A smart highway integrated management system based on the Internet of Things

Abstract

The present invention belongs to the field of artificial intelligence technology. The present invention discloses an intelligent highway integrated management system based on the Internet of Things, including: collecting lane speed limit data, vehicle speed data, vehicle condition data, highway environment data and lane specified speed data; based on vehicle status data, using a multi-task DNN model to predict emission health data, engine health data and tire health data, and using an adaptive weighting method to obtain vehicle diagnostic index data; setting a health threshold to determine whether the vehicle is healthy; using an SVR model to predict a safe speed based on the vehicle diagnostic index data and highway environment data of healthy vehicles; setting safe highway range data to determine whether healthy vehicles are speeding or driving at a snail's pace, and issuing early warning reminders; improving the traffic safety and environmental protection performance of highways, and also improving emergency response capabilities through intelligent speed limit management and real-time monitoring, and effectively reducing accident risks.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent highway integrated management system based on the Internet of Things. Background Art

[0002] The existing smart highway integrated management system has many problems. First, the data between different devices is often isolated. For example, on-board equipment and highway equipment are isolated and not integrated through IoT technology.

[0003] Second, existing smart highway systems rely solely on basic sensors to monitor vehicle health and lack in-depth health prediction and diagnostic capabilities.

[0004] Furthermore, existing smart highway systems typically only consider road speed limits or basic environmental parameters when predicting safe speeds, ignoring the vehicle's health during driving. This results in limited safe speed predictions.

[0005] In addition, existing speeding and slow speed detection mostly relies on single vehicle speed data, and lacks comprehensive consideration of lane speed limits and predicted safe speeds. This causes the system to misjudge the vehicle's driving status and is unable to issue accurate warnings.

[0006] In view of this, the present invention proposes an intelligent highway integrated management system based on the Internet of Things to solve the above problems. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solutions: a smart highway integrated management system based on the Internet of Things, comprising:

[0008] Data acquisition module: obtains lane speed limit data, vehicle speed data, vehicle condition data, high-speed environment data and lane speed data;

[0009] Vehicle Health Monitoring Module: Based on vehicle status data, a multi-task DNN model is used to predict emissions health data, engine health data, and tire health data. Based on the predicted emissions health data, engine health data, and tire health data, an adaptive weighting method is used to obtain vehicle diagnostic index data.

[0010] Health status management module: sets health thresholds based on historical vehicle diagnostic index data, compares vehicle diagnostic index data with health thresholds, and determines whether the vehicle is healthy;

[0011] Safe speed prediction module: Based on vehicle diagnostic index data of healthy vehicles and high-speed environment data, it uses the SVR model to predict the safe speed;

[0012] Highway warning module: The lane speed limit data and the predicted safe speed are averaged to obtain the safe highway range data, and the vehicle speed data is compared with the safe highway range data to determine whether a healthy vehicle is speeding or driving at a snail's pace, and issue a warning reminder.

[0013] Furthermore, the vehicle condition data includes emitted NOx concentration data, emitted CO2 concentration data, emitted CO concentration data, engine speed data, engine temperature data, engine load data, oil pressure data, vehicle acceleration data, brake fluid pressure data, brake temperature data, and brake pedal pressure data;

[0014] High-speed environmental data includes temperature data, humidity data, wind speed data, visibility data and traffic flow data.

[0015] Furthermore, the visibility data is obtained by:

[0016] A high-definition camera is used to collect real-time images, and the mean filtering method is used to denoise the collected images to obtain denoised images;

[0017] The denoised image is convolved using the Laplace operator to obtain the gradient of each pixel in the denoised image. The formula is: Among them, I(x,y) represents the pixel value of the image at position (x,y), x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, K(i,j) represents the convolution kernel of the Laplace operator, i is the row offset relative to the center position in the convolution kernel, j is the column offset relative to the center position in the convolution kernel, and I(x+i,y+j) represents the pixel value covered by the convolution kernel in the denoised image. Represents the denoised image gradient obtained after convolution;

[0018] The gradient of each pixel in the denoised image is averaged to obtain the sharpness index of the denoised image. The formula is: Where G*H represents the total number of pixels in the denoised image, G represents the width of the denoised image, H represents the height of the denoised image, and SI represents the sharpness index;

[0019] Visibility data is calculated by linear regression, and the calculation formula is: VJ = ap * SI+bp, where ap is the slope and bp is the intercept. The specific values ​​of ap and bp are calculated by the least squares method.

[0020] Furthermore, the specific method of using the multi-task DNN model to predict emission health data, engine health data, and tire health data based on vehicle status data includes:

[0021] Step F1: performing missing value, outlier and normalization processing on the vehicle status data to obtain pre-processed vehicle status data;

[0022] Step F2: input a sample set, including SR group samples, each group of samples including a pre-processed vehicle condition data and corresponding emission health data, engine health data and tire health data;

[0023] Step F3, initialize the input layer, hidden layer and output layer;

[0024] For the input layer, set the number of neurons to be equal to the number of input data; for the hidden layer, use grid search to set N hidden layers, and set M_J neurons in each hidden layer; for the output layer, set 3 neurons, which is the same as the number of tasks;

[0025] Initialize model weights and biases;

[0026] Step F4: Pass the sample from the input layer to the first hidden layer, and perform linear and ReLU activation functions to obtain the output of the first hidden layer. Pass the output of the first hidden layer to the next hidden layer, repeat the linear calculation and ReLU activation function until the last hidden layer is completed, and obtain the output of the last hidden layer. Pass the output of the last hidden layer to the output layer to obtain three task results: the current iteration prediction values ​​of emission health data, engine health data, and tire health data.

[0027] Step F5, loss calculation: For each task result output, use the mean square error as the loss function to calculate the error between the current iteration prediction value and the true value, and perform weighted summation to obtain the total loss function error. The formula is: Among them, rw is the task index, 3 is the total number of tasks, WC rw is the loss function error of the rwth task, λ rw is the weight of the rwth task, WC total is the total loss function, where the true value refers to the input emission health data, engine health data and tire health data;

[0028] Step F6: Use backpropagation to calculate the gradient of the total loss function error with respect to weights and biases layer by layer, and use the Adam optimizer to update the weights and biases of each layer;

[0029] Step F7: When the total loss function error is less than the set error threshold, the iteration is stopped and the final results of the three tasks are output, namely, predicted emission health data, predicted engine health data, and predicted tire health data.

[0030] Furthermore, the specific method of using the adaptive weighting method to obtain vehicle diagnostic index data includes:

[0031] Based on the predicted emission health data, predicted engine health data and predicted tire health data, the vehicle diagnostic index data is calculated using an adaptive weighting method. The formula is: H cl =a1*H emission +a2*H engine +a3*H tire , where H cl is the vehicle diagnostic index data, H emission 、H engine and H tire are the predicted emission health data, predicted engine health data and predicted tire health data, respectively. a1, a2 and a3 are the actual weights of the predicted emission health data, predicted engine health data and predicted tire health data, respectively. The actual weights of the predicted emission health data, predicted engine health data and predicted tire health data are obtained by the joint error method.

[0032] Furthermore, the specific method of obtaining the weights of the predicted emission health data, the predicted engine health data, and the predicted tire health data by the combined error method includes:

[0033] Calculate the error between the final result of each task and the actual data: Among them, H rw,m is the final result of the mth sample of the rwth task, is the true value of the mth sample of the rwth task, m is the sample index, MSE rw is the error of the rwth task, and the initial weight is adjusted according to the error of each task:

[0034] Calculate the error correlation between tasks and use the error correlation to correct the preliminary weight to get the actual weight. The formula is: Among them, corr rw,fw represents the error correlation between the rwth task and the fwth task, which is calculated by the Pearson correlation method. rw and fw are the indexes of the tasks. represents the initial weight gap between the rwth task and the fwth task, a rw represents the actual weight of the rwth task, that is, the actual weights of the three tasks a1, a2 and a3, and γ represents the adjustment coefficient, which is set by empirical method.

[0035] Furthermore, the specific method of setting a health threshold based on historical vehicle diagnostic index data and comparing the vehicle diagnostic index data with the health threshold to determine whether the vehicle is healthy includes:

[0036] Based on historical vehicle diagnostic indicator data, calculate the average and standard deviation to set the health threshold,

[0037] When the vehicle diagnostic index data is greater than or equal to the set health threshold, it means that the vehicle is in a healthy state and driving monitoring continues;

[0038] When the vehicle diagnostic index data is less than the set health threshold, it indicates that the vehicle is in an unhealthy state. An early warning is issued in the vehicle, triggering the on-board navigation to find the nearest service area for parking, and reminding other vehicles through vehicle network communication.

[0039] Furthermore, the specific method of using the SVR model to predict the safe speed based on the vehicle diagnostic index data of the healthy vehicle and the high-speed environment data includes:

[0040] The vehicle diagnostic index data and high-speed environment data are processed for missing values, outliers, normalization, and unified timestamps to obtain fully pre-processed vehicle diagnostic index data and high-speed environment data;

[0041] Input FG sample set, including N_n groups of FG samples, each group of FG samples includes a set of fully pre-processed vehicle diagnostic index data and high-speed environment data, as well as corresponding safe speed data;

[0042] Use grid search to initialize hyperparameters, including penalty coefficient, epsilon parameter, gamma parameter, and number of iterations; use random method to initialize weight vector qz0 and bias term pl0;

[0043] Calculate the initial loss function value of the model As the starting point for optimization, represents regularization, qz0 represents the initialization weight vector, pl0 represents the initialization bias term, represents the error term, C is the penalty coefficient, N_n is the total number of samples in the FG sample set, Q is the sample index, ξ represents the error of the Q-th group of samples on the upper deviation, ξ * It represents the error of the Qth group of samples on the lower deviation;

[0044] The LAMB model is used to update the weight vector and bias term in each iteration, and a new loss function is calculated during the update process;

[0045] After each round of iteration, the current loss function value is compared with the change in the previous round. When the loss change between two adjacent iterations is less than the preset error threshold, the iteration is stopped and the predicted safe speed data is output.

[0046] Furthermore, the specific method of using the LAMB model to update the weight vector and bias term includes:

[0047] Initialize the learning rate, first-order moment attenuation factor, and second-order moment attenuation factor; initialize the first-order moment and second-order moment to zero, and initialize the bias correction term to zero;

[0048] In each iteration, the gradient of the loss function with respect to the weight vector and the bias term is calculated;

[0049] Calculate the first and second moments of the weight vector and bias term based on the gradient;

[0050] Performing bias correction on the first-order moment and the second-order moment of the weight vector and the bias term to obtain the bias-corrected first-order moment and the bias-corrected second-order moment of the weight vector and the bias term;

[0051] The gradients of the weight vector and the bias term are normalized using the bias-corrected second-order moment to obtain the normalized gradients of the weight vector and the bias term;

[0052] Calculate the adaptive learning rate based on the first-order moment of bias correction to obtain the adaptive learning rate of the weight vector and bias term;

[0053] The updated weight vector and bias term are obtained according to their adaptive learning rates and their normalized gradients.

[0054] Furthermore, the lane speed limit data and the predicted safe speed are averaged to obtain safe high-speed range data, and the vehicle speed data is compared with the safe high-speed range data to determine whether a healthy vehicle is speeding or slowing down, and to issue an early warning. Specific methods include:

[0055] The lane speed limit data and the predicted safe speed are averaged to obtain the safe high-speed range data. The upper limit of the safe high-speed range data is: The lower limit data of the safe high-speed range is: Among them, GS upper The lane speed limit data specified for the lane where the vehicle is located, GS lower The lane speed limit data specified for the lane where the vehicle is located, AD safe To predict safe speed;

[0056] Compare the vehicle speed data with the safe high-speed range data. If the vehicle speed data exceeds the upper limit of the safe high-speed range, the vehicle is speeding and a deceleration warning is issued;

[0057] When the vehicle speed data is lower than the lower limit of the safe high-speed range, the vehicle is slowing down and an acceleration warning reminder is issued.

[0058] The technical effects and advantages of the present invention's intelligent highway integrated management system based on the Internet of Things are as follows:

[0059] This invention significantly improves highway management efficiency and safety through the integration of innovative technologies. The data acquisition module collects lane speed limit data, vehicle speed data, vehicle condition data, and highway environment data in real time. Combined with a multi-task DNN model, it accurately predicts vehicle health and enables comprehensive monitoring of the health of key components such as emissions, engines, and tires.

[0060] The system uses an adaptive weighting method to comprehensively process predicted data, ensuring the accuracy of vehicle diagnostic results. By setting health thresholds based on historical vehicle data, the system can determine in real time whether the vehicle is in a healthy state and issue warnings when an anomaly is detected, ensuring that the driver can take timely action.

[0061] For healthy vehicles, the system combines the SVR model to predict the vehicle's safe speed, and calculates the safe driving range based on the lane speed limit and the predicted safe speed, providing real-time warnings for speeding or slow driving. This not only improves the safety and environmental protection of the vehicle, but also realizes the intelligent speed limit management of highways based on the Internet of Things, enhances the synergy between vehicles and the road environment, and thus greatly improves the intelligence level and emergency response capabilities of highways. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a schematic diagram of an intelligent highway integrated management system based on the Internet of Things of the present invention;

[0063] Figure 2 This is a schematic diagram of an intelligent highway comprehensive management method based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0065] Example 1, please refer to Figure 1 As shown, the embodiment of the present invention provides an IoT-based intelligent highway integrated management system, including:

[0066] Data acquisition module: obtains lane speed limit data, vehicle speed data, vehicle condition data, high-speed environment data and lane speed data;

[0067] Vehicle Health Monitoring Module: Based on vehicle status data, a multi-task DNN model is used to predict emissions health data, engine health data, and tire health data. Based on the predicted emissions health data, engine health data, and tire health data, an adaptive weighting method is used to obtain vehicle diagnostic index data.

[0068] Health status management module: sets health thresholds based on historical vehicle diagnostic index data, compares vehicle diagnostic index data with health thresholds, and determines whether the vehicle is healthy;

[0069] Safe speed prediction module: Based on vehicle diagnostic index data of healthy vehicles and high-speed environment data, it uses the SVR model to predict the safe speed;

[0070] Highway warning module: The lane speed limit data and the predicted safe speed are averaged to obtain the safe highway range data, and the vehicle speed data is compared with the safe highway range data to determine whether a healthy vehicle is speeding or driving at a snail's pace, and issue a warning reminder.

[0071] Lane speed limit data: The lane keeping system uses the vehicle's built-in system to obtain the specified lane speed limit data and transmit it to the data acquisition module. The lane keeping system includes upper and lower speed limits. For example, the upper speed limit of the innermost lane is 120 km / h and the lower speed limit is 110 km / h.

[0072] Vehicle condition data includes NOx concentration data, CO2 concentration data, CO concentration data, engine speed data, engine temperature data, engine load data, oil pressure data, vehicle acceleration data, brake fluid pressure data, brake temperature data, and brake pedal pressure data;

[0073] The vehicle speed data, NOx concentration data, CO2 concentration data, CO concentration data, engine speed data, engine temperature data, engine load data, oil pressure data, vehicle acceleration data, brake fluid pressure data, brake temperature data, and brake pedal pressure data are acquired in real time through the on-board diagnostic interface, and the vehicle data is transmitted to the data acquisition module through the connection with the Internet of Vehicles.

[0074] High-speed environmental data includes temperature data, humidity data, wind speed data, visibility data, and traffic flow data;

[0075] By installing a temperature sensor, a humidity sensor, a wind speed sensor, a high-definition camera, and a radar sensor every 1 km along the highway, real-time temperature data, humidity data, wind speed data, images, and traffic flow data are collected;

[0076] Visibility data can be obtained by:

[0077] A high-definition camera is used to collect real-time images, and the mean filtering method is used to denoise the collected images to obtain denoised images;

[0078] The denoised image is convolved using the Laplace operator to obtain the gradient of each pixel in the denoised image. The formula is: Among them, I(x,y) represents the pixel value of the image at position (x,y), x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, K(i,j) represents the convolution kernel of the Laplace operator, i is the row offset relative to the center position in the convolution kernel, j is the column offset relative to the center position in the convolution kernel, and I(x+i,y+j) represents the pixel value covered by the convolution kernel in the denoised image. Represents the denoised image gradient obtained after convolution;

[0079] The gradient of each pixel in the denoised image is averaged to obtain the sharpness index of the denoised image. The formula is: Where G*H represents the total number of pixels in the denoised image, G represents the width of the denoised image, H represents the height of the denoised image, and SI represents the sharpness index;

[0080] The sharpness index is verified to be linearly related to visibility through existing references. The visibility data is calculated by linear regression using the following formula: VJ = ap*SI + bp, where ap is the slope and bp is the intercept. The specific values ​​of ap and bp are obtained by the least squares method.

[0081] The gradient reflects the degree of change of each pixel in the denoised image, that is, the sharpness of the image. For the gradient of each pixel, a larger gradient value usually means that there are obvious edges or details at that location, while a smaller gradient value indicates that the denoised image area is relatively smooth or uniform. Therefore, by calculating the average value of the gradients of all pixels, the sharpness index of the entire image can be obtained, that is, the richness or clarity of the image details. The larger the average value of the gradient in the image, the higher the image details and sharpness, and thus the larger the sharpness index obtained, which reflects the clarity or visibility of the image.

[0082] By capturing images with high-definition cameras and combining them with advanced image processing techniques such as mean filtering, Laplace operator, and gradient calculation, the sharpness index of the image can be accurately extracted, thereby effectively evaluating visibility. This method not only reduces hardware costs but also can obtain high-precision visibility data in real time and non-invasively. At the same time, a linear regression model is used to associate image features with visibility data, further improving the accuracy and robustness of the prediction, making it particularly suitable for real-time monitoring in complex environments.

[0083] Based on vehicle status data, the specific methods of using a multi-task DNN model to predict emission health data, engine health data, and tire health data include:

[0084] Step F1: performing missing value, outlier and normalization processing on the vehicle status data to obtain pre-processed vehicle status data;

[0085] Step F2: input a sample set, including SR group samples, each group of samples including a pre-processed vehicle condition data and corresponding emission health data, engine health data and tire health data;

[0086] Step F3, initialize the input layer, hidden layer and output layer;

[0087] For the input layer, set the number of neurons to be equal to the number of input data; for the hidden layer, use grid search to set N hidden layers, and set M_J neurons in each hidden layer; for the output layer, set 3 neurons, which is the same as the number of tasks;

[0088] Initialize model weights and biases;

[0089] Step F4: Pass the sample from the input layer to the first hidden layer, and calculate the output of the first hidden layer through the linear and ReLU activation functions. The formula is: 1 =ReLU(W 1 X+b 1 ), where W 1 is the weight of the first hidden layer, X is the input sample, b 1 is the bias term of the first hidden layer, h 1 is the output of the first hidden layer;

[0090] The output of the first hidden layer is passed to the next hidden layer, and the linear calculation and ReLU activation function are repeated until the last hidden layer ends. The formula is: 2 =ReLU(W 2 h 1 +b 2 ), where W 2 is the weight of the second hidden layer, b 2 is the bias term of the second hidden layer, h 2 is the output of the second hidden layer;

[0091] The output of the hidden layer is passed to the output layer to obtain three task results, which are the predicted values ​​of emission health data, engine health data, and tire health data in the current iteration. The formula is: in, is the predicted value output by the current iteration, i.e., the three task results, is the current iteration prediction value of the emission health data, is the current iteration prediction value of the engine health data, is the current iteration prediction value of tire health data, W out is the weight of the output layer, b out is the bias term of the output layer;

[0092] Step F5, loss calculation: For each task result output, use the mean square error as the loss function to calculate the error between the current iteration prediction value and the true value, and perform weighted summation to obtain the total loss function error. The formula is: Among them, rw is the task index, 3 is the total number of tasks, WC rw is the loss function error of the rwth task, λ rw is the weight of the rwth task, WC total is the total loss function, where the true value refers to the input emission health data, engine health data and tire health data;

[0093] Step F6: Use backpropagation to calculate the gradient of the total loss function error with respect to weights and biases layer by layer, and use the Adam optimizer to update the weights and biases of each layer;

[0094] Step F7: When the total loss function error is less than the set error threshold, stop the iteration and output the final results of the three tasks, namely, predicted emission health data, predicted engine health data, and predicted tire health data; the error threshold is obtained by taking the average value through Q_I experiments;

[0095] Multiple task DNN models can jointly learn potential correlations, improve prediction accuracy, and reduce the risk of overfitting. In addition, multi-task learning improves data utilization efficiency, reduces the consumption of computing resources, and reduces the complexity of model deployment and maintenance.

[0096] The specific methods of obtaining vehicle diagnostic index data using the adaptive weighting method include:

[0097] Based on the predicted emission health data, predicted engine health data and predicted tire health data, the vehicle diagnostic index data is calculated using an adaptive weighting method. The formula is: H cl =a1*H emission +a2*H engine +a3*H tire , where H cl is the vehicle diagnostic index data, H emission 、H engine and H tireare the predicted emission health data, predicted engine health data and predicted tire health data, respectively. a1, a2 and a3 are the actual weights of the predicted emission health data, predicted engine health data and predicted tire health data, respectively. The actual weights of the predicted emission health data, predicted engine health data and predicted tire health data are obtained by the joint error method.

[0098] The specific method of obtaining the weights of the predicted emission health data, the predicted engine health data, and the predicted tire health data by the joint error method includes:

[0099] Calculate the final result of each task, that is, the error between the predicted value and the actual data: Among them, H rw,m is the final result of the mth sample of the rwth task, is the true value of the mth sample of the rwth task, m is the sample index, MSE rw is the error of the rwth task, and the initial weight is adjusted according to the error of each task:

[0100] Calculate the error correlation between tasks and use the error correlation to correct the preliminary weight to get the actual weight. The formula is: Among them, corr rw,fw represents the error correlation between the rwth task and the fwth task, which is calculated by the Pearson correlation method. rw and fw are the indexes of the tasks. represents the initial weight gap between the rwth task and the fwth task, a rw represents the actual weight of the rwth task, i.e., the actual weight of predicting emission health data, predicting engine health data, and predicting tire health data. γ represents the adjustment coefficient, which is set by empirical method.

[0101] Calculating the error of each task and the error correlation between tasks can more accurately evaluate the relative importance of each task. Using error correlation to correct the initial weights helps reduce interference and deviation between tasks and improve the overall quality of the fusion results.

[0102] The health threshold is set based on historical vehicle diagnostic index data. The vehicle diagnostic index data is compared with the health threshold to determine whether the vehicle is healthy. The specific methods include:

[0103] Based on historical vehicle diagnostic indicator data, calculate the average and standard deviation to set health thresholds;

[0104] When the vehicle diagnostic index data is greater than or equal to the set health threshold, it means that the vehicle is in a healthy state and driving monitoring continues;

[0105] When the vehicle diagnostic index data is less than the set health threshold, it indicates that the vehicle is in an unhealthy state. An early warning is issued in the car, triggering the on-board navigation to find the nearest service area for parking, and reminding other vehicles through vehicle network communication.

[0106] Based on vehicle diagnostic index data of healthy vehicles and highway environment data, the specific methods of using the SVR model to predict safe speed include:

[0107] The vehicle diagnostic index data and high-speed environment data are processed for missing values, outliers, normalization, and unified timestamps to obtain fully pre-processed vehicle diagnostic index data and high-speed environment data;

[0108] Input FG sample set, including N_n groups of FG samples, each group of FG samples includes a set of fully pre-processed vehicle diagnostic index data and high-speed environment data, as well as corresponding safe speed data;

[0109] Use grid search to initialize hyperparameters, including penalty coefficient, epsilon parameter, gamma parameter, and number of iterations; use random method to initialize weight vector qz0 and bias term pl0;

[0110] Calculate the initial loss function value of the model As the starting point for optimization, represents regularization, qz0 represents the initialization weight vector, pl0 represents the initialization bias term, represents the error term, C is the penalty coefficient, N_n is the total number of samples in the FG sample set, Q is the sample index, ξ represents the error of the Q-th group of samples on the upper deviation, ξ * It represents the error of the Qth group of samples on the lower deviation;

[0111] The LAMB model is used to update the weight vector and bias term in each iteration, and a new loss function is calculated during the update process;

[0112] After each iteration, the current loss function value is compared with the change in the previous round. When the loss change between two adjacent iterations is less than the preset error threshold, the iteration is stopped and the predicted safe speed data is output. The preset error threshold is obtained through experimental calculation.

[0113] The specific methods of using the LAMB model to update the weight vector and bias terms include:

[0114] Randomly initialize the learning rate and the number of LAMB model iterations; initialize the first-order moment and the second-order moment to zero, and initialize the bias correction term to zero;

[0115] In each iteration, the gradient of the loss function with respect to the weight vector and the bias term is calculated as follows: Among them, t is the index of the number of iterations of the LAMB model, g t,qz and g t,pl are the gradients of the weight vector and bias term at the tth iteration, is the partial derivative of the loss function with respect to the weight vector and bias term in the tth round;

[0116] The first-order moment and second-order moment of the weight vector and bias term are calculated based on the gradient. The first-order moment formula is: M t,qz =β1M t,qz-1 +(1-β1)g t,qz , M t,pl =β1M t,pl-1 +(1-β1)g t,pl , where M t,qz and M t,pl are the weight vector and the first-order moment of the bias term at the tth iteration, M t,qz-1 and M t,pl-1 are the weight vector and the first-order moment of the bias term at the t-1th iteration, respectively. β1 is the attenuation factor of the first-order moment, which is set to 0.9 according to the reference;

[0117] The second-order moment formula is: Among them, v t,qz and v t,pl are the weight vector and the second-order moment of the bias term at the tth iteration, v t,qz-1 and v t,pl-1 are the weight vector and the second-order moment of the bias term at the t-1th iteration, respectively. β2 is the attenuation factor of the second-order moment, which is set to 0.999 according to the reference;

[0118] Among them, the first-order moment reflects the trend of the gradient, and the second-order moment reflects the amplitude of the gradient change;

[0119] The first-order moment and second-order moment of the weight vector and the bias term are bias-corrected to obtain the bias-corrected first-order moment and bias-corrected second-order moment of the weight vector and the bias term. The calculation formula for the bias-corrected first-order moment is: and are the weight vector and bias term bias correction first-order moment of the t-th iteration respectively;

[0120] The calculation formula for the bias-corrected second-order moment is: and are the bias-corrected second-order moments of the weight vector and bias term at the t-th iteration, respectively;

[0121] The gradient of the weight vector and the bias term is normalized using the bias-corrected second-order moment to obtain the normalized gradient of the weight vector and the bias term. The formula is: g′ t,qz and g′ t,pl are the normalized gradients of the weight vector and bias term at the tth iteration, ∈ is a constant, which is set to 10 according to relevant literature. -8 , used to avoid division by zero errors;

[0122] The adaptive learning rate is calculated based on the first-order moment of deviation correction to obtain the adaptive learning rate of the weight vector and the bias term. The formula is: Among them, ||qz t || is the weight vector qz t The L2 norm of and The L2 norm of the first-order moment of the weight vector and bias term deviation correction; η t,qz and η t,pl are the adaptive learning rates of the weight vector and bias term respectively;

[0123] The updated weight vector and bias term are obtained according to the adaptive learning rate of the weight vector and bias term and the normalized gradient of the weight vector and bias term. The formula is: in, and are the updated weight vector and bias term respectively;

[0124] The LAMB optimizer can dynamically adjust the learning rate by combining adaptive adjustment of the first-order and second-order moments, thereby avoiding the slow convergence or instability caused by the fixed learning rate in SVR; LAMB improves the stability and accuracy of weight updates through gradient normalization and bias correction, especially on high-dimensional and complex datasets, which can more effectively accelerate convergence and improve model performance; in addition, the LAMB optimizer can better cope with changes in different tasks and data, provide more flexible and efficient optimization strategies, and can find the global optimal solution more quickly.

[0125] The lane speed limit data and the predicted safe speed are averaged to obtain safe high-speed range data. The vehicle speed data is compared with the safe high-speed range data to determine whether healthy vehicles are speeding or slowing down. Specific methods for issuing early warnings include:

[0126] The lane speed limit data and the predicted safe speed are averaged to obtain the safe high-speed range data. The upper limit of the safe high-speed range data is: The lower limit data of the safe high-speed range is: Among them, GS upper The lane speed limit data specified for the lane where the vehicle is located, GS lower The lane speed limit data specified for the lane where the vehicle is located, AD safe To predict safe speed;

[0127] Compare the vehicle speed data with the safe high-speed range data. If the vehicle speed data exceeds the upper limit of the safe high-speed range, the vehicle is speeding and a deceleration warning is issued;

[0128] When the vehicle speed data is lower than the lower limit of the safe high-speed range, the vehicle is slowing down and an acceleration warning is issued;

[0129] By averaging lane speed limit data and predicted safe speeds to obtain safe high-speed range data, the system can more accurately consider actual road conditions and the needs of safe vehicle driving. Compared with simply using fixed lane speed limits or predicted safe speeds, this method can dynamically adjust the safe driving range based on different roads and actual vehicle conditions, providing a more intelligent and personalized early warning mechanism. When the vehicle speed exceeds the safe high-speed range, the system can promptly remind the driver to take corresponding safety measures to avoid speeding or slow driving, thereby improving driving safety and reducing the risk of traffic accidents.

[0130] This embodiment significantly improves highway management efficiency and safety through the integration of innovative technologies. The data acquisition module collects lane speed limit data, vehicle speed data, vehicle condition data, and highway environment data in real time. Combined with a multi-task DNN model, it accurately predicts vehicle health and enables comprehensive monitoring of the health of key components such as emissions, engines, and tires.

[0131] The system uses an adaptive weighting method to comprehensively process predicted data, ensuring the accuracy of vehicle diagnostic results. By setting health thresholds based on historical vehicle data, the system can determine in real time whether the vehicle is in a healthy state and issue warnings when an anomaly is detected, ensuring that the driver can take timely action.

[0132] For healthy vehicles, the system combines the SVR model to predict the vehicle's safe speed, and calculates the safe driving range based on the lane speed limit and the predicted safe speed, providing real-time warnings for speeding or slow driving. This not only improves the safety and environmental protection of the vehicle, but also realizes the intelligent speed limit management of highways based on the Internet of Things, enhances the synergy between vehicles and the road environment, and thus greatly improves the intelligence level and emergency response capabilities of highways.

[0133] Example 2, please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A smart highway integrated management method based on the Internet of Things is provided, including:

[0134] Step S1, collecting lane speed limit data, vehicle speed data, vehicle condition data, high-speed environment data and lane prescribed speed data;

[0135] Step S2: Based on the vehicle status data, a multi-task DNN model is used to predict emission health data, engine health data, and tire health data; based on the predicted emission health data, engine health data, and tire health data, an adaptive weighting method is used to obtain vehicle diagnostic index data;

[0136] Step S3: Setting a health threshold based on historical vehicle diagnostic index data, comparing the vehicle diagnostic index data with the health threshold to determine whether the vehicle is healthy;

[0137] Step S4: using the SVR model to predict the safe speed based on the vehicle diagnostic index data and high-speed environment data of the healthy vehicle;

[0138] Step S5: Average the lane speed limit data and the predicted safe speed to obtain safe high-speed range data, compare the vehicle speed data with the safe high-speed range data, determine whether a healthy vehicle is speeding or driving at a snail's pace, and issue an early warning.

[0139] Example 3. This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned smart highway integrated management system based on the Internet of Things is implemented.

[0140] Since the electronic device introduced in this embodiment is an electronic device used to implement a smart highway integrated management system based on the Internet of Things in the embodiment of this application, based on the smart highway integrated management system based on the Internet of Things introduced in the embodiment of this application, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not introduced in detail here. As long as technical personnel in this field implement the electronic device used in a smart highway integrated management system based on the Internet of Things in the embodiment of this application, it falls within the scope of protection of this application.

[0141] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0142] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A smart highway integrated management system based on the Internet of Things, characterized by: include: Data acquisition module: obtains lane speed limit data, vehicle speed data, vehicle condition data, high-speed environment data and lane speed data; Vehicle health monitoring module: Based on vehicle status data, a multi-task DNN model is used to predict emission health data, engine health data, and tire health data; Based on the predicted emission health data, engine health data and tire health data, an adaptive weighting method is used to obtain vehicle diagnostic index data; Health status management module: sets health thresholds based on historical vehicle diagnostic index data, compares vehicle diagnostic index data with health thresholds, and determines whether the vehicle is healthy; Safe speed prediction module: Based on vehicle diagnostic index data of healthy vehicles and high-speed environment data, it uses the SVR model to predict the safe speed; Highway Warning Module: This module averages lane speed limit data and predicted safe speed to obtain safe highway range data. It then compares vehicle speed data with the safe highway range data to determine whether healthy vehicles are speeding or slowing down, and issues warnings. The specific method of obtaining vehicle diagnostic index data using the adaptive weighting method includes: Based on the predicted emission health data, predicted engine health data and predicted tire health data, the vehicle diagnostic index data is calculated using an adaptive weighting method. The formula is: H cl =a1*H emission +a2*H engine +a3*H tire , where H cl is the vehicle diagnostic index data, H enission 、H engine and H tire are the predicted emission health data, the predicted engine health data, and the predicted tire health data, respectively; a1, a2, and a3 are the actual weights of the predicted emission health data, the predicted engine health data, and the predicted tire health data, respectively; wherein the actual weights of the predicted emission health data, the predicted engine health data, and the predicted tire health data are obtained by the joint error method; The specific method of obtaining the weights of the predicted emission health data, the predicted engine health data, and the predicted tire health data by the combined error method includes: Calculate the error between the final result of each task and the actual data: Among them, H rw,m is the final result of the mth sample of the rwth task, is the true value of the mth sample of the rwth task, m is the sample index, MSE rw is the error of the rwth task, and the initial weight is adjusted according to the error of each task: Calculate the error correlation between tasks and use the error correlation to correct the preliminary weight to get the actual weight. The formula is: Among them, corr rw,fw represents the error correlation between the rwth task and the fwth task, which is calculated by the Pearson correlation method. rw and fw are the indexes of the tasks. represents the initial weight gap between the rwth task and the fwth task, a rw represents the actual weight of the rwth task, that is, the actual weights of the three tasks a1, a2 and a3, and γ represents the adjustment coefficient.

2. The intelligent highway integrated management system based on the Internet of Things according to claim 1 is characterized in that: The vehicle condition data includes NOx concentration data, CO2 concentration data, CO concentration data, engine speed data, engine temperature data, engine load data, oil pressure data, vehicle acceleration data, brake fluid pressure data, brake temperature data, and brake pedal pressure data; High-speed environmental data includes temperature data, humidity data, wind speed data, visibility data and traffic flow data.

3. The intelligent highway integrated management system based on the Internet of Things according to claim 2 is characterized in that: The visibility data is obtained by: A high-definition camera is used to collect real-time images, and the mean filtering method is used to denoise the collected images to obtain denoised images; The denoised image is convolved using the Laplace operator to obtain the gradient of each pixel in the denoised image. The formula is: Among them, I(x,y) represents the pixel value of the image at position (x,y), x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, K(i,j) represents the convolution kernel of the Laplace operator, i is the row offset relative to the center position in the convolution kernel, j is the column offset relative to the center position in the convolution kernel, and I(x+i,y+j) represents the pixel value covered by the convolution kernel in the denoised image. Represents the denoised image gradient obtained after convolution; The gradient of each pixel in the denoised image is averaged to obtain the sharpness index of the denoised image. The formula is: Where G*H represents the total number of pixels in the denoised image, G represents the width of the denoised image, H represents the height of the denoised image, and SI represents the sharpness index; Visibility data is calculated by linear regression, and the calculation formula is: VJ = ap * SI+bp, where ap is the slope and bp is the intercept. The specific values ​​of ap and bp are calculated by the least squares method.

4. The intelligent highway integrated management system based on the Internet of Things according to claim 3 is characterized in that: The specific method of using the multi-task DNN model to predict emission health data, engine health data, and tire health data based on vehicle status data includes: Step F1: performing missing value, outlier and normalization processing on the vehicle condition data to obtain pre-processed vehicle condition data; Step F2: input a sample set, including SR group samples, each group of samples including a pre-processed vehicle condition data and corresponding emission health data, engine health data and tire health data; Step F3, initialize the input layer, hidden layer and output layer; For the input layer, set the number of neurons to be equal to the number of input data. For the hidden layer, use grid search to set N hidden layers, with M_J neurons in each hidden layer. For the output layer, set 3 neurons, which is the same as the number of tasks. Initialize the model weights and biases. Step F4: Pass the sample from the input layer to the first hidden layer, and perform linear and ReLU activation functions to obtain the output of the first hidden layer. Pass the output of the first hidden layer to the next hidden layer, repeat the linear calculation and ReLU activation function until the last hidden layer is completed, and obtain the output of the last hidden layer. Pass the output of the last hidden layer to the output layer to obtain three task results: the current iteration prediction values ​​of emission health data, engine health data, and tire health data. Step F5, Loss Calculation: For each task result output, use the mean square error as the loss function to calculate the error between the current iteration prediction value and the true value, and perform a weighted sum to obtain the total loss function error. The true value refers to the input emission health data, engine health data, and tire health data. Step F6: Use backpropagation to calculate the gradient of the total loss function error with respect to weights and biases layer by layer, and use the Adam optimizer to update the weights and biases of each layer; Step F7: When the total loss function error is less than the set error threshold, the iteration is stopped and the final results of the three tasks are output, namely, predicted emission health data, predicted engine health data, and predicted tire health data.

5. The intelligent highway integrated management system based on the Internet of Things according to claim 4 is characterized in that: The specific method of setting a health threshold based on historical vehicle diagnostic index data and comparing the vehicle diagnostic index data with the health threshold to determine whether the vehicle is healthy includes: Based on historical vehicle diagnostic indicator data, calculate the average and standard deviation to set the health threshold, When the vehicle diagnostic index data is greater than or equal to the set health threshold, it means that the vehicle is in a healthy state and driving monitoring continues; When the vehicle diagnostic index data is less than the set health threshold, it indicates that the vehicle is in an unhealthy state. An early warning is issued in the vehicle, triggering the on-board navigation to find the nearest service area for parking, and reminding other vehicles through vehicle network communication.

6. The intelligent highway integrated management system based on the Internet of Things according to claim 5 is characterized in that: The specific method of using the SVR model to predict the safe speed based on the vehicle diagnostic index data of the healthy vehicle and the high-speed environment data includes: The vehicle diagnostic index data and high-speed environment data are processed for missing values, outliers, normalization, and unified timestamps to obtain fully pre-processed vehicle diagnostic index data and high-speed environment data; Input FG sample set, including N_n groups of FG samples, each group of FG samples includes a set of fully pre-processed vehicle diagnostic index data and high-speed environment data, as well as corresponding safe speed data; Use grid search to initialize hyperparameters, including penalty coefficient, epsilon parameter, gamma parameter, and number of iterations; use random method to initialize weight vector qz L and the bias term pl L ; Calculate the initial loss function value of the model As the starting point for optimization, represents regularization, qz0 represents the initialization weight vector, pl0 represents the initialization bias term, represents the error term, C is the penalty coefficient, N_n is the total number of samples in the FG sample set, Q is the sample index, ξ represents the error of the Q-th group of samples on the upper deviation, ξ * It represents the error of the Qth group of samples on the lower deviation; The LAMB model is used to update the weight vector and bias term in each iteration, and a new loss function is calculated during the update process; After each round of iteration, the current loss function value is compared with the change in the previous round. When the loss change between two adjacent iterations is less than the preset error threshold, the iteration is stopped and the predicted safe speed data is output.

7. The intelligent highway integrated management system based on the Internet of Things according to claim 6 is characterized in that: The specific method of using the LAMB model to update the weight vector and bias term includes: Initialize the learning rate, first-order moment attenuation factor, and second-order moment attenuation factor; initialize the first-order moment and second-order moment to zero, and initialize the bias correction term to zero; In each iteration, the gradient of the loss function with respect to the weight vector and the bias term is calculated; Calculate the first and second moments of the weight vector and bias term based on the gradient; Performing bias correction on the first-order moment and the second-order moment of the weight vector and the bias term to obtain the bias-corrected first-order moment and the bias-corrected second-order moment of the weight vector and the bias term; The gradients of the weight vector and the bias term are normalized using the bias-corrected second-order moment to obtain the normalized gradients of the weight vector and the bias term; Calculate the adaptive learning rate based on the first-order moment of bias correction to obtain the adaptive learning rate of the weight vector and bias term; The updated weight vector and bias term are obtained according to their adaptive learning rates and their normalized gradients.

8. The intelligent highway integrated management system based on the Internet of Things according to claim 7 is characterized in that: The specific method of averaging the lane speed limit data and the predicted safe speed to obtain the safe high-speed range data, comparing the vehicle speed data with the safe high-speed range data, determining whether a healthy vehicle is speeding or slowing down, and issuing an early warning includes: The lane speed limit data and the predicted safe speed are averaged to obtain the safe high-speed range data. The upper limit of the safe high-speed range data is: The lower limit data of the safe high-speed range is: Among them, GS upper The lane speed limit data specified for the lane where the vehicle is located, GS lower The lane speed limit data specified for the lane where the vehicle is located, AD safe To predict safe speed; Compare the vehicle speed data with the safe high-speed range data. If the vehicle speed data exceeds the upper limit of the safe high-speed range, the vehicle is speeding and a deceleration warning is issued; When the vehicle speed data is lower than the lower limit of the safe high-speed range, the vehicle is slowing down and an acceleration warning reminder is issued.

Images