Detection device and prediction method for predicting icing time of water film on road surface

Abstract

The invention discloses a detection device and a prediction method for predicting icing time of a pavement water film, and belongs to the technical field of sensors. The detection device comprises an icing detection sensor used for detecting whether the water film on the road surface is iced or not, a temperature sensor used for detecting the temperature of the contact face of the road surface and the water film, a water film thickness sensor used for detecting the thickness of the water film on the road surface and a processor. The microprocessor is used for collecting sensing data of the icing detection sensor, the temperature sensor and the water film thickness sensor and outputting the sensing data. The prediction method comprises the following steps: constructing an icing time prediction model based on a CNN-GRU network and training the icing time prediction model; and icing time prediction is carried out by adopting the trained model and sensing data. According to the invention, the thickness of the water film and the temperature of the contact surface of the pavement water film can be accurately measured, and efficient and accurate icing time prediction and safety early warning can be carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of sensors, and in particular, to a detection device and a prediction method for predicting the icing time of a road surface water film. Background Art

[0002] Weather conditions have an important impact on the icing process and ice layer formation. Especially in a low-temperature environment, the thickness of the water film and temperature changes directly determine the icing speed. Once the ice layer is formed, it may have a significant impact on traffic safety, the stability of building facilities, etc. Therefore, real-time monitoring and accurate prediction of the icing time to improve safety. Currently, there are some icing warning systems based on temperature and humidity, but the accuracy and real-time performance of these systems still have certain deficiencies. They often rely on single weather data or simplified physical models, resulting in a certain increase in the deviation of the prediction results. Traditional warning methods mostly judge the icing risk through weather forecasts. Although they can provide a reference, they lack real-time and location measurements. They can only roughly predict the upcoming icing within a certain city range and cannot accurately predict the icing time of a certain road section according to the actual water film thickness and temperature changes of that road section. In recent years, some studies have begun to combine sensor technology to monitor the water film and temperature and use this data for icing prediction. However, the existing sensor technology still faces some challenges, such as insufficient accuracy in data collection, difficulty in effectively distinguishing the icing process under different water film thicknesses, and a lack of sufficient optimization algorithms in practical applications to improve the prediction accuracy. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a detection device and a prediction method for predicting the icing time of a road surface water film, which can accurately measure the water film thickness and the contact surface temperature of the road surface water film, and perform efficient and accurate icing time prediction and safety warning.

[0004] To achieve the above object, the present invention is implemented by the following technical solutions:

[0005] In a first aspect, the present invention provides a detection device for predicting the icing time of a road surface water film, including:

[0006] An icing detection sensor for detecting whether the water film on the road surface is frozen;

[0007] A temperature sensor for detecting the contact surface temperature of the road surface and the water film;

[0008] A water film thickness sensor for detecting the water film thickness on the road surface;

[0009] A microprocessor for collecting the sensing data of the icing detection sensor, the temperature sensor, and the water film thickness sensor and outputting them.

[0010] Optionally, the icing detection sensor uses a ring-shaped columnar capacitor, which includes an electrode column and an electrode ring coaxially sleeved outside the electrode column. A low-dielectric constant material is filled between the electrode ring and the electrode column, and the filled low-dielectric constant material serves as the excitation end, while the electrode ring and the electrode column serve as the grounding end.

[0011] Optionally, the temperature sensor uses a PT100 platinum resistance, and the contact temperature between the road surface and the water film is obtained by measuring the resistance value of the PT100 platinum resistance;

[0012]

[0013] In the formula, is the resistance value of the PT100 platinum resistance at temperature and 0 °C, is the temperature coefficient of the PT100 platinum resistance.

[0014] Optionally, the water film thickness sensor is wrapped in an electromagnetic shielding cover, and includes a substrate and a coaxial feed patch antenna attached to the substrate. The transmitting end of the coaxial feed patch antenna emits a microwave signal, and the microwave signal is reflected by the water film and then returns to the receiving end of the coaxial feed patch antenna to calculate the water film thickness of the road surface :

[0015]

[0016] In the formula, is the horizontal distance from the transmitting end to the receiving end, is the microwave speed and frequency, is the phase change amount of the microwave signal from the transmitting end through the water film reflection to the receiving end.

[0017] Optionally, it further includes a PPO housing and a metal protective housing. The icing detection sensor, the temperature sensor, and the water film thickness sensor are arranged in the PPO housing, the PPO housing is embedded in the metal protective housing, and the detection surfaces of the icing detection sensor and the temperature sensor are flush with the upper surfaces of the PPO housing and the metal protective housing.

[0018] In a second aspect, the present invention provides a method for predicting the icing time of a road surface water film. Based on the above detection device, the prediction method includes:

[0019] Obtain the contact temperature and the water film thickness, and generate sample data by combining the voltage analog quantities output by the temperature sensor and the water film thickness sensor;

[0020] Construct an icing time prediction model based on the CNN-GRU network;

[0021] Divide the sample data into a training set and a test set, and train and test the icing time prediction model with the training set and the test set;

[0022] Use the trained icing time prediction model to predict the icing time of the road surface water film.

[0023] Optionally, the icing time prediction model includes an input layer, a CNN layer, a GRU layer, and an output layer;

[0024] The input layer is used to input the input data into the CNN layer;

[0025] The CNN layer includes a convolutional layer and a pooling layer. The convolutional layer is used to extract features from the input data, and the pooling layer is used to perform dimensionality reduction on the extracted features;

[0026] The GRU layer is used to perform time series prediction on the features after dimensionality reduction;

[0027] The output layer is used to output the prediction result.

[0028] Optionally, the CNN layer adopts the SE attention mechanism.

[0029] Optionally, when training the icing time prediction model, the GOSO snake optimization algorithm is used for hyperparameter optimization.

[0030] Optionally, the prediction method further includes risk warning according to the contact surface temperature, the water film thickness, and the predicted icing time:

[0031] If , and , and , then the risk level is low risk;

[0032] If , and , and , then the risk level is medium risk;

[0033] If , and , and , then the risk level is high risk;

[0034] Wherein, is the contact surface temperature, the water film thickness, and the predicted icing time, are the first temperature threshold and the second temperature threshold, ; are the first thickness threshold and the second thickness threshold, ; are the first time threshold and the second time threshold, .

[0035] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0036] A detection device and a prediction method for predicting the icing time of a road surface water film provided by the present invention accurately measure the water film thickness and temperature by setting up the detection device, and combine deep learning technologies such as a convolutional neural network (CNN), a self-attention mechanism (SE), and a gated recurrent unit (GRU), as well as a snake optimization algorithm (GOSO) for global optimization, further improving the accuracy and computational efficiency of the prediction model, thereby realizing the prediction of the icing time, and enhancing the real-time performance and reliability of subsequent road safety warning and meteorological monitoring systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a schematic diagram of the principle framework of the detection device provided by an embodiment of the present invention;

[0038] Figure 2 is a schematic structural diagram of an ice detection sensor provided by an embodiment of the present invention;

[0039] Figure 3 is a schematic structural diagram of a water film thickness sensor provided by an embodiment of the present invention;

[0040] Figure 4 is a schematic structural diagram of the detection device provided by an embodiment of the present invention;

[0041] Figure 5 is a schematic flowchart of the prediction method provided by an embodiment of the present invention;

[0042] Figure 6 is a schematic structural diagram of an icing time prediction model provided by an embodiment of the present invention;

[0043] Figure 7 is a schematic diagram of the principle of the GOSO snake optimization algorithm provided by an embodiment of the present invention;

[0044] Figure 8 is a schematic structural diagram of the training of a convolutional neural network provided by an embodiment of the present invention;

[0045] Figure 9 is a schematic diagram of the training process of a convolutional neural network provided by an embodiment of the present invention;

[0046] Figure 10 is a schematic diagram of the training result of a convolutional neural network provided by an embodiment of the present invention;

[0047] Figure 11 is a total data linear prediction fitting diagram of a convolutional neural network provided by an embodiment of the present invention;

[0048] Figure 12 It is a schematic flow diagram of road icing warning provided by an embodiment of the present invention;

[0049] Figure 13 It is a schematic diagram of the complete working process provided by an embodiment of the present invention;

[0050] The markings in the figure are:

[0051] 1. Ice detection sensor; 2. Temperature sensor; 3. PPO housing; 4. Metal protective shell; 5. Water film thickness sensor; 6. Electromagnetic shielding cover; 7. Electrode post; 8. Electrode ring; 9. Low dielectric constant material; 10. Substrate; 11. Coaxial feed patch antenna. Specific embodiments

[0052] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and cannot be used to limit the protection scope of the present invention.

[0053] Embodiment 1:

[0054] As Figure 1 shown, an embodiment of the present invention provides a detection device for predicting the icing time of a water film on a road surface, including an ice detection sensor 1, a temperature sensor 2, a water film thickness sensor 5, and a microprocessor. The ice detection sensor 1 is used to detect whether the water film on the road surface is frozen. The temperature sensor 2 is used to detect the contact surface temperature between the road surface and the water film. The water film thickness sensor 5 is used to detect the thickness of the water film on the road surface; the microprocessor is used to collect the sensing data of the ice detection sensor 1, the temperature sensor 2, and the water film thickness sensor 5 and output it. If the water film is frozen, there is no need to collect the sensing data and output it anymore.

[0055] The microprocessor is a central processing unit composed of one or a few large-scale integrated circuits, such as an STM32 single-chip microcomputer. The microprocessor converts the analog signal output by the sensor into a digital signal through an analog-to-digital converter (ADC), and then performs subsequent operations and processing.

[0056] The detection device usually further includes a power management module and a host computer. The power management module is used to supply power to the ice detection sensor 1, the temperature sensor 2, the water film thickness sensor 5, and the microprocessor. The microprocessor uploads the converted digital signal to the host computer through RS485 or a network for operation and processing.

[0057] As Figure 2As shown, specifically in this embodiment, the icing detection sensor 1 adopts an annular columnar capacitor. The annular columnar capacitor includes an electrode column 7 and an electrode ring 8 coaxially sleeved outside the electrode column 7. A low dielectric constant material 9 is filled between the electrode ring 8 and the electrode column 7. The filled low dielectric constant material 9 serves as the excitation end, and the electrode ring 8 and the electrode column 7 serve as the grounding ends.

[0058] Utilizing the principle that when a dielectric is under the action of an external electric field, polarization occurs, and when a conductive medium covers the electrodes, the capacitance between the electrodes increases. The two electrode plates of the detection module and the dielectric filled therein form a parallel plate capacitor. For a capacitor with fixed plate spacing and area, its capacitance value increases with the increase of the dielectric constant of the medium. According to this principle, by using a Wheatstone bridge to measure the capacitance value, the dielectric constant of the covering can be deduced by detecting the capacitance value, thereby identifying and distinguishing the types of coverings. The initial capacitance value is:

[0059]

[0060] In the formula, represents the theoretical initial capacitance value of the annular columnar capacitor, represents the relative dielectric constant of the low dielectric constant material, represents the dielectric constant of vacuum, represents the length of the columnar capacitor, represents the outer diameter of the inner ring, represents the inner diameter of the outer ring.

[0061] The initial capacitance value of the annular columnar capacitor is proportional to the relative dielectric constant When ice or water comes into contact, the capacitance value will change. By calibrating the capacitance value of the annular columnar capacitor, water and ice can be defined. When the road surface covering is water, the sensor continuously collects data; when the road surface covering is detected as ice, the sensor stops data collection.

[0062] Specifically in this embodiment, the temperature sensor 2 adopts a PT100 platinum resistance. The measurement range of the PT100 platinum resistance is usually between -200°C and 850°C, which can cover a wide temperature range. Platinum metal has high chemical stability, and the temperature measurement of the PT100 platinum resistance has good long-term stability. The resistance of the PT100 changes approximately linearly with temperature, so it is easy to establish an accurate corresponding relationship with temperature. The contact temperature between the road surface and the water film is obtained by measuring the resistance value of the PT100 platinum resistance;

[0063]

[0064] In the formula, is the PT100 platinum resistance at temperature and the resistance value at 0 °C, is the temperature coefficient of the PT100 platinum resistance, usually 0.00385 / °C.

[0065] As Figure 3 shown, specifically in this embodiment, the water film thickness sensor 5 uses the microwave remote sensing method, and the detection antenna is its core component. Since the antenna does not need to be in direct contact with the medium for measurement, the sensor is less affected by the medium properties. Also, due to the advantages of low power consumption, good repeatability, and strong stability of this technology, the microwave sensor technology has been widely used in the fields of measuring complex relative permittivity, permeability, geometric structure parameters (thickness), etc.

[0066] Specifically set as: The water film thickness sensor 5 is wrapped in the electromagnetic shielding cover 6, including the substrate 10 and the coaxial feed patch antenna 11 attached to the substrate 10. The transmitting end of the coaxial feed patch antenna 11 emits a microwave signal, and the microwave signal returns to the receiving end of the coaxial feed patch antenna 11 after being reflected by the water film. When the microwave signal is emitted from the bottom of the liquid towards the top of the liquid, part of the microwave will form a reflected wave at the separation interface between the water film and the air. The amplitude and phase of the reflection coefficient of this reflected wave are related to the thickness, permittivity, and permeability of the dielectric layer, and the phase of the reflection coefficient changes periodically, and this period is related to the permittivity, permeability of the sensor housing and the water film, and the frequency of the microwave signal. Therefore, since the permittivity and permeability of water are basically unchanged, under the conditions of the appropriate thickness, relative permittivity, and relative permeability of the sensor housing, using a special amplitude and phase detection chip to detect the phase difference between the incident wave and the reflected wave can calculate the water film thickness. Calculating the water film thickness of the road surface :

[0067]

[0068] In the formula, is the horizontal distance from the transmitting end to the receiving end, is the microwave speed and frequency, is the phase change amount of the microwave signal from the transmitting end through the water film reflection to the receiving end.

[0069] As Figure 4 shown, specifically in this embodiment, the detection device further includes a PPO housing 3 and a metal protective shell 4. The icing detection sensor 1, the temperature sensor 2, and the water film thickness sensor 5 are arranged in the PPO housing 3, and the PPO housing 3 is embedded in the metal protective shell 4. The detection surfaces of the icing detection sensor 1 and the temperature sensor 2 are flush with the upper surfaces of the PPO housing 3 and the metal protective shell 4. Usually, the icing detection sensor 1 and the temperature sensor 2 are integrated and arranged together.

[0070] PPO is a thermoplastic engineering plastic with excellent performance as the shell material of the water film thickness sensor 5. It has good electrical insulation, heat resistance, dimensional stability and other characteristics. It has high hardness and strong rigidity, which is suitable for road detection environment. Its relative dielectric constant is 2.6~2.8. As a non-magnetic material, the magnetic permeability of PPO is close to 1, which is very suitable as the shell of the water film thickness detection and prediction sensor. The whole is encapsulated by metal, which provides good mechanical strength and stability for the sensor, and can effectively protect the sensor shell from adverse factors such as pressure, vibration, impact and chemical corrosion, thereby significantly extending its service life and operational stability. At the same time, the metal package has good sealing performance, which can effectively prevent moisture and other substances from invading the interior of the sensor, improve its waterproof ability, and improve the reliability of the device in complex environments. In addition, the metal package has a stable structure, which will be embedded in the sensor or other areas at this time, and the material is easy to maintain and replace, so as to achieve daily maintenance. The metal package also has excellent thermal performance and anti-electromagnetic interference target capability, so that the embedded road sensor can maintain stable operation in a strictly required environment and continue to provide continuous and reliable data support.

[0071] In summary, the present invention integrates and packages functional modules such as high-frequency antennas, microwave detection circuits, and temperature sensors, uses a metal casing for reinforcement and protection, and is installed in the ground through an embedded design. This design enables the sensor to have good anti-interference performance, long-term durability, and comprehensive environmental resistance, and is suitable for complex meteorological conditions and long-term monitoring needs. The present invention can not only achieve icing prediction, but can also be expanded to monitor key road parameters such as water film thickness and humidity. Its modified prediction system can perform long-term trend analysis on the collected data, provide data support and decision-making basis for meteorology, traffic planning, and road operations, and has broad practical value and promotion significance.

[0072] When the detection device provided by the embodiment of the present invention is actually applied, multiple detection devices can be evenly distributed in space to provide higher spatial resolution. Compared with the traditional single sensor, this configuration can more accurately measure the local changes in water film thickness and temperature, capture the possible unevenness and slight differences on the road surface, and help to make more accurate predictions about the icing process. Through integrated data processing technology, each sensor unit will fuse the collected data to effectively reduce the impact of external noise and interference, thereby improving the stability and accuracy of the measurement. Even in complex external environments, the sensor can still provide reliable results to ensure the accuracy of the prediction of the icing time.

[0073] Embodiment 2:

[0074] like Figure 5As shown in the figure, an embodiment of the present invention provides a method for predicting the icing time of a road surface water film. Based on the detection device provided in Embodiment 1, the prediction method includes the following steps:

[0075] Step S1: Obtain the contact surface temperature and the water film thickness, and generate sample data by combining the voltage analog quantities output by the temperature sensor and the water film thickness sensor.

[0076] The voltage analog quantities output by the water film thickness sensor include: the amplitude voltage before phase shift, the amplitude voltage after phase shift, the phase voltage before phase shift, and the phase voltage after phase shift. Therefore, in this embodiment, the sample data contains a total of 7 characteristic values:

[0077] The contact surface temperature obtained by the temperature sensor and the voltage analog quantity output;

[0078] The water film thickness obtained by the water film thickness sensor and the amplitude voltage before phase shift, the amplitude voltage after phase shift, the phase voltage before phase shift, and the phase voltage after phase shift output.

[0079] Step S2: Construct an icing time prediction model based on the CNN-GRU network, as Figure 6 shown, specifically including an input layer, a CNN layer, a GRU layer, and an output layer; the input layer is used to input the input data into the CNN layer; the CNN layer includes a convolutional layer and a pooling layer, the convolutional layer is used to extract features from the input data, and the pooling layer is used to perform dimensionality reduction processing on the extracted features; the GRU layer is used to perform time series prediction on the features after dimensionality reduction processing; the output layer is used to output the prediction result.

[0080] The CNN layer is based on the CNN convolutional neural network. When processing multi-dimensional time series, this network can perform local feature extraction on the input sequence through convolutional kernels. Assume that the input multi-dimensional time series is , the convolutional layer uses multiple different convolutional kernels to extract features, and each convolutional kernel represents a feature extraction mode. Assume that one of the convolutional kernels is , ( , i ), in the case of a step size of 1 and a padding of 0, the result after performing a convolution operation on the input sequence x is expressed as , , where: u is the value after convolution of the sequence x; is the value after introducing non-linearity through the activation function of the convolutional kernel; represents the weight value in the convolutional kernel; represents the parameter in the input matrix corresponding to the convolutional kernel sliding p, q; is the activation function.

[0081] Subsequently, the pooling layer selects the feature information extracted by the convolutional layer. The main function of the pooling layer is to reduce the dimension of the feature information, retain the receptive field of the features, and reduce the number of variables in the data, thereby reducing the computational load. The pooling operation usually compresses the features by taking the maximum or average value of a certain area in the feature map. Then, the fully connected layer is used to integrate all the features and perform the final output, and its output result is , where 、 are the output results of two adjacent connected layers respectively; 、 、 are the activation function, weight matrix, and bias matrix of the i-th fully connected layer respectively.

[0082] To further improve the feature extraction ability and model performance of the CNN layer, the Squeeze-and-Excitation (SE) architecture is proposed as an innovative feature recalibration mechanism, namely the SE attention mechanism. In traditional CNNs, the convolutional layer slides the convolutional kernel on the input feature map to extract local features, and constructs deep feature representations by stacking multiple convolutional layers. However, this processing method often ignores the differences between the features of different channels, that is, the importance of the feature maps of each channel for the final task may be different. To solve this problem, the SE architecture is proposed, which explicitly models and recalibrates the channel features to improve the model's sensitivity to useful features and suppress irrelevant features. The core idea of the SE architecture is to dynamically adjust the output feature map of the convolutional layer through two steps: Squeeze and Excitation.

[0083] The GRU algorithm is used to predict the icing time. The GRU algorithm is a variant of the RNN algorithm. Different from the standard RNN, the GRU introduces two gating mechanisms: the update gate and the reset gate, to control the flow of information and solve the problem of gradient disappearance in traditional RNNs. The update gate determines how much past information to retain at the current time step, while the reset gate determines how to combine past information and the current input. These two gates control the propagation of information in the network, helping the GRU to better capture long-term dependencies. Due to their design, GRUs are usually easier to train than traditional RNNs and perform better in many sequence modeling tasks, especially in short sequences or tasks that require long-term memory.

[0084] (1) Reset gate: Used to control the influence of past information on the current moment. The output of the reset gate is between 0 and 1, indicating how much past information to retain. Its formula is as follows:

[0085]

[0086] Among them, is the weight matrix of the reset gate, is the bias term, is the sigmoid function, is the hidden state at the previous moment, is the input at the current moment.

[0087] (2) Update gate: It is used to control the trade-off between the input at the current moment and the past hidden state. The output of the update gate is between 0 and 1, indicating how much of the past hidden state to retain. Its formula is as follows:

[0088]

[0089] Among them, is the weight matrix of the update gate, is the bias term.

[0090] (3) Hidden state: The candidate hidden state refers to the state of the candidate memory cell at a certain time step. It is calculated from the input at the current time step and the hidden state at the previous time step, and is used to update the content of the memory cell. Its formula is as follows:

[0091]

[0092] The final hidden state is the hidden state at the last time step, which can be regarded as the encoding result of the network for the entire sequence information. The formula is as follows:

[0093]

[0094] Among them, is the candidate hidden state, is the final hidden state, represents element-wise multiplication, is the weight matrix, is the bias term.

[0095] Through the above mechanism, GRU can learn long-term dependencies more flexibly, while alleviating the problem of gradient vanishing, making the training more stable and effective.

[0096] Step S3: Divide the sample data into a training set and a test set, and train and test the icing time prediction model through the training set and the test set.

[0097] During the training process, the GOSO snake optimization algorithm is used for hyperparameter optimization. The present invention introduces the snake optimization algorithm (GOSO) to optimize the model hyperparameters. Derived from traditional optimization methods, GOSO can quickly search for the optimal parameter combination globally, thereby improving the efficiency and accuracy of model training and prediction. This strategy optimization not only reduces the complexity of model tuning but also enhances the adaptability to complex non-linear structural ice processes.

[0098] The inspiration for the snake optimization algorithm comes from the mating behavior of snakes. If the temperature is low and food is available, the mating behavior of snakes occurs; otherwise, snakes will only search for food or eat the existing food. Based on this, the search process of the snake optimization algorithm is divided into two stages: exploration and exploitation. Exploration describes the environmental factors, namely cold places and food. In this stage, there is no situation where snakes only search for food in their surrounding environment. Exploitation includes many transition stages to improve the search efficiency of the algorithm. If food is available but the temperature is high, snakes will only focus on eating the available food. Finally, if food is available and the area is cold, the mating process occurs. There are some situations during the mating process, namely the combat mode or the mating mode. In the combat mode, each male will fight to get the best female, and each female will strive to select the best male. In the mating mode, the occurrence of the mating behavior of each pair of snakes depends on the quantity of food availability. In the search space, if the mating behavior occurs, the female is likely to lay eggs, which hatch into new snakes.

[0099] As Figure 7 shown, the mathematical description of snake population initialization is as follows:

[0100]

[0101] In the formula: is the position of the i-th snake; r is a random number in the range of [0, 1]; and are the upper and lower bounds of the problem to be solved, respectively.

[0102] Assume that the number of males is 50% and the number of females is 50%. The population is divided into two groups: the male group and the female group. The following two formulas are used to divide the population:

[0103]

[0104]

[0105] In the formula: is the size of the snake population; is the number of males; is the number of females.

[0106] Evaluate each group and define the temperature and the amount of food. Find the best individuals in each group to obtain the best male , the best female and the location of the food . The temperature can be defined by the following formula:

[0107]

[0108] where: t is the current iteration number; T is the maximum iteration number.

[0109] The amount of food can be defined by the following formula:

[0110]

[0111] where: is a constant, taking 0.5.

[0112] If Q < Threshold (threshold Threshold = 0.25), the snakes search for food by choosing any random location and update their positions. To simulate the exploration phase, as described below:

[0113]

[0114] where: is the male position; is the position of a randomly selected male; rand is a random number in the range [0,1]; is the ability of the male to search for food, and the calculation formula is as follows:

[0115]

[0116] where: is the fitness value of the position of a randomly selected male, is the fitness value of the male position; is a constant, taking 0.05.

[0117] *

[0118] where: is the female position; is the position of a randomly selected female; rand is a random number in the range [0,1].[[]END]]

[0119] is the ability of the female to search for food, and the calculation formula is as follows:

[0120]

[0121] In the formula: is the fitness value of the position of a randomly selected male , and is the fitness value of the male position .

[0122] Under the condition of Q > Threshold, if temperature > Threshold(0.6), then the temperature is in the hot state. The snake will only search for food, and the position update formula is as follows:

[0123]

[0124] In the formula: is the position of a snake individual (male or female); is the best position of the snake individual; rand is a random number in the range of [0, 1]; is a constant, taking 2.

[0125] Under the condition of Q > Threshold, if temperature < Threshold(0.6), then the temperature is in the cold state. The snake will be in the combat mode or the mating mode.

[0126] (1) Combat mode

[0127]

[0128] In the formula: is the position of the i-th male; is the best position in the female snake group; rand is a random number in the range of [0, 1]; is the combat ability of the male.

[0129]

[0130] In the formula: is the position of the i-th female; is the best position in the male snake group; rand is a random number in the range of [0, 1]; is the combat ability of the female.

[0131] and can be calculated by the following formula:

[0132]

[0133]

[0134] In the formula: The best position in the female snake group The fitness value; The best position in the male snake group The fitness value; is The fitness value of the snake individual.

[0135] (2) Mating pattern

[0136]

[0137]

[0138] Where: Is the position of the i-th male; Is the position of the i-th female; rand is a random number in the range [0,1]; and Are divided into the mating abilities of males and females, which can be calculated by the following formula:

[0139]

[0140]

[0141] Where: Is the fitness value of the i-th male position; is The fitness value of the i-th female position. If the eggs hatch, select the worst male and female and replace them.

[0142]

[0143]

[0144] Wherein, Is the worst position in the male snake group; Is the worst position in the female snake group.

[0145] Such as Figure 8As shown, it is the training structure diagram of a convolutional neural network. Sequence is the sequence input, which consists of 7 eigenvalue of the input data. seqfold is the sequence folding layer, conv_1 is the first two-dimensional convolutional layer, relu_1 is the first activation layer, conv_2 is the second two-dimensional convolutional layer, relu_2 is the second activation layer. gapool is the two-dimensional global average pooling layer, fc_2 is the first fully connected layer, relu_3 is the third activation layer, fc_3 is the second fully connected layer, sigmoid is the activation layer, multiplication is the dot product layer, sequnfold is the sequence unfolding layer, flatten is the network flattening layer, bilstm is the bidirectional recurrent neural network layer, and reqressionoutput is the regression output layer.

[0146] As Figure 9 , Figure 10 shown, they are respectively the training process and the training result diagram of the convolutional neural network. During the experiment, the data collected by the sensor was shuffled to form a data set for input into the convolutional neural network. Among them, 70% was used as the training set and 30% was used as the test set. It was trained for 1200 rounds, with 7 iterations per round and a maximum of 8400 iterations. The initial error RMSE was greater than 0.3 and it automatically stopped when the training ended. The training error RMSE decreased to 0.005 as the number of times increased. The error of the test using the test set after training was also relatively low. The training error is not only related to the training model but also related to the data set. A large amount of data and a good model can improve the accuracy of use and reduce the error.

[0147] As Figure 11 shown, it is the linear prediction fitting diagram of the total data. This diagram shows the fitting prediction results of all samples. The horizontal axis represents the true value of the icing time and the vertical axis represents the predicted value of the icing time. It can be seen from the figure that the scatter points are roughly distributed near a diagonal line, indicating that there is a strong linear relationship between the predicted value and the true value of the model. The slope of the trend line is positive and close to 1, indicating that the deviation between the predicted value and the true value is small. However, some scatter points deviate far from the trend line, especially in the regions where the true value is relatively low and relatively high, which indicates that there may be certain errors in the prediction of the model in these regions. Generally speaking, the prediction effect of the model is good, but the prediction accuracy in the extreme value region needs to be further improved.

[0148] Step S4: Use the trained icing time prediction model to predict the icing time of the road surface water film.

[0149] Furthermore, the prediction method also includes risk warning according to the contact surface temperature, water film thickness and the predicted icing time:

[0150] If , and and then the risk level is low risk (green);

[0151] If and and then the risk level is medium risk (yellow);

[0152] If and and then the risk level is high risk (red);

[0153] Among them, is the contact surface temperature, water film thickness and predicted icing time, are the first temperature threshold and the second temperature threshold, ; are the first thickness threshold and the second thickness threshold, ; are the first time threshold and the second time threshold, .

[0154] The thresholds can be calibrated according to actual needs. As shown in Figure 12 , in the road icing warning system, is calibrated to -2°C and 0°C, is calibrated to 0.1mm and 0.5mm, is calibrated to 30min and 60min.

[0155] 1. Green (low risk): The temperature is higher than 0°C, the icing time exceeds 60 minutes or there is no risk. The road is dry or the water film is extremely thin (<0.1mm), and normal driving is possible, but weather changes need to be monitored. 2. Yellow (medium risk): The temperature is between -2°C and 0°C, and the icing time is 30 - 60 minutes. There is a small amount of water accumulation on the road (0.1mm - 0.5mm). It is recommended to limit the speed to 70% - 80%, and spread snow melting agents on bridges, slopes, etc. Drivers need to slow down, keep a safe distance, and avoid sudden braking.

[0156] 3. Red (high risk): The temperature is lower than -2°C, and the icing time is less than 30 minutes. The water film on the road is relatively thick (>0.5mm), and the risk is extremely high. It is recommended to limit the speed to less than 50%, spread high-concentration snow melting agents or anti-slip materials on key sections, and close the road if necessary. Drivers should avoid traveling. If driving is necessary, anti-slip chains need to be installed and the vehicle should be driven at a low speed.

[0157] In summary, the complete working process of the embodiment of the present invention is as shown in Figure 13As shown, first, the system is initialized, and then the sensor starts data acquisition. Next, the sensor begins to measure the road surface covering. When the detected road surface covering is water, the sensor continuously acquires data; when the detected road surface covering is ice, the sensor stops data acquisition. Then, signal conditioning and data processing are performed on the collected temperature, water film thickness, temperature voltage, amplitude voltage before phase shift, amplitude voltage after phase shift, phase voltage before phase shift, and phase voltage after phase shift. Next, the processed data is sent into a convolutional neural network, and the CNN-GRU model is used to predict the icing time of the water film. At the same time, factors such as vehicle tires and the road surface are combined to divide the warning level, thereby facilitating the transportation department to provide appropriate driving suggestions and necessary traffic control for vehicles on the highway.

[0158] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0159] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0160] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0161] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are executed on the computer or other programmable apparatus to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps for implementing the functions specified in one block or a plurality of blocks

[0162] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art of the present technology, several improvements and modifications can be made without departing from the technical principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A detection device for predicting the icing time of road surface water film, characterized in that, Including: An icing detection sensor for detecting whether the water film on the road surface freezes; A temperature sensor for detecting the contact surface temperature of the road surface and the water film; A water film thickness sensor for detecting the thickness of the water film on the road surface; A microprocessor for collecting and outputting the sensing data of the icing detection sensor, the temperature sensor, and the water film thickness sensor.

2. The detection device for predicting the icing time of road surface water film according to claim 1, characterized in that, The icing detection sensor uses a ring-shaped columnar capacitor, which includes an electrode column and an electrode ring coaxially sleeved outside the electrode column. A low-dielectric constant material is filled between the electrode ring and the electrode column, and the filled low-dielectric constant material serves as the excitation end, while the electrode ring and the electrode column serve as the grounding end.

3. The detection device for predicting the icing time of road surface water film according to claim 1, characterized in that, The temperature sensor uses a PT100 platinum resistance, and the contact surface temperature of the road surface and the water film is obtained by measuring the resistance value of the PT100 platinum resistance; Wherein, is the resistance value of the PT100 platinum resistance at temperature and 0°C, is the temperature coefficient of the PT100 platinum resistance.

4. The detection device for predicting the icing time of road surface water film according to claim 1, wherein, The water film thickness sensor is wrapped in an electromagnetic shielding cover and includes a substrate and a coaxial-fed patch antenna attached to the substrate. The transmitting end of the coaxial-fed patch antenna emits a microwave signal, and the microwave signal is reflected by the water film and then returns to the receiving end of the coaxial-fed patch antenna to calculate the water film thickness on the road surface : Wherein, is the horizontal distance from the transmitting end to the receiving end, is the microwave speed and frequency, is the phase change amount when the microwave signal starts from the transmitting end, is reflected by the water film and reaches the receiving end.

5. The detection device for predicting the icing time of road surface water film according to claim 1, wherein, It further includes a PPO housing and a metal protective housing. The icing detection sensor, the temperature sensor, and the water film thickness sensor are arranged in the PPO housing, the PPO housing is embedded in the metal protective housing, and the detection surfaces of the icing detection sensor and the temperature sensor are flush with the upper surfaces of the PPO housing and the metal protective housing.

6. A method for predicting the freezing time of a road surface water film, characterized in that, Based on the detection device according to any one of claims 1-5, the prediction method includes: Obtaining the contact surface temperature and the water film thickness, and generating sample data in combination with the voltage analog quantities output by the temperature sensor and the water film thickness sensor; Constructing an icing time prediction model based on the CNN-GRU network; Dividing the sample data into a training set and a test set, and training and testing the icing time prediction model through the training set and the test set; Using the trained icing time prediction model to predict the icing time of the road surface water film.

7. The method for predicting the icing time of the road surface water film according to claim 6, wherein, The icing time prediction model includes an input layer, a CNN layer, a GRU layer, and an output layer; The input layer is used to input the input data into the CNN layer; The CNN layer includes a convolutional layer and a pooling layer. The convolutional layer is used to extract features from the input data, and the pooling layer is used to perform dimensionality reduction processing on the extracted features; The GRU layer is used to perform time series prediction on the features after dimensionality reduction processing; The output layer is used to output the prediction result.

8. The method for predicting the icing time of a road surface water film according to claim 1, wherein The CNN layer adopts the SE attention mechanism.

9. The method for predicting the freezing time of the road surface water film according to claim 1, characterized in that When training the icing time prediction model, the GOSO snake optimization algorithm is used for hyperparameter optimization.

10. The method for predicting the freezing time of the road surface water film according to claim 1, characterized in that, The prediction method further includes performing risk warning according to the contact surface temperature, the water film thickness, and the predicted icing time: If and and then the risk level is low risk; If and and then the risk level is medium risk; If and and , the risk level is high risk; wherein, are the contact surface temperature, the water film thickness, and the predicted icing time, are the first temperature threshold and the second temperature threshold, ; are the first thickness threshold and the second thickness threshold, ; are the first time threshold and the second time threshold, .

Images