feature extraction and normalization includes feature extraction, global normalization, and spatial relationship encoding; The feature extraction and global normalization yield a preliminary normalized feature vector; The spatial relationship encoding specifically involves extracting a subset of location-related features from the preliminary normalized feature vector and inputting it into a composite encoder composed of three parallel encoding functions.

3. The method for dynamic detection of field strength of vehicle-mounted ETC in mountainous environments according to claim 2, characterized in that, The three encoding functions are the horizontal plane geometric encoding function, the vertical profile geometric encoding function, and the line-of-sight condition encoding function.

4. A vehicle-mounted ETC field strength dynamic detection system for mountainous environments, based on the vehicle-mounted ETC field strength dynamic detection method for mountainous environments as described in any one of claims 1 to 3, characterized in that, include: The data acquisition and annotation module is used to synchronously acquire and record raw data from multiple sources at a fixed frequency to form an initial data stream. The data processing and feature extraction and enhancement module is used for multidimensional heterogeneous data feature extraction and normalization, and environmental coupling feature enhancement. The field strength dynamic detection model construction and training module is used to construct dynamic multipath interference features. It adopts a gated feature fusion network with gated recurrent units to model the signal time-series dependence for dynamic field strength detection. The vehicle-mounted ETC field strength dynamic detection module is used for real-time detection based on a trained field strength dynamic detection model.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions for a vehicle-mounted ETC field strength dynamic detection method for mountainous environments. The program instructions for the vehicle-mounted ETC field strength dynamic detection method for mountainous environments can be executed by one or more processors to implement the steps of the vehicle-mounted ETC field strength dynamic detection method for mountainous environments as described in any one of claims 1 to 3.

Images