Highway pavement temperature inversion method and system fusing weather and traffic information

Abstract

The application discloses a highway pavement temperature inversion method and system fusing meteorological and traffic information, and the method comprises the following steps: acquiring meteorological information and traffic information of a target highway section, performing time synchronization, space matching and preprocessing on multi-source data, and constructing an inversion input dataset; performing feature extraction on meteorological information, road category information and traffic time series information based on a multi-input branch neural network, comprehensively processing in a fusion layer, and establishing a pavement temperature inversion model; introducing a physical loss function with embedded energy balance constraints in the model training process to enhance the physical consistency of the inversion result; and using the trained model to perform inversion calculation on the pavement temperature of the target section at the corresponding time. The method can comprehensively consider the differences between meteorological conditions and road operating environments, improve the accuracy and stability of pavement temperature inversion, and is suitable for highway pavement temperature monitoring and traffic safety guarantee.

Description

Technical Field

[0001] This invention relates to a method for retrieving highway pavement temperature, and more particularly to a method and system for retrieving highway pavement temperature by integrating meteorological and traffic information. Background Technology

[0002] Road surface temperature is a crucial environmental factor affecting highway traffic safety. Especially under conditions of low temperatures, rain, snow, and day-night cycles, even minute changes in road surface temperature can directly lead to icing, snow accumulation, or slippery surfaces, thus causing traffic accidents. Currently, highway road surface temperature acquisition mainly relies on point-based road surface temperature sensors or simple model inversion methods that only consider a few meteorological factors such as air temperature and radiation. These methods have limited accuracy under complex weather conditions and diverse road structures. In reality, radiation environment, ventilation conditions, road surface heat exchange characteristics, and traffic flow all have direct or indirect impacts on the road surface thermal environment, and existing technologies lack a road surface temperature inversion method that simultaneously integrates meteorological and traffic information. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a method and system for retrieving highway pavement temperature by integrating meteorological and traffic information. By comprehensively considering meteorological factors and traffic information, a pavement temperature retrieval model is established to achieve refined and dynamic retrieval of highway pavement temperature.

[0004] Technical solution: The method described in this invention includes the following steps:

[0005] Obtain meteorological information corresponding to the target highway section, including temperature, humidity, wind speed, net radiation, and precipitation information;

[0006] Obtain traffic information corresponding to the target highway segment. The traffic information includes road category information that characterizes the road structure characteristics and traffic time sequence information that characterizes the traffic operation pattern.

[0007] Meteorological and traffic information is synchronized in time, matched spatially, and subjected to quality control. Normalization preprocessing is performed to construct a unified inversion input dataset for model training.

[0008] A highway pavement temperature inversion model based on a multi-input branch neural network is constructed for different data types. A physical loss function with embedded energy balance constraints is designed. Based on the inversion input dataset, deep learning is used to fuse preprocessed meteorological information and traffic information to train the inversion model. The multi-input branch neural network structure includes a meteorological information branch, a road category information branch, a traffic time series information branch, a feature fusion layer, and an output layer. Meteorological information is extracted nonlinearly through a feedforward neural network to characterize the complex nonlinear relationship between meteorological elements and pavement temperature. Road category information is transformed into a continuous feature vector through embedding mapping to express the implicit differences in thermal environment of different road structures. Traffic time series information is processed through a time series neural network to extract the operational features that evolve over time. The outputs of each branch are combined in the feature fusion layer to form a comprehensive feature representation, and the pavement temperature inversion result is obtained through the output layer.

[0009] Using a trained highway pavement temperature inversion model, the pavement temperature is inverted and calculated based on meteorological and traffic information for the target time and target highway segment, and the pavement temperature inversion estimation results for the corresponding time and segment are obtained.

[0010] Furthermore, a meteorological element input vector is constructed based on meteorological information, represented as:

[0011] ;

[0012] in, and These are the road segment index and the time index, respectively. The input vector for the meteorological elements at this index. , , , and These are the temperature, humidity, wind speed, net radiation, and precipitation at that index location, respectively.

[0013] Furthermore, road category information is represented using discrete variables as follows:

[0014] ;

[0015] in, For road segment indexing, The road category label at this index. Total number of road categories;

[0016] Traffic time-series information is constructed as a time feature vector, represented as:

[0017] ;

[0018] in, For time indexing, This is the traffic time-series feature vector at this index. For the hour corresponding to this index, Encode the date type corresponding to this index. and These represent the sine and cosine functions, respectively.

[0019] Furthermore, the meteorological and traffic information is synchronized in time, spatially matched, and quality controlled, and normalized preprocessing is performed to construct a unified inversion input dataset; including:

[0020] By using time synchronization processing, data with different time resolutions are uniformly mapped onto the same time axis;

[0021] Spatial matching processing is used to ensure that meteorological information is matched one-to-one with the corresponding highway sections;

[0022] Outliers are eliminated through quality control.

[0023] Normalize continuous input variables to bring variables with different dimensions and ranges to the same scale.

[0024] After the above processing, meteorological information, road category information, traffic time series information, and corresponding road surface temperature observations are combined to form a unified inversion input dataset.

[0025] Furthermore, the meteorological information is processed through a feedforward neural network to extract its effective features, represented as follows:

[0026] ;

[0027] in, This represents the feature vector output by the meteorological information branch. This represents a feedforward neural network representing a branch of meteorological information. For meteorological information input, The parameter set for the meteorological information branch;

[0028] Road category information is mapped through an embedding layer, transforming discrete category information into a continuous vector representation, in the form of:

[0029] ;

[0030] in, It is the embedding vector corresponding to the road category information. It is an embedding matrix. Road category information The corresponding one-hot encoding;

[0031] Traffic temporal information is processed through a long short-term memory network to capture the evolution of traffic changes over time, represented as:

[0032] ;

[0033] in, It is the feature vector output by the traffic time-series information branch. It is a time feature extraction function for Long Short-Term Memory (LSTM) networks. From time arrive Input of traffic timing information It is the parameter set for the timing branch;

[0034] The output features of all branch networks are fused together by concatenation or weighted summation to obtain a comprehensive feature vector. :

[0035] ;

[0036] fused feature vector The input is sent to an output layer, and the final output is a road surface temperature inversion estimate. :

[0037] ;

[0038] in, It is the mapping function of the output layer. It is the set of parameters for the output layer.

[0039] Furthermore, the physical loss function with embedded energy balance constraints Represented as:

[0040] ;

[0041] in, The data fitting loss represents the road surface temperature inversion estimation. Compared with the measured road surface temperature Mean squared error within a training batch; These are the physical constraint weighting coefficients. For physical constraint loss, it represents The average of the absolute values ​​of all samples within a training batch; The energy balance residual is expressed as:

[0042] ;

[0043] in, Net radiation, For sensible heat flux, For geothermal flux;

[0044] ;

[0045] ;

[0046] in, air density, The specific heat of air at constant pressure. For road surface temperature inversion estimation, and These are the road segment index and the time index, respectively. The temperature at that index. For turbulent thermal resistance, The thermal conductivity is an equivalent parameter. This refers to the temperature of the lower layer of the road surface.

[0047] In another embodiment of the present invention, a highway pavement temperature retrieval system integrating meteorological and traffic information is also provided, comprising:

[0048] The meteorological information acquisition unit is used to acquire meteorological information corresponding to the target highway section. The meteorological information includes temperature, humidity, wind speed, net radiation, and precipitation information.

[0049] The traffic information acquisition unit is used to acquire traffic information corresponding to the target highway segment. The traffic information includes road category information that characterizes the road structure characteristics and traffic time sequence information that characterizes the traffic operation pattern.

[0050] The dataset construction unit is used to synchronize meteorological and traffic information in time, match spatially, and control quality, and to perform normalization preprocessing to build a unified inversion input dataset for model training.

[0051] The model building and training unit is used to construct highway pavement temperature inversion models based on multi-input branch neural networks for different data types. A physical loss function with embedded energy balance constraints is designed, and deep learning is used to fuse preprocessed meteorological and traffic information to train the inversion model. The multi-input branch neural network structure includes a meteorological information branch, a road category information branch, a traffic time-series information branch, a feature fusion layer, and an output layer. Meteorological information is extracted nonlinearly through a feedforward neural network to characterize the complex nonlinear relationship between meteorological elements and pavement temperature. Road category information is transformed into continuous feature vectors through embedding mapping to express the implicit differences in thermal environment among different road structures. Traffic time-series information is processed through a time-series neural network to extract operational characteristics that evolve over time. The outputs of each branch are combined in the feature fusion layer to form a comprehensive feature representation, and the pavement temperature inversion result is obtained through the output layer.

[0052] The inversion calculation unit is used to perform inversion calculations on the pavement temperature based on meteorological and traffic information for the target time and target highway segment using a trained highway pavement temperature inversion model, and obtain the pavement temperature inversion estimation results for the corresponding time and segment.

[0053] In another embodiment of the present invention, an electronic device is also provided, comprising:

[0054] Memory, used to store computer programs;

[0055] A processor for executing the computer program to implement the method.

[0056] In another embodiment of the present invention, a non-volatile storage medium is also provided for storing a computer program, wherein the computer program implements the method described thereon when executed by a processor.

[0057] In another embodiment of the present invention, a computer program product is also provided, including a computer program / instructions that, when executed by a processor, implement the method described herein.

[0058] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are as follows: (1) In the process of road surface temperature inversion, the present invention not only introduces meteorological information such as air temperature, humidity, wind speed, and net radiation, but also integrates road category information that characterizes road structure characteristics and traffic time series information that characterizes traffic operation patterns, so that the inversion model can simultaneously consider external meteorological forcing and differences in road operation environment, overcoming the problem of insufficient applicability caused by relying only on a single or a small number of meteorological elements in the prior art; (2) In view of the differences in data type and statistical characteristics of meteorological information, road category information and traffic time series information, the present invention constructs a multi-input branch neural network model for different data types, extracts features of different types of information respectively and fuses them at a high level, avoiding the problem of insufficient feature expression caused by simple splicing of heterogeneous data in the prior art, and improving the model's utilization efficiency of multi-source information; (3) The present invention embeds a physical loss function of energy balance constraint in the model training process, so that the model output results meet the data fitting accuracy while conforming to the physical laws of road surface heat balance, effectively reducing the risk of non-physical results generated by pure data-driven methods under extreme weather or sample sparse conditions, and improving the stability and reliability of the model. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method of the present invention;

[0060] Figure 2 This is a schematic diagram of a highway pavement temperature inversion model based on a multi-input branch neural network.

[0061] Figure 3 A schematic diagram of the physical loss function for embedding energy balance constraints. Detailed Implementation

[0062] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0063] This invention realizes a highway pavement temperature inversion method that integrates meteorological and traffic information and takes into account both data-driven and physical constraints, which improves the inversion accuracy and enhances the physical consistency of the model results.

[0064] like Figure 1 As shown, the method of the present invention specifically includes the following steps:

[0065] S1. Obtain meteorological information corresponding to the target highway section, including but not limited to temperature, humidity, wind speed, net radiation, and precipitation information;

[0066] Meteorological information is the primary external forcing factor influencing road surface temperature changes, directly determining the characteristics of radiative exchange, turbulent heat exchange, and wet processes between the road surface and the atmosphere. Specifically, air temperature, humidity, wind speed, net radiation, and precipitation are selected as meteorological input elements. These elements characterize the road surface thermal environment from the perspectives of turbulent heat exchange, radiative exchange, and wet processes, respectively. By unifying various meteorological elements into a meteorological element input vector, a structured and computable input format can be provided for subsequent model construction.

[0067] Specifically, for any target road segment With any time For example, construct the meteorological element input vector:

[0068] ;

[0069] in, and These are the road segment index and the time index, respectively. The input vector for the meteorological elements at this index. , , , and These are the temperature, humidity, wind speed, net radiation, and precipitation at this index location, respectively. Net radiation The meteorological information is obtained by combining shortwave and longwave radiation, and the sources of the meteorological information include, but are not limited to, automatic weather stations along the road, meteorological observation stations along the route, and numerical model analysis products.

[0070] S2. Obtain traffic information corresponding to the target highway segment. The traffic information includes, but is not limited to, road category information characterizing road structure features and traffic time sequence information characterizing traffic operation patterns. The road category information includes bridge segments, tunnel segments, and ramp segments, etc.

[0071] Unlike traditional methods that only consider meteorological factors, this invention takes into account the significant impact of the road operating environment itself on the road surface thermal environment, thus introducing traffic information as an auxiliary input. In this embodiment, traffic information mainly includes two categories: one is road category information, used to characterize the differences in heat exchange conditions among different road structures. For example, bridge sections typically have strong ventilation and high heat dissipation capacity, while tunnel sections are less affected by radiation and have a relatively stable thermal environment; ramp sections also have their own characteristics in terms of structure and operation. The other category is traffic temporal information, used to characterize the regularity of road operating status changes over time, such as diurnal variations and periodic operating characteristics. By introducing road category information as discrete variables and constructing traffic temporal information into a time feature vector, this invention provides the necessary information foundation for subsequent models to distinguish different road operating environments.

[0072] Specifically, road category information is represented using discrete variables:

[0073] ;

[0074] in, For road segment indexing, The road category label at this index includes, but is not limited to, ordinary roadbed, bridge, tunnel, and ramp. This represents the total number of road categories. Traffic time-series information can be constructed as a time feature vector:

[0075] ;

[0076] in, For time indexing, This is the traffic time-series feature vector at this index. For the hour corresponding to this index, Encode the date type corresponding to this index, including but not limited to weekdays, weekends, and public holidays. and These represent sine and cosine functions, respectively, used to map hourly variables to periodic features, enabling the inversion model to continuously and stably learn the diurnal variation characteristics of road surface temperature and avoid inversion errors caused by discontinuous time boundaries of day and night alternation.

[0077] S3. Perform time synchronization, spatial matching, and quality control on the meteorological and traffic information, and perform normalization preprocessing to construct a unified inversion input dataset, which is used for model training.

[0078] Unified data preprocessing was performed on the acquired meteorological and traffic information to ensure consistency across multiple data sources in terms of time, space, and numerical scale. First, time synchronization processing mapped data from different time resolutions onto the same time axis, avoiding feature misalignment caused by time inconsistencies. Second, spatial matching processing, such as bilinear interpolation, ensured a one-to-one correspondence between meteorological information and corresponding highway segments. Third, quality control eliminated obvious outliers, improving the reliability of the input data. Subsequently, continuous input variables were normalized to ensure variables with different dimensions and value ranges were at the same scale, preventing any single type of variable from dominating the model learning process due to excessively large numerical scales.

[0079] After the above processing, meteorological information, road category information, traffic time series information, and corresponding road surface temperature observations are combined to form a unified inversion input dataset for model training.

[0080] S4. Construct highway pavement temperature inversion models based on multi-input branch neural networks for different data types. Design a physical loss function that incorporates embedded energy balance constraints. Use deep learning to fuse preprocessed meteorological and traffic information to train the inversion model; such as Figure 2 As shown, the specific steps include:

[0081] S41. Since meteorological information, road category information, and traffic time series information have significant differences in data type and physical meaning, this invention adopts a multi-input branch neural network structure to extract features from different types of inputs.

[0082] The multi-input branch neural network structure includes a meteorological information branch, a road category information branch, a traffic time-series information branch, a feature fusion layer, and an output layer. Meteorological information, as a continuous numerical input, undergoes nonlinear feature extraction via a feedforward neural network to characterize the complex nonlinear relationship between meteorological elements and road surface temperature. Road category information, as discrete structural information, is transformed into a continuous feature vector through embedding mapping to express the implicit differences in thermal environment among different road structures. Traffic time-series information is processed by a time-series neural network to extract operational characteristics evolving over time. The outputs of each branch are combined in the feature fusion layer to form a comprehensive feature representation, and the output layer retrieves the road surface temperature. This structure matches the model's processing method with the physical properties and statistical characteristics of the input information, thus improving the model's expressive power.

[0083] Specifically, meteorological information input The effective features are extracted by processing the data through a feedforward neural network, and are represented as follows:

[0084] ;

[0085] in, This represents the feature vector output by the meteorological information branch. This represents a feedforward neural network representing a branch of meteorological information. This is the parameter set for the meteorological information branch.

[0086] Road category information By using an embedding layer for mapping, discrete category information is transformed into a continuous vector representation, in the form of:

[0087] ;

[0088] in, It is the embedding vector corresponding to the road category information. It is an embedding matrix. Road category information The corresponding one-hot encoding.

[0089] Traffic timing information The data is processed using a temporal neural network (a long short-term memory network is used in this embodiment) to capture the evolution of traffic changes over time, represented as follows:

[0090] ;

[0091] in, It is the feature vector output by the traffic time-series information branch. It is a time feature extraction function for Long Short-Term Memory (LSTM) networks. From time arrive Input of traffic timing information It is the parameter set of the timing branch.

[0092] The output features of all branch networks are fused together by concatenation or weighted summation to obtain a comprehensive feature vector. :

[0093] ;

[0094] fused feature vector The input is sent to an output layer, and the final output is a road surface temperature inversion estimate. :

[0095] ;

[0096] in, It is the mapping function of the output layer. It is the set of parameters for the output layer.

[0097] S42. To avoid the pure data-driven model from producing results that violate physical laws during training, this invention introduces energy balance constraints during the model training phase.

[0098] In the theory of pavement thermal processes, pavement temperature is determined by multiple heat fluxes, including net radiation, sensible heat flux, and geothermal flux. Based on this physical understanding, such as... Figure 3 As shown, this implementation constructs a physical constraint term centered on the energy balance residual, used to measure whether the road surface temperature output by the model satisfies the basic energy conservation relationship. By incorporating the energy balance residual into the loss function, the model is constrained by physical laws while minimizing prediction errors, thus maintaining the rationality and stability of the inversion results under conditions such as extreme weather, sparse samples, or complex road structures. Finally, the total loss function is composed of a weighted average of the data fitting loss and the physical constraint loss, with the weighting coefficients used to balance accuracy requirements with physical consistency requirements.

[0099] Specifically, net radiation The sensible heat flux has been provided in the meteorological information corresponding to the target highway section. and geothermal flux They can be expressed as parameterized as follows:

[0100] ;

[0101] ;

[0102] in, air density, The specific heat of air at constant pressure. The turbulent thermal resistance can be obtained through empirical relationships with wind speed. The thermal conductivity is an equivalent parameter. For road surface temperature inversion estimation, This represents the temperature of the lower layer of the road surface. Therefore, an energy balance residual is generated. :

[0103] ;

[0104] Therefore, physical constraint loss occurs. ,express The absolute values ​​of all samples within a training batch are averaged.

[0105] In addition, there is data fitting loss. This indicates the road surface temperature inversion estimation. Compared with the measured road surface temperature Mean squared error within a training batch.

[0106] Finally, the physical loss function embedded with energy balance constraints for:

[0107] ;

[0108] in, These are the physical constraint weighting coefficients, used to balance fitting accuracy and physical consistency.

[0109] S43. During the model training phase, the model parameters are optimized by minimizing the physical loss function that embeds the energy balance constraint using the constructed inversion input dataset. Gradient descent-type optimization methods can be used during training, and validation data can be used to monitor changes in model error to avoid overfitting.

[0110] The model parameters obtained through training enable the multi-input branch neural network to simultaneously learn the combined effects of meteorological factors, road structure differences, and traffic temporal characteristics on road surface temperature.

[0111] S5. Using the highway pavement temperature inversion model trained in step S4, calculate the pavement temperature based on meteorological and traffic information for the target time and target highway section, and obtain the pavement temperature inversion estimation results for the corresponding time and section.

[0112] The inversion results can be used for real-time road surface temperature monitoring and can further serve applications such as road surface icing risk assessment, traffic safety early warning, and road operation management.

[0113] In another embodiment of the present invention, a highway pavement temperature retrieval system integrating meteorological and traffic information includes:

[0114] The meteorological information acquisition unit is used to acquire meteorological information corresponding to the target highway section. The meteorological information includes temperature, humidity, wind speed, net radiation, and precipitation information.

[0115] The traffic information acquisition unit is used to acquire traffic information corresponding to the target highway segment. The traffic information includes road category information that characterizes the road structure characteristics and traffic time sequence information that characterizes the traffic operation pattern.

[0116] The dataset construction unit is used to synchronize meteorological and traffic information in time, match spatially, and control quality, and to perform normalization preprocessing to build a unified inversion input dataset for model training.

[0117] The model building and training unit is used to construct highway pavement temperature inversion models based on multi-input branch neural networks for different data types. A physical loss function with embedded energy balance constraints is designed, and deep learning is used to fuse preprocessed meteorological and traffic information to train the inversion model. The multi-input branch neural network structure includes a meteorological information branch, a road category information branch, a traffic time-series information branch, a feature fusion layer, and an output layer. Meteorological information is extracted nonlinearly through a feedforward neural network to characterize the complex nonlinear relationship between meteorological elements and pavement temperature. Road category information is transformed into continuous feature vectors through embedding mapping to express the implicit differences in thermal environment among different road structures. Traffic time-series information is processed through a time-series neural network to extract operational characteristics that evolve over time. The outputs of each branch are combined in the feature fusion layer to form a comprehensive feature representation, and the pavement temperature inversion result is obtained through the output layer.

[0118] The inversion calculation unit is used to perform inversion calculations on the pavement temperature based on meteorological and traffic information for the target time and target highway segment using a trained highway pavement temperature inversion model, and obtain the pavement temperature inversion estimation results for the corresponding time and segment.

[0119] In another embodiment of the present invention, an electronic device includes:

[0120] Memory, used to store computer programs;

[0121] A processor for executing the computer program to implement the method.

[0122] In another embodiment of the present invention, a non-volatile storage medium is provided for storing a computer program, wherein the computer program implements the method described thereon when executed by a processor.

[0123] In another embodiment of the present invention, a computer program product includes a computer program / instructions that, when executed by a processor, implement the method described herein.

Claims

1. A method for retrieving highway pavement temperature by integrating meteorological and traffic information, characterized in that, Includes the following steps: Obtain meteorological information corresponding to the target highway section, including temperature, humidity, wind speed, net radiation, and precipitation information; Obtain traffic information corresponding to the target highway segment. The traffic information includes road category information that characterizes the road structure characteristics and traffic time sequence information that characterizes the traffic operation pattern. Meteorological and traffic information is synchronized in time, matched spatially, and subjected to quality control. Normalization preprocessing is performed to construct a unified inversion input dataset for model training. A highway pavement temperature inversion model based on a multi-input branch neural network is constructed for different data types. A physical loss function with embedded energy balance constraints is designed. Based on the inversion input dataset, deep learning is used to fuse preprocessed meteorological and traffic information to train the inversion model. The multi-input branch neural network structure includes a meteorological information branch, a road category information branch, a traffic time series information branch, a feature fusion layer, and an output layer. Meteorological information is extracted nonlinearly through a feedforward neural network to characterize the complex nonlinear relationship between meteorological elements and pavement temperature. Road category information is transformed into continuous feature vectors through embedding mapping to express the implicit differences in thermal environment among different road structures. Traffic time series information is processed through a time series neural network to extract operational features that evolve over time. The outputs of each branch are combined in the feature fusion layer to form a comprehensive feature representation, and the pavement temperature inversion result is obtained through the output layer. The physical loss function with embedded energy balance constraints is used. Represented as: ； in, The data fitting loss represents the road surface temperature inversion estimation. Compared with the measured road surface temperature Mean squared error within a training batch; These are the physical constraint weighting coefficients. For physical constraint loss, it represents The average of the absolute values ​​of all samples within a training batch; The energy balance residual is expressed as: ； in, Net radiation, For sensible heat flux, For geothermal flux; ； ； in, air density, The specific heat of air at constant pressure. For road surface temperature inversion estimation, and These are the road segment index and the time index, respectively. The temperature at that index. For turbulent thermal resistance, The thermal conductivity is an equivalent parameter. Temperature of the lower layer of the road surface; Using a trained highway pavement temperature inversion model, the pavement temperature is inverted and calculated based on meteorological and traffic information for the target time and target highway segment, and the pavement temperature inversion estimation results for the corresponding time and segment are obtained.

2. The method according to claim 1, characterized in that, The meteorological element input vector is constructed based on meteorological information and represented as follows: ； in, and These are the road segment index and the time index, respectively. The input vector for the meteorological elements at this index. , , , and These are the temperature, humidity, wind speed, net radiation, and precipitation at that index location, respectively.

3. The method according to claim 1, characterized in that, Road category information is represented using discrete variables as follows: ； in, For road segment indexing, For the road category label at this index, Total number of road categories; Traffic time-series information is constructed as a time feature vector, represented as: ； in, For time indexing, This is the traffic time-series feature vector at this index. For the hour corresponding to this index, Encode the date type corresponding to this index. and These represent the sine and cosine functions, respectively.

4. The method according to claim 1, characterized in that, The meteorological and traffic information is synchronized in time, spatially matched, and quality controlled, and normalized preprocessing is performed to construct a unified inversion input dataset; including: By using time synchronization processing, data with different time resolutions are uniformly mapped onto the same time axis; Spatial matching processing is used to ensure that meteorological information is matched one-to-one with the corresponding highway sections; Outliers are eliminated through quality control. Normalize continuous input variables to bring variables with different dimensions and ranges to the same scale. After the above processing, meteorological information, road category information, traffic time series information, and corresponding road surface temperature observations are combined to form a unified inversion input dataset.

5. The method according to claim 1, characterized in that, Meteorological information is processed through a feedforward neural network to extract its effective features, which are represented as follows: ； in, This represents the feature vector output by the meteorological information branch. This represents a feedforward neural network representing a branch of meteorological information. For meteorological information input, The parameter set for the meteorological information branch; Road category information is mapped through an embedding layer, transforming discrete category information into a continuous vector representation, in the form of: ； in, It is the embedding vector corresponding to the road category information. It is an embedding matrix. Road category information The corresponding one-hot encoding; Traffic temporal information is processed through a long short-term memory network to capture the evolution of traffic changes over time, represented as: ； in, It is the feature vector output by the traffic time-series information branch. It is a time feature extraction function for Long Short-Term Memory (LSTM) networks. From time arrive Input of traffic timing information It is the set of parameters for the timing branch; The output features of all branch networks are fused together by concatenation or weighted summation to obtain a comprehensive feature vector. : ； fused feature vector The input is sent to an output layer, and the final output is a road surface temperature inversion estimate. : ； in, It is the mapping function of the output layer. It is the set of parameters for the output layer.

6. A highway pavement temperature retrieval system integrating meteorological and traffic information, characterized in that, include: The meteorological information acquisition unit is used to acquire meteorological information corresponding to the target highway section. The meteorological information includes temperature, humidity, wind speed, net radiation, and precipitation information. The traffic information acquisition unit is used to acquire traffic information corresponding to the target highway segment. The traffic information includes road category information that characterizes the road structure characteristics and traffic time sequence information that characterizes the traffic operation pattern. The dataset construction unit is used to synchronize meteorological and traffic information in time, match spatially, and control quality, and to perform normalization preprocessing to build a unified inversion input dataset for model training. The model building and training unit is used to construct highway pavement temperature inversion models based on multi-input branch neural networks for different data types. A physical loss function with embedded energy balance constraints is designed. Deep learning is used to fuse preprocessed meteorological and traffic information to train the inversion model. The multi-input branch neural network structure includes meteorological information branches, road category information branches, traffic time-series information branches, a feature fusion layer, and an output layer. Meteorological information is extracted nonlinearly through a feedforward neural network to characterize the complex nonlinear relationship between meteorological elements and pavement temperature. Road category information is transformed into continuous feature vectors through embedding mapping to express the implicit differences in thermal environment among different road structures. Traffic time-series information is processed through a time-series neural network to extract operational characteristics evolving over time. The outputs of each branch are combined in the feature fusion layer to form a comprehensive feature representation, and the pavement temperature inversion result is obtained through the output layer. The physical loss function with embedded energy balance constraints is used. Represented as: ； in, The data fitting loss represents the road surface temperature inversion estimation. Compared with the measured road surface temperature Mean squared error within a training batch; These are the physical constraint weighting coefficients. For physical constraint loss, it represents The average of the absolute values ​​of all samples within a training batch; The energy balance residual is expressed as: ； in, Net radiation, For sensible heat flux, For geothermal flux; ； ； in, air density, The specific heat of air at constant pressure. For road surface temperature inversion estimation, and These are the road segment index and the time index, respectively. The temperature at that index. For turbulent thermal resistance, The thermal conductivity is an equivalent parameter. Temperature of the lower layer of the road surface; The inversion calculation unit is used to perform inversion calculations on the pavement temperature based on meteorological and traffic information for the target time and target highway segment using a trained highway pavement temperature inversion model, and obtain the pavement temperature inversion estimation results for the corresponding time and segment.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1-5.

8. A non-volatile storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1-5.

9. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method described in any one of claims 1-5.

Images