TPA-LSTM-based vehicle-type-based dynamic early warning and speed guiding method for wet and slippery road surface in shunting area

Abstract

The invention relates to a TPA-LSTM-based shunting area wet and slippery road surface vehicle type-based dynamic early warning and speed guiding method. The method comprises the following steps: S1, collecting traffic flow operation time sequence data in a highway shunting area wet and slippery road surface scene; s2, predicting speed limit values of different vehicle types for safe passing on the wet and slippery road surface in the shunting area; s3, formulating early warning information release content; s4, determining an early warning information issuing mode; s5, determining an information issuing position of the highway shunting area, wherein the information issuing position is comprehensively determined according to an existing shunting area ETC portal layout principle in combination with a warning information issuing standard; and S6, early warning information is pushed to the vehicle end, and the ETC door frame receives the optimal passing speed information of the wet and slippery road surface pushed by the traffic management background and issues the information to vehicle-mounted OBU equipment with a voice broadcast function of the vehicle. According to the method, the real-time performance, the dynamic adaptability, the personalized early warning and the driver response effect are remarkably improved, so that the occurrence rate of accidents in the diversion area in a wet and slippery road scene in rainy days is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of traffic safety technology, and more specifically to a dynamic early warning and speed guidance method for different vehicle types on slippery roads in a diversion area based on TPA-LSTM. Background Art

[0002] Existing domestic and foreign accident data show that the accident rate in the highway diversion area is about 30% higher than that in general sections. In addition, traffic accidents often occur under bad weather conditions, and excessive speed is an important cause of accidents, especially on slippery roads in the diversion area on rainy days. Due to frequent lane changes in the diversion area and the reduced friction coefficient between the tire and the ground in slippery road scenes, when the driver is driving at high speed, the vehicle will slip and lose control when changing lanes, and the incidence of rear-end collisions caused by increased braking distance will increase greatly. Therefore, how to effectively guide vehicles to slow down before slippery roads in the highway diversion area is an important means to reduce the accident rate.

[0003] At present, in the scenario of slippery road surface in the highway diversion area, the warning methods to guide drivers to slow down in advance mainly include setting traditional speed limit signs, variable information boards or laying longitudinal deceleration markings. Although these methods have certain speed guidance capabilities, they still have shortcomings. For example: the speed limit value of traditional speed limit signs is fixed and lacks dynamic update capabilities; the longitudinal deceleration markings on the road surface use the principle of visual perception to make drivers have the illusion of speeding too fast, thereby inducing drivers to slow down, but their effects are fixed and limited, and they cannot be dynamically adjusted according to real-time road conditions and weather changes; although variable information boards can dynamically update speed limit information, they have information update lags, making it difficult to provide drivers with accurate safe driving speeds based on the real-time slipperiness of the road surface. In addition, the above three speed guidance methods also have a limitation, that is, they cannot dynamically adjust the speed limit according to different vehicle models. The existing speed guidance methods for slippery roads in highway diversion areas have poor real-time performance, insufficient dynamics and lack of vehicle model adaptability. Since the slippery road conditions and vehicle behavior characteristics will change dynamically over time, the traditional time series prediction model has unreasonable information weight distribution when dealing with nonlinear time series relationships. There are problems in calculating the risk warning speed limit values ​​for different vehicle models in the slippery road scenario in the diversion area, and there is a lack of adaptability for different vehicle models in the speed guidance. The speed limit warning for slippery roads in the diversion area has poor real-time performance and insufficient dynamics. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a dynamic warning and speed guidance method for different vehicle types on slippery roads in diversion areas based on TPA-LSTM (Temporal Pattern Attention Long Short-Term Memory), which significantly improves the response efficiency and accuracy of speed warning, reduces the risk of sideslipping or rear-end collision accidents caused by vehicle type differences in diversion areas on slippery roads, and is conducive to improving driving safety.

[0005] The technical solution adopted by the present invention to solve the technical problem is to construct a dynamic warning and speed guidance method for slippery road types in a diversion area based on TPA-LSTM, including the following steps:

[0006] S1. Collect traffic flow operation timing data in the slippery road scenario in the highway diversion area;

[0007] S2: Predict the speed limits for different vehicle types to safely pass through slippery roads in the diversion area. Based on the TPA-LSTM deep learning model, combined with slippery road environment data and characteristics of different vehicle types, the optimal speed limit for each vehicle type is predicted in real time;

[0008] S3. Formulate the content of early warning information release;

[0009] S4. Determine the method of issuing warning information, and convey the warning information to the driver based on the functional characteristics of the ETC gantry and the voice broadcast function of the on-board OBU equipment;

[0010] S5. Determine the information release location of the highway diversion area. The information release location is determined based on the existing ETC gantry layout principles of the diversion area and combined with the warning information release standards;

[0011] S6: Push warning information to the vehicle end. The ETC gantry receives the optimal speed information for slippery roads pushed by the traffic management background, and sends the information to the vehicle's on-board OBU device with voice broadcast function, reminding the driver to slow down by voice.

[0012] According to the above scheme, in step S1, the collected traffic flow operation time series data includes timestamp, vehicle type, speed, acceleration, rainfall and road friction coefficient, and the collected traffic flow operation time series data type is divided into two categories: small passenger cars and large trucks.

[0013] According to the above scheme, a speedometer is used to obtain the timestamp, speed and acceleration data, and the acceleration is obtained by derivation calculation of the speed data; the rain gauge is used to collect real-time rainfall; and the road friction coefficient sensor is responsible for monitoring the road friction coefficient.

[0014] According to the above scheme, in step S2, based on the TPA-LSTM deep learning model, the associated variables of rainfall and road friction coefficient in a slippery road environment are introduced into the model input variables, and the input traffic flow operation timing data is normalized using the Min-Max normalization method, and the data is normalized to the interval [0, 1] to improve the convergence speed and prediction accuracy of the model. The normalization formula is:

[0015]

[0016] Where, X normalized is the normalized value of sample X, X max is the maximum value of the sample data, X min is the minimum value of the sample data. In order to train and evaluate the prediction effect of the TPA-LSTM model, the data set is divided into a training set and a test set. The training set is used to optimize the model parameters, and the test set is used to verify the generalization ability of the model. The ratio of the training set to the test set is 7:3;

[0017] The model uses time series features, including timestamp t, vehicle model s t , speed v t , acceleration a t The speed and acceleration of the car and truck are v tc 、v tt and a tc 、a tt , rainfall t and road friction coefficient μ t Input, train the model to predict the speed change at future moments. For each time step t, the input feature set HI t It is expressed as:

[0018] HI t =[t,s t , v t , a t , r t , μ t ];

[0019] In order to effectively capture the temporal characteristics and regularities of the data, the time window size and time step are set. The time window T is used to include continuous historical time step data during prediction, and the time window length is set to 50, that is, {X t-T+1 , X t-T+2 , X t-T+3 ..., X t-T+50 As the input feature, the time step is set to 1, that is, the model advances 1 time step each time to maintain the time series characteristics of continuous data and ensure the accuracy of the prediction results;

[0020] The TPA-LSTM model is based on the traditional LSTM model. By adding the temporal pattern attention mechanism, the model's key feature extraction ability at different time series is enhanced. The key modules of the model include LSTM units and temporal attention mechanism. The structure includes input layer, LSTM layer, fully connected layer and output layer. The LSTM layer consists of two layers of stacked LSTM units, each layer contains 64 neurons, which extract the temporal features of the input data. The fully connected layer is used to map the output of LSTM to the prediction result, and the output layer finally gives the predicted speed value. The forget gate f in the LSTM layer t The formula for determining whether information is retained is:

[0021] f t =σ(W f ·[h t-1 , x t ]+b f );

[0022] Input Gate i t Control the influence of the current input information, input gate i t and candidate memory cell c t The formulas are:

[0023] i t =σ(W i ·[h t-1 , x t ]+b i );

[0024] c t =tanhW c ·[h t-1 , x t ]+b c ;

[0025] Cell state update C t Combined with forget gate f t and input gate i t The update formula is:

[0026] C t =f t ×C t-1 +i t ×c t ;

[0027] Output gate o t Generate the output at the current moment, output gate o t and the hidden state h t The formulas are:

[0028] o t =σ(W o ·[h t-1, x t ]+b o );

[0029] h t =o t tanh(c t );

[0030] In the formula, o t represents the sigmoid activation function, tanh represents the tanh activation function, W f , W i , W c and W o is the model parameter matrix;

[0031] In the TPA-LSTM model, the temporal attention mechanism enhances the model's attention to key moments by dynamically adjusting the weights of time steps. First, according to the hidden state h of the current time step, t And the hidden states of other time steps, the attention weight α is calculated through nonlinear transformation t , the specific formula is as follows:

[0032]

[0033] In the formula, α t represents the attention weight at time step t, W, U and b are learnable parameters. Finally, the model performs a weighted average on the hidden state to generate a weighted hidden state H to capture the dynamic information of the key time step. The formula is:

[0034]

[0035] During the training process, the mean square error (MSE) loss function is used to measure the error between the predicted speed and the actual speed. The formula is:

[0036]

[0037] In the formula, v i is the true value, V i is the speed prediction value, n is the number of samples, Adam is selected as the optimizer, and the initial learning rate is set to 0.001 to improve the convergence speed and avoid falling into the local optimal solution. The early stopping strategy is adopted in the training process. If the loss function does not decrease significantly for 5 consecutive times on the test set, the training is stopped in advance to prevent overfitting. After the model training is completed, the root mean square error (RMSE) and mean absolute error (MAE) indicators are used to evaluate the generalization ability of the model through the test set. The root mean square error and mean absolute error calculation formulas are:

[0038]

[0039] After completing model training and verification, the TPA-LSTM model can be used for real-time speed prediction by combining the real-time collected timestamp t, speed v t , acceleration a t , rainfall t and road friction coefficient μ t Data, for each time step t, the feature set HI t =[t,s t , v t , a t , r t , μ t ] Input model, TPA-LSTM forward propagates the time series data through LSTM unit and time pattern attention mechanism to generate the speed prediction value of the next time step. Finally, the model outputs the predicted speed HI′ t =[V i ] is used to push speed limit warning information to the vehicle side, thereby ensuring the safe passage speed of the vehicle in different slippery road scenarios.

[0040] According to the above scheme, in step S3, the warning information content is constructed as: "The road surface is slippery at X meters ahead, please reduce the speed to X kilometers per hour."

[0041] According to the above scheme, in step S4, the warning method uses an intelligent OBU device to receive the warning information sent by the ETC gantry, and uses the voice broadcast function of the OBU to broadcast the road danger to the driver in real time, reminding the driver to slow down in advance to ensure driving safety.

[0042] According to the above scheme, in the warning method, the ETC gantry is set as the main receiving and sending device of the warning information. When slippery road conditions are detected in the diversion area, the traffic management background will trigger the control link to generate warning information and send the information to the ETC gantry. The wireless communication module of the ETC gantry uses dedicated short-range communication technology to transmit the warning information to the on-board OBU device. The parsing module of the OBU device is started to process and decode the information content to identify the specific warning content.

[0043] According to the above scheme, in step S5, the release location of the slippery road warning information in the highway diversion area is set upstream of the diversion area, which is greater than 500 meters.

[0044] According to the above scheme, in step S6, a dynamic adaptive risk warning strategy for slippery road surface in the diversion area is adopted. The rainfall and road friction coefficient of the slippery road surface in the diversion area are monitored and obtained in real time through the traffic management background. The different types and speed information of vehicles approaching the ETC gantry are considered, and the speed limit value is dynamically adjusted and optimized in combination with the TPA-LSTM deep learning model. The linkage between the ETC gantry and the on-board OBU equipment is utilized to convey the warning information to the driver in the form of voice.

[0045] According to the above scheme, the specific method of the dynamic adaptive risk warning strategy for slippery road surface in the diversion area includes the following steps:

[0046] S601. In a slippery road scenario, when an OBU equipped with a voice broadcast function approaches the ETC gantry upstream of the highway diversion area, the ETC gantry exchanges information with the vehicle's OBU device through DSRC. During this process, the speedometer collects the vehicle's approaching speed time series data, and the ETC gantry obtains the vehicle's model information and uploads the real-time data of the vehicle model, speed, and acceleration to the traffic management backend;

[0047] S602. The traffic management backend relies on the TPA-LSTM large model in the system, combined with the current environmental conditions and vehicle information of the diversion area, to predict the optimal safe speed of the vehicle on slippery roads, and sends the predicted speed limit information to the ETC gantry upstream of the diversion area. After receiving the speed limit information, the ETC gantry pushes the warning information to the vehicle OBU device through DSRC. The OBU device then informs the driver of the warning information through voice broadcast, prompting him to slow down safely.

[0048] The implementation of the TPA-LSTM-based dynamic warning and speed guidance method for slippery roads in the diversion area and different vehicle types has the following beneficial effects:

[0049] 1. The present invention effectively solves the problems of the inadequacy of existing warning technologies in terms of real-time and dynamic performance. By collecting rainfall and road friction coefficient in real time in the slippery road scene in the highway diversion area, and combining the vehicle speed, acceleration and vehicle model information, the TPA-LSTM model can generate more accurate speed limit prediction values ​​according to the real-time changing road conditions and vehicle dynamic behavior. By introducing deep learning models and vehicle-road-cloud collaborative mechanisms, the ability to adaptively adjust speed limit information according to real-time road conditions is achieved, which significantly improves the response efficiency and accuracy of speed warnings.

[0050] 2. The speed warning method of the present invention shows unique advantages in vehicle type adaptability. Based on the wireless communication mechanism between the ETC gantry and the intelligent OBU device, it accurately obtains and identifies the vehicle type information of each vehicle, and provides personalized speed warnings according to different vehicle types. By dynamically adjusting the speed limit information, the risk of skidding or rear-end collisions caused by vehicle type differences in the slippery road diversion area is further reduced. The personalized speed limit strategies for different vehicle types are conducive to improving driving safety.

[0051] 3. The present invention transmits speed warning information to the driver in the form of OBU voice broadcast through the linkage between the ETC gantry and the intelligent OBU device, overcoming the problem that the traditional visual speed limit sign may be ignored. The voice broadcast can significantly improve the driver's reaction speed and compliance rate, thereby realizing the deceleration operation more quickly and effectively. It not only improves the timeliness of the warning information transmission, but also significantly improves the driver's perception of the warning information, further enhances the guarantee of safe driving, and provides a new way for the highway traffic management platform to push warning information.

[0052] 4. The present invention has significantly improved real-time performance, dynamic adaptability, personalized warning and driver response effects, thereby greatly reducing the incidence of diversion area accidents on slippery road scenarios in rainy days. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0054] Figure 1 It is the technical roadmap of the present invention's method for dynamic early warning and speed guidance of slippery road surface in diversion area based on TPA-LSTM;

[0055] Figure 2 It is a schematic diagram of the process of pushing slippery road warning information for a highway diversion area according to the present invention;

[0056] Figure 3 It is a schematic diagram of the TPA-LSTM model structure of the present invention. DETAILED DESCRIPTION

[0057] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0058] In the risk scenario of slippery road surface in the highway diversion area, the traffic management backend uses roadside sensing equipment to collect environmental information such as rainfall and road friction coefficient in the risk scenario, and uploads the relevant data to the traffic management backend. When the ETC gantry near the upstream of the highway diversion area detects an incoming vehicle, the speed radar on the gantry records the vehicle speed and acceleration information. At the same time, the ETC gantry interacts with the on-board OBU device through the dedicated short-range communication technology (DSRC) to obtain the vehicle model information. Subsequently, the ETC gantry uploads the vehicle model, speed, acceleration and other data to the traffic management backend. Based on the pre-trained TPA-LSTM deep learning model embedded in the system, the traffic management backend combines the real-time environment and vehicle information to predict the optimal safe speed limit value for the vehicle under slippery road conditions, and sends this speed limit information to the ETC gantry. The ETC gantry pushes the warning information containing the speed limit value to the intelligent on-board OBU device with voice broadcast function on the vehicle side through DSRC. The OBU informs the driver of the warning information through voice broadcast, prompting him to safely slow down and ensure driving safety.

[0059] The present invention proposes a speed warning method that can dynamically adjust the speed limit information according to different vehicle models and real-time slippery road conditions. The method monitors and obtains the rainfall and road friction coefficient of the slippery road in the diversion area in real time through the traffic management background, and combines the different vehicle models and speed data of vehicles near the ETC gantry as model input. Based on the TPA-LSTM deep learning model, the speed limit value is dynamically adjusted and optimized and sent to the ETC gantry. At the same time, the ETC gantry is linked with the on-board OBU equipment to convey the warning information to the driver in the form of voice. The specific contents are as follows:

[0060] 1) Calculation of slippery road risk warning speed limit value in diversion area based on TPA-LSTM model

[0061] TPA-LSTM is an improved long short-term memory network model that introduces a module based on the attention mechanism to automatically select and focus on key time steps to extract the most important time information for the prediction task. Since the slippery road conditions and vehicle behavior characteristics change dynamically over time, traditional time series prediction models may face the problem of unreasonable information weight distribution when dealing with nonlinear time series relationships. The advantage of TPA-LSTM is that it can adaptively weight important features in the time series (such as sudden slippery changes or vehicle acceleration / deceleration periods) through the attention mechanism, thereby improving the modeling ability of complex and long time series dependencies. This improvement helps to improve the performance of the model under irregular time series or noisy data, making the prediction results more accurate and robust.

[0062] Considering the advantages of TPA-LSTM model in time series data prediction, TPA-LSTM deep learning model is used to predict the speed limit of various types of vehicles. At the same time, rainfall and road friction coefficient, which are closely related to slippery road environment, are introduced into the model input variables to more accurately reflect the characteristics of complex road conditions.

[0063] Based on the collected slippery road time series data in the diversion area, the TPA-LSTM model uses time series data as input, including timestamp, vehicle type, speed, acceleration, rainfall, and road friction coefficient, extracts time-dependent features through the LSTM layer, and uses the attention mechanism to focus on the important features of key time steps, thereby predicting the future safe passage speed of the vehicle. The optimization model considers the associated variables in the slippery road environment in the data input items, which can better fit the complex feature relationship of the slippery road scene to a certain extent and meet the speed prediction needs under different road conditions.

[0064] 2) Method for pushing warning information for different vehicle models based on ETC gantry and intelligent OBU equipment

[0065] Risk warning method based on ETC gantry and intelligent OBU equipment to obtain vehicle model information: When the vehicle OBU device enters the communication range of the ETC gantry, the ETC gantry and OBU establish wireless communication through dedicated short-range communication technology (DSRC). At this time, the OBU device will transmit the identification information of the vehicle type to the ETC gantry. This communication mechanism ensures that the traffic management background can obtain accurate vehicle model data in real time based on the ETC gantry, and issue personalized warning information for different vehicle models accordingly.

[0066] With the development of technology, more and more vehicle-mounted OBU devices have built-in voice function modules. As the receiving end of the ETC gantry, the intelligent vehicle-mounted OBU device can automatically receive data signals from the ETC gantry. Once the warning information is received, the information is immediately processed and decoded to extract the specific warning content. Subsequently, the warning information is converted into a voice signal through the built-in speech synthesis and broadcast module, and the driver is reminded to slow down through voice broadcast. This warning method is based on the ETC gantry and OBU equipment. While realizing the core functions of non-stop toll collection, it further expands its application scenarios and provides a new way for the highway diversion area to push personalized warning information for different models. The warning information is transmitted to the driver in the form of voice broadcast, which can improve the driver's reaction speed and compliance rate.

[0067] 3) Dynamic speed limit warning response strategy based on real-time slippery road conditions in the diversion area

[0068] Taking advantage of the existing vehicle-road-cloud integration technology in real-time information sharing, early warning and emergency response, the proposed dynamic speed limit early warning response strategy based on the real-time slippery road conditions in the diversion area is as follows: In the slippery road scenario in the diversion area of ​​the highway, the traffic management background uses the roadside sensing equipment (road friction coefficient sensor and rain gauge) to collect environmental information such as the friction coefficient and rainfall of the road in real time, and uploads the relevant data to the traffic management background. When the ETC gantry upstream of the diversion area detects an incoming vehicle, the speed radar on the gantry will record the vehicle speed and acceleration information. At the same time, the ETC gantry interacts with the intelligent OBU device through DSRC to obtain the vehicle model information. Subsequently, the ETC gantry uploads the acquired model, speed, acceleration and other data to the traffic management background. Based on the TPA-LSTM deep learning model pre-trained and embedded in the system, the traffic management background uses the real-time environment and vehicle information data as the model input data to predict the optimal safe speed limit value of the vehicle under slippery road conditions, and sends this speed limit information to the ETC gantry. The ETC gantry will push the warning information including the speed limit value to the intelligent OBU device on the vehicle side through DSRC. The OBU will inform the driver of the warning information through voice broadcast, prompting him to safely slow down and ensure driving safety.

[0069] This risk warning strategy uses vehicle-road-cloud integrated technology, real-time environmental perception, intelligent data analysis and rapid interaction between vehicle and roadside equipment to achieve dynamic speed limit warnings on slippery roads in diversion areas: roadside perception equipment collects road condition data in real time, and the background deep learning model quickly calculates the safe speed limit value and immediately transmits the warning information to the vehicle-side equipment, ensuring the real-time and dynamic nature of the speed limit warning response and improving driving safety.

[0070] Example

[0071] like Figure 1 As shown, a dynamic early warning method for multiple vehicle types on slippery roads in a diversion area based on TPA-LSTM includes the following steps:

[0072] S1. Collect traffic flow operation timing data in slippery road scenarios in highway diversion areas. The collected data include timestamp, vehicle type, speed, acceleration, rainfall, and road friction coefficient.

[0073] In order to collect the time series data of traffic flow operation in the slippery road scene of the highway diverging area, the typical diverging area scene type in the highway was selected. Figure 2 As shown in Figure 1, existing studies have shown that more than 70% of ramp types have the following characteristics: the number of lanes upstream and downstream of the main line is the same, a deceleration lane is set up in the diversion area by widening the lane, and the exit ramp is a single lane. Figure 2 Typical ramp types are shown in order to effectively collect time series data on traffic flow operation.

[0074] In order to collect the timestamp, vehicle type, speed, acceleration, rainfall and road friction coefficient in the traffic flow time series data, a speedometer is used to obtain the timestamp, speed and acceleration data, where the acceleration is obtained by derivation of the speed data; the rain gauge is used to collect real-time rainfall; and the road friction coefficient sensor is responsible for monitoring the road friction coefficient. In order to provide speed warnings for different vehicle types, the collected data types are divided into two categories: small passenger cars and large trucks, in order to optimize their driving speed and improve road traffic safety. In order to provide speed warnings for small passenger cars and large trucks respectively, when collecting vehicle operation data in the slippery road scene in the diversion area, it is necessary to collect data for the two different vehicle types separately.

[0075] In order to ensure the training effect of the TPA-LSTM large model in step S2 and improve the generalization ability of the model, sufficient and appropriate time series data must be provided. The duration of the time series data should be reasonably selected according to the fluctuation period of the traffic flow, the requirements of the prediction scenario, and the data sampling frequency. Therefore, for the slippery road scene in the diversion area, it is necessary to cover different types of rainfall conditions (such as light rain, moderate rain and heavy rain) and traffic flow conditions at different times (such as peak and off-peak periods). At the same time, in order to avoid prediction bias caused by seasonal changes, it is recommended to select rainy day data covering different seasons to capture the diverse impacts of rainy days on traffic flow in each season. To ensure the timeliness and accuracy of the data, the data sampling frequency should be above 10Hz.

[0076] S2. Predict the speed limits for different types of vehicles for safe passage on slippery roads in diversion areas. Based on the TPA-LSTM deep learning model, combined with slippery road environment data and characteristics of different types of vehicles, the optimal speed limit for each type of vehicle is predicted in real time.

[0077] In order to predict the vehicle speed in slippery road scenarios, the prediction method based on the TPA-LSTM model introduces the associated variables in the slippery road environment: rainfall and road friction coefficient; the proposed optimization model can better adapt to the speed prediction needs of slippery road scenarios in the diversion area by considering the rich feature relationships in the slippery road scenario. Compared with the traditional TPA-LSTM model that does not introduce associated variables in the slippery road environment, the optimization model can better fit the complex feature relationships of the slippery road scenario and meet the speed prediction needs under different road conditions.

[0078] like Figure 3As shown in the figure, the framework structure of the TPA-LSTM model. Before building the prediction model, the original data of passenger cars and trucks collected in step S1 are first cleaned and preprocessed to ensure the accuracy and stability of the model prediction. To this end, the input data is normalized, and the Min-Max normalization method is used to normalize the data to the interval [0, 1] to improve the convergence speed and prediction accuracy of the model. The normalization formula is:

[0079]

[0080] Where, X normalized is the normalized value of sample X, X max is the maximum value of the sample data, X min is the minimum value of the sample data. In order to train and evaluate the prediction effect of the TPA-LSTM model, the data set is divided into a training set and a test set. The training set is used to optimize the model parameters, and the test set is used to verify the generalization ability of the model. The ratio of the two is 7:3.

[0081] The model uses time series features, including timestamp t, vehicle model s t , speed v t , acceleration a t (The speed and acceleration of the car and truck are v tc 、v tt and a tc 、a tt ), rainfall t and road friction coefficient μ t The model is trained to predict the speed change at future moments. For each time step t, the input feature set HI t It is expressed as:

[0082] HI t =[t,s t , v t , a t , r t , μ t ];

[0083] In order to effectively capture the temporal characteristics and regularities of the data, the model needs to set the time window size and time step. Among them, the time window T is used to include continuous historical time step data during prediction, and the time window length is set to 50, that is, {X t-T+1 , X t-T+2 , X t-T+3 ......, X t-T+50} as input features. The time step is set to 1, that is, the model advances 1 time step each time to maintain the time series characteristics of continuous data and ensure the accuracy of the prediction results.

[0084] The TPA-LSTM model is based on the traditional LSTM model. By adding a temporal pattern attention mechanism, the model's ability to extract key features in different time series is enhanced. The key modules of the model include LSTM units and temporal attention mechanisms. Its structure includes input layer, LSTM layer, fully connected layer and output layer. The LSTM layer consists of two layers of stacked LSTM units, each layer contains 64 neurons, which extract the temporal features of the input data. The fully connected layer is used to map the output of the LSTM to the prediction result, and the output layer finally gives the predicted speed value. The forget gate f in the LSTM layer t The formula for determining whether information should be retained is:

[0085] f t =σ(W f ·[h t-1 , x t ]+b f );

[0086] Input Gate i t Control the influence of the current input information, input gate i t and candidate memory cell c t The formulas are:

[0087] i t =σ(W i ·[h t-1 , x t ]+b i );

[0088] c t =tanhW c ·[h t-1 , x t ]+b c ;

[0089] Cell state update C t Combined with forget gate f t and input gate i t The update formula is:

[0090] C t =f t ×C t-1 +i t ×c t ;

[0091] Output gate o t Generate the output at the current moment, output gate o t and the hidden state h t The formulas are:

[0092] o t =σ(W o ·[ht-1 , x t ]+b o );

[0093] h t =o t tanh(c t );

[0094] In the formula, o t represents the sigmoid activation function, tanh represents the tanh activation function, W f , W l , W c and W o is the model parameter matrix.

[0095] In the TPA-LSTM model, the temporal attention mechanism enhances the model's attention to key moments by dynamically adjusting the weights of time steps, enabling the model to more effectively capture important patterns in time series data, thereby improving the accuracy of predictions, especially in vehicle speed prediction scenarios on slippery roads with long time dependencies or complex time features. In implementation, the temporal pattern attention mechanism first calculates the hidden state h of the current time step based on the current state h. t And the hidden states of other time steps, the attention weight α is calculated through nonlinear transformation t , the specific formula is as follows:

[0096]

[0097] In the formula, α t represents the attention weight at time step t, and W, U, and b are learnable parameters. Finally, the model performs a weighted average of the hidden states to generate a weighted hidden state H to capture the dynamic information of the key time step. The formula is:

[0098]

[0099] During the training process, the mean square error (MSE) loss function is used to measure the error between the predicted speed and the actual speed. The formula is:

[0100]

[0101] In the formula, v i is the true value, V iis the speed prediction value, and n is the number of samples. Adam is selected as the optimizer, and the initial learning rate is set to 0.001 to improve the convergence speed and avoid falling into the local optimal solution. The early stopping strategy is adopted in the training process. If there is no significant decrease in the loss function for 5 consecutive times on the test set, the training is stopped in advance to prevent overfitting. After the model training is completed, the root mean square error (RMSE) and mean absolute error (MAE) indicators are used to evaluate the generalization ability of the model through the test set. The root mean square error and mean absolute error calculation formulas are:

[0102]

[0103] After completing model training and verification, the TPA-LSTM model can be used for real-time speed prediction. t , acceleration a t , rainfall t and road friction coefficient μ t Data, for each time step t, the feature set HI t =[t, s t , v t , a t , r t , μ t ] Input the model, TPA-LSTM forward propagates the time series data through LSTM units and temporal pattern attention mechanism to generate the speed prediction value for the next time step. Finally, the model outputs the predicted speed HI′ t =[V i ] is used to push speed limit warning information to the vehicle side, thereby ensuring the safe passage speed of the vehicle in different slippery road scenarios.

[0104] S3. Formulate the content of early warning information release, and follow the principles of clarity, specificity and simplicity to avoid information overload, insufficient information, or contradictory and ambiguous content.

[0105] In today's information-based driving environment, effectively transmitting warning information to the on-board OBU device and broadcasting it by voice is one of the effective measures to improve driving safety. However, relevant research data show that information overload, insufficient content or ambiguity will bring additional burden to the driver's brain and physiological state to a certain extent. In view of this, it is very important to design a reasonable set of warning voice broadcast content. According to the requirements of the industry standard "JT / T 1354-2020 Requirements for Voice Navigation Phrases for Vehicle Travel", the ideal warning information voice broadcast phrases should be semantically clear, simple and clear, and should not confuse the driver. Following the navigation phrase structure of the above standard, that is, the principle of "'ahead' + distance + indication category + auxiliary category", taking the ETC gantry deployed 500 meters upstream of the diversion area and the speed limit of 80km / h as an example, the warning information content is constructed as: "The road surface is slippery 500 meters ahead, please reduce the speed to 80 kilometers per hour". The above broadcast content is not only logically clear, but also efficient in information transmission, making it easy for drivers to quickly capture key information and ensuring that drivers can quickly understand and take corresponding measures at critical moments.

[0106] S4. Determine the method of issuing warning information, and convey the warning information to the driver based on the functional characteristics of the ETC gantry and the voice broadcast function of the on-board OBU equipment;

[0107] The way to release warning information relies on the existing ETC gantry and intelligent on-board OBU equipment. Based on the above equipment, a warning method for slippery road risks in the diversion area of ​​highways is proposed. While realizing the core function of non-stop toll collection, the ETC gantry and OBU equipment further expand its application scenarios, providing a new way to push warning information in the diversion area of ​​highways. Intelligent OBU equipment is used to receive the warning information sent by the ETC gantry, and the voice broadcast function of the OBU is used to broadcast the dangerous road conditions to the driver in real time, reminding the driver to slow down in advance to ensure driving safety.

[0108] In this early warning mode, the ETC gantry is set as the main receiving and sending device for early warning information. The existing ETC gantry not only undertakes the electronic toll collection function, but also has a strong data transmission capability. The traffic management background conducts real-time analysis of the road condition data collected by the gantry, and can dynamically monitor the slippery road conditions in the diversion area of ​​the highway. When slippery road conditions are detected in the diversion area, the traffic management background will trigger the control link to generate early warning information and send the information to the ETC gantry. Subsequently, the wireless communication module of the ETC gantry uses the dedicated short-range communication technology (DSRC) to transmit the early warning information to the on-board OBU device. When the vehicle enters the communication coverage of the ETC gantry, the early warning information will be quickly transmitted to the OBU device through the data link.

[0109] As the receiving end of the ETC gantry, the intelligent vehicle-mounted OBU device has a built-in automatic receiving module that can automatically receive data signals from the ETC gantry. Once the warning information is received, the parsing module of the OBU device is immediately activated to process and decode the information content and identify the specific warning content. The warning function of the OBU device relies on its internal voice broadcast system, which includes a voice synthesis and broadcast module. The parsed warning information will be automatically converted into a voice signal. The built-in voice synthesis system of the OBU is based on a preset voice model library to convert text information into voice content that is easy for the driver to understand, and broadcast it to the driver through the OBU speaker, such as "The road is slippery 500 meters ahead, please reduce the speed to 80 kilometers per hour."

[0110] S5. Determine the information release location of the highway diversion area. The information release location is determined based on the existing ETC gantry layout principle of the diversion area and combined with the warning information release standards;

[0111] Reasonable warning information release location can effectively help drivers discover potential risks in advance, so as to quickly take risk avoidance measures. According to existing data, the straight-line distance between the ETC gantry layout and the exit ramp of the diversion area should be no less than 1km in principle, and can be appropriately relaxed to no less than 500m when conditions are limited. At the same time, according to the relevant provisions of "JTG D82-2009 Highway Traffic Signs and Markings Setting Specifications", taking the design speed of 120 kilometers per hour on the expressway as an example, under heavy traffic conditions, when vehicles need to slow down or change lanes, the warning sign should be more than 360m away from the starting position of the danger zone.

[0112] In order to make full use of the existing ETC gantry equipment, avoid repeated deployment of ETC equipment and other infrastructure in the same area, which leads to waste of resources, and reduce the comprehensive cost of issuing warning information, the release location of the slippery road warning information in the highway diversion area is set upstream of the diversion area, with a distance of more than 500 meters. This can not only effectively combine the layout of the existing ETC gantry, but also ensure that drivers receive sufficient early warning reminders within a reasonable distance, thereby improving driving safety and optimizing resource allocation.

[0113] S6: Push warning information to the vehicle end. The ETC gantry receives the optimal speed information for slippery roads pushed by the traffic management background, and sends the information to the vehicle's on-board OBU device with voice broadcast function, reminding the driver to slow down by voice.

[0114] The dynamic adaptive risk warning strategy for slippery roads in the diversion area monitors and obtains the rainfall and friction coefficient of slippery roads in the diversion area in real time through the traffic management background, considers the different models and speed information of vehicles near the ETC gantry, and combines the TPA-LSTM deep learning model to dynamically adjust and optimize the speed limit. At the same time, the linkage between the ETC gantry and the on-board OBU equipment is used to convey warning information to the driver in the form of voice. Specifically, this strategy:

[0115] In a slippery road scenario, when an OBU vehicle equipped with a voice broadcast function approaches the ETC gantry upstream of the highway diversion area, the ETC gantry exchanges information with the vehicle's OBU device through DSRC. During this process, the speedometer collects the vehicle's approaching speed time series data, and the ETC gantry obtains the vehicle's model information and uploads real-time data such as model, speed, acceleration, etc. to the traffic management backend.

[0116] The traffic management backend relies on the TPA-LSTM large model in the system, combined with the current environmental conditions and vehicle information of the diversion area, to predict the optimal safe speed of vehicles on slippery roads, and send the predicted speed limit information to the ETC gantry upstream of the diversion area. After receiving the speed limit information, the ETC gantry pushes the warning information to the vehicle OBU device through DSRC. The OBU device then informs the driver of the warning information through voice broadcast, prompting him to slow down safely.

[0117] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A dynamic warning and speed guidance method for slippery road types in diversion areas based on TPA-LSTM, characterized in that: The following steps are involved: S1. Collect traffic flow operation timing data in the slippery road scenario in the highway diversion area; S2: Predict the speed limits for different vehicle types to safely pass through slippery roads in the diversion area. Based on the TPA-LSTM deep learning model, combined with slippery road environment data and characteristics of different vehicle types, the optimal speed limit for each vehicle type is predicted in real time; S3. Formulate the content of early warning information release; S4. Determine the method of issuing warning information, and convey the warning information to the driver based on the functional characteristics of the ETC gantry and the voice broadcast function of the on-board OBU equipment; S5. Determine the information release location of the highway diversion area. The information release location is determined based on the existing ETC gantry layout principles of the diversion area and combined with the warning information release standards; S6: Push warning information to the vehicle end. The ETC gantry receives the optimal speed information for slippery roads pushed by the traffic management background, and sends the information to the vehicle's on-board OBU device with voice broadcast function, reminding the driver to slow down by voice.

2. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 1 is characterized in that: In step S1, the collected traffic flow operation time series data include timestamp, vehicle type, speed, acceleration, rainfall and road friction coefficient, and the collected traffic flow operation time series data types are divided into two categories: small passenger cars and large trucks.

3. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 2 is characterized in that: A speedometer is used to obtain the timestamp, speed and acceleration data, and the acceleration is obtained by performing a derivative calculation on the speed data; the rain gauge is used to collect real-time rainfall; and the road friction coefficient sensor is responsible for monitoring the road friction coefficient.

4. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 1 is characterized in that: In step S2, based on the TPA-LSTM deep learning model, the associated variables of rainfall and road friction coefficient in a slippery road environment are introduced into the model input variables, and the input traffic flow operation timing data is normalized using the Min-Max normalization method. The normalization formula is: Where, X normalized is the normalized value of sample X, x max is the maximum value of the sample data, X min is the minimum value of the sample data, The model uses time series features, including timestamp t, vehicle model s t , speed v t , acceleration a t The speed and acceleration of the car and truck are v tc 、v tt and a tc 、a tt , rainfall t and road friction coefficient μ t Input, train the model to predict the speed change at future moments. For each time step t, the input feature set HI t It is expressed as: HI t =[t,s t ,v t ,a t ,r t ,μ t ]; In order to effectively capture the temporal characteristics and regularities of the data, the time window size and time step are set. The time window T is used to include continuous historical time step data during prediction, and the time window length is set to 50, that is, {X t-T+1 , X t-T+2 , X t-T+3 ......, X t-T+50 } as input features, the time step is set to 1; The TPA-LSTM model is based on the traditional LSTM model. The key modules of the model include LSTM units and temporal attention mechanism. The structure includes input layer, LSTM layer, fully connected layer and output layer. The LSTM layer consists of two layers of stacked LSTM units, each layer contains 64 neurons, which extract the temporal characteristics of the input data. The fully connected layer is used to map the output of LSTM to the prediction result, and the output layer finally gives the predicted speed value. The forget gate f in the LSTM layer t The formula for determining whether information is retained is: f t =σ(W f ·[h t-1 ,x t ]+b f ); Input Gate i t Control the influence of the current input information, input gate i t and candidate memory cell c t The formulas are: i t =σ(W i ·[h t-1 ,x t ]+b i ); c t =tanhW c ·[h t-1 ,x t ]+b c ; Cell state update C t Combined with forget gate f t and input gate i t The update formula is: C t =f t ×C t-1 +i t ×c t ; Output gate o t Generate the output at the current moment, output gate o t and the hidden state h t The formulas are: the t =σ(W o ·[h t-1 ,x t ]+b o ); h t =o t fishy t ); In the formula, o t represents the sigmoid activation function, tanh represents the tanh activation function, W f , W i , W c and W o is the model parameter matrix; In the TPA-LSTM model, the temporal attention mechanism enhances the model's attention to key moments by dynamically adjusting the weights of time steps. First, according to the hidden state h of the current time step, t And the hidden states of other time steps, the attention weight α is calculated through nonlinear transformation t , the specific formula is as follows: In the formula, α t represents the attention weight at time step t, W, U and b are learnable parameters. Finally, the model performs a weighted average on the hidden state to generate a weighted hidden state H to capture the dynamic information of the key time step. The formula is: During the training process, the mean square error (MSE) loss function is used to measure the error between the predicted speed and the actual speed. The formula is: In the formula, v i is the true value, V i is the speed prediction value, n is the number of samples, Adam is selected as the optimizer, and the initial learning rate is set to 0.001 to improve the convergence speed and avoid falling into the local optimal solution. The early stopping strategy is adopted in the training process. If the loss function does not decrease significantly for 5 consecutive times on the test set, the training is stopped in advance to prevent overfitting. After the model training is completed, the root mean square error (RMSE) and mean absolute error (MAE) indicators are used to evaluate the generalization ability of the model through the test set. The root mean square error and mean absolute error calculation formulas are: After completing model training and verification, the TPA-LSTM model can be used for real-time speed prediction by combining the real-time collected timestamp t, speed v t , acceleration a t , rainfall t and road friction coefficient μ t Data, for each time step t, the feature set HI t =[t, s t , v t , a t , r t , μ t ] Input model, TPA-LSTM forward propagates the time series data through LSTM unit and time pattern attention mechanism to generate the speed prediction value of the next time step. Finally, the model outputs the predicted speed HI′ t =[V i ], used to push speed limit warning information to the vehicle.

5. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 1 is characterized in that: In step S3, the warning information content is constructed as: "The road surface is slippery at X meters ahead, please reduce the speed to X kilometers per hour." 6. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 1 is characterized in that: In step S4, the warning method uses an intelligent OBU device to receive the warning information sent by the ETC gantry, and uses the voice broadcast function of the OBU to broadcast the road danger to the driver in real time, reminding the driver to slow down in advance to ensure driving safety.

7. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 6 is characterized in that: In the warning method, the ETC gantry is set as the main receiving and sending device for warning information. When slippery road conditions are detected in the diversion area, the traffic management background will trigger the control link to generate warning information and send the information to the ETC gantry. The wireless communication module of the ETC gantry uses dedicated short-range communication technology to transmit the warning information to the on-board OBU device. The parsing module of the OBU device is started to process and decode the information content to identify the specific warning content.

8. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 1 is characterized in that: In step S5, the release location of the slippery road warning information in the highway diversion area is set to be upstream of the diversion area, which is greater than 500 meters away.

9. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 1 is characterized in that: In step S6, a dynamic adaptive risk warning strategy for slippery roads in the diversion area is adopted. The rainfall and road friction coefficient of the slippery roads in the diversion area are monitored and acquired in real time through the traffic management background. The different vehicle types and speed information of vehicles approaching the ETC gantry are considered, and the speed limit value is dynamically adjusted and optimized in combination with the TPA-LSTM deep learning model. The linkage between the ETC gantry and the on-board OBU equipment is utilized to convey the warning information to the driver in the form of voice.

10. The method for dynamic early warning and speed guidance of slippery road types in diversion areas based on TPA-LSTM according to claim 9, characterized in that: The specific method of the dynamic adaptive risk warning strategy for slippery road surface in the diversion area includes the following steps: S601. In a slippery road scenario, when an OBU equipped with a voice broadcast function approaches the ETC gantry upstream of the highway diversion area, the ETC gantry exchanges information with the vehicle's OBU device through DSRC. During this process, the speedometer collects the vehicle's approaching speed time series data, and the ETC gantry obtains the vehicle's model information and uploads the real-time data of the vehicle model, speed, and acceleration to the traffic management backend; S602. The traffic management backend relies on the TPA-LSTM large model in the system, combined with the current environmental conditions and vehicle information of the diversion area, to predict the optimal safe speed of the vehicle on slippery roads, and sends the predicted speed limit information to the ETC gantry upstream of the diversion area. After receiving the speed limit information, the ETC gantry pushes the warning information to the vehicle OBU device through DSRC. The OBU device then informs the driver of the warning information through voice broadcast, prompting him to slow down safely.

Images