Data-mechanism fused airport runway surface water ice condition prediction method

Abstract

The invention discloses a data-mechanism fused airport runway surface water and ice condition prediction method. The method comprises the following steps: establishing a theoretical model of a runway surface temperature field and a water and ice condition; thermal parameter inversion is carried out on the theoretical model based on actually measured data, and a dynamic theoretical model adapted to an actual application scene is obtained through multi-time iteration optimization of model parameters; and utilizing the fused time sequence data and the dynamic theoretical model to predict the water and ice condition on the surface of the airport runway, and providing dynamic evolution information of the icing process. According to the method, the problems of insufficient real-time performance and accuracy of runway icing prediction in the prior art are solved, the icing time, the ice layer thickness and the ice layer range of the runway surface can be accurately predicted, real-time deicing decision support is provided, and the runway operation safety and the airport operation efficiency are effectively improved.

Description

Technical Field

[0001] The invention relates to the field of airport engineering, and in particular to a method for predicting water ice conditions on an airport runway surface based on data-mechanism fusion, which is used to improve the safety of runway operation. Background Art

[0002] The ice condition of the airport runway surface is directly related to the safety of aircraft take-off and landing, especially in cold climates or winter. The ice on the runway may cause the friction coefficient of the runway to drop significantly, thereby increasing the taxiing distance during aircraft take-off and landing, and even causing runway skidding or other serious accidents. Therefore, accurately predicting the ice condition of the runway surface is crucial to airport operation management.

[0003] In the prior art, the following two main methods are usually used to predict the water ice condition of the runway: (1) Statistical method based on experience: This method relies on historical meteorological data and statistical models to predict the ice condition of the runway surface. Although this method can provide a reference in some cases, it lacks an in-depth understanding of the physical process, especially in the face of extreme weather or complex meteorological conditions, and the prediction accuracy is often not high. In addition, statistical models are difficult to update in a timely manner, and when environmental conditions change, the prediction model cannot respond quickly. (2) Prediction method based on physical models: This type of method mainly uses heat transfer and phase change theory to construct a physical model, and predicts the water ice condition by calculating the heat exchange process between the runway surface and the surrounding environment. However, traditional physical models are usually too idealized and fail to fully consider the complexity of environmental factors (such as solar radiation, wind speed, humidity changes, etc.) and the multi-phase dynamic changes of water, ice, and salt. In addition, model parameters are often based on assumptions or experimental data, which leads to poor adaptability of the model in different application scenarios. Especially when environmental conditions change rapidly, the model is difficult to adjust in real time, and the prediction accuracy is greatly limited.

[0004] The limitations of traditional methods include: (1) Lack of real-time and adaptability: Traditional methods, whether statistical models or physical models, often lack sufficient real-time and adaptability when faced with rapid changes in meteorological conditions. In particular, in the case of snowfall, rainfall and drastic temperature changes, existing methods are difficult to accurately predict the time and range of runway icing. (2) Inaccurate physical model parameters: The parameters in the physical model are often obtained in a laboratory environment and fail to fully reflect the complex conditions of the actual airport runway. In particular, when faced with runways of different materials and different climatic conditions, the applicability and accuracy of the model are greatly reduced. (3) Insufficient integration of data and models: With the development of intelligent runway technology, more and more real-time perception data can be used for runway monitoring, but existing prediction methods are usually based on a single mode of data or model, which makes it difficult to effectively combine the advantages of the two, thus failing to give full play to the role of real-time data, resulting in inaccurate prediction results. (4) Lack of physical consistency of prediction results: Although simple data-driven models can provide certain prediction capabilities when dealing with water ice conditions on the runway surface, due to the lack of physical model constraints, the prediction results often lack physical consistency. This may lead to deviations or unreasonable prediction results in actual applications, thereby affecting the accuracy of de-icing decisions.

[0005] In summary, the existing technologies in the prediction of runway water ice conditions have many pain points, such as poor real-time performance, insufficient adaptability, inaccurate model parameters, and insufficient fusion of data and models. Therefore, it is necessary to study a method for predicting the water ice condition on the airport runway surface to significantly improve the accuracy and real-time performance of the prediction, and to provide scientific support for the de-icing management of the airport runway. Summary of the invention

[0006] The purpose of the present invention is to provide a data-mechanism fusion method for predicting water ice conditions on the surface of an airport runway. This method improves the accuracy and real-time performance of runway water ice condition prediction by combining theoretical models of pavement temperature field and water ice conditions with real-time perception data, thereby providing a scientific basis for airport de-icing decisions and ensuring the safe operation of airport runways.

[0007] The technical solution adopted to achieve the purpose of the present invention is a data-mechanism fusion method for predicting water ice conditions on the surface of an airport runway, the method comprising:

[0008] S1. Establish theoretical models of pavement temperature field and water ice conditions;

[0009] S2. Invert the thermal parameters of the theoretical model based on the measured data, and optimize the model parameters through multiple iterations to obtain a dynamic theoretical model that adapts to the actual application scenario;

[0010] S3. Utilize the fused time series data and the dynamic theoretical model to predict the water ice condition on the airport runway surface and provide dynamic evolution information of the icing process.

[0011] The method of the present invention has the following advantages:

[0012] (1) Through the deep integration of data and mechanism, the accuracy and timeliness of the prediction of water ice conditions on the runway surface are improved.

[0013] (2) Through the dynamic optimization of the model and the fusion application of real-time data, the prediction results are more in line with the actual situation and can effectively guide the airport's de-icing decisions.

[0014] (3) This method provides an intelligent solution for airport runway management, reducing the impact of runway icing on flight safety and punctuality.

[0015] The present invention solves the problem of insufficient real-time and accuracy of runway icing prediction in the prior art, and can accurately predict the icing time, ice thickness and range of the runway surface, provide real-time deicing decision support, and effectively improve the safety of runway operation and the efficiency of airport operation. Practical applications show that the present invention can significantly warn of runway icing in advance, optimize deicing operations, and reduce operating costs, and has broad application prospects and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Flowchart of the data-mechanism fusion method for predicting water ice conditions on airport runway surfaces.

[0017] Figure 2 Schematic diagram of the multi-layered system of the runway.

[0018] Figure 3 Schematic diagram of the structure of the physical information neural network used for solving the problem.

[0019] Figure 4 Flowchart of parameter inversion and model optimization method.

[0020] Figure 5 The model framework for predicting road surface water ice conditions.

[0021] Figure 6 It is a single LSTM basic unit. DETAILED DESCRIPTION

[0022] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, the method for predicting water ice conditions on the surface of an airport runway based on data-mechanism fusion of the present invention comprises the following steps:

[0024] S1. Establish a theoretical model of pavement temperature field and water ice conditions. This theoretical model is used to reflect the dynamic thermal interaction process between water-ice-salt multiphase changes and environment-pavement.

[0025] The theoretical model of the pavement temperature field and water ice condition of the present invention is established based on the basic theory of heat transfer and phase change theory, including three parts: thermal balance of the environment-runway system, heat transfer of the multi-layered structure of the runway, and phase change of the water-ice-salt multiphase system. The three parts are described in detail below.

[0026] S1.1. The environment-runway heat balance is expressed by the environment-runway heat balance equation, which is as follows:

[0027] The environment-runway heat balance equation is used to describe the heat exchange process between the runway surface and the surrounding environment, including solar radiation, air convection, road surface reflection, surface heat conduction, and the impact of aircraft wake on the runway surface temperature. Each component in the heat balance equation is expressed in the form of heat flux, with the unit of W·m -2 According to the three basic modes of heat transfer, the environment-runway heat balance equation includes radiation, convection and conduction.

[0028] S1.1.1 The radiation part of the environment-runway heat balance equation is as follows:

[0029] The radiation heat flux is mainly composed of three parts: solar shortwave radiation heat flux Q ss , atmospheric long-wave radiation heat flux Q la , Road surface reverse long-wave radiation heat flux Q lp For two types of long-wave radiation Q la and Q lp Superposition, we get the long-wave radiation heat transfer heat flux Q out , solar shortwave radiation heat flux Q ss Heat flux Q with long-wave radiation out The calculation formula is as follows:

[0030] Q ss =(1 p )q s

[0031] In the formula, q s is the solar radiation intensity, unit: W·m -2 ; α p Pavement reflectivity, unitless.

[0032]

[0033] In the formula, ε p is the emissivity of the road surface, unitless; σ is the Stefan-Boltzmann constant 5.67×10 -8 W / (m 2 ·K4 );T p is the road surface temperature, T a is the air temperature, all in Kelvin temperature, unit K.

[0034] When the runway surface is not frozen and is covered with water, the solar shortwave radiation heat flux Q ss It is divided into two parts, one part is directly absorbed by the water layer surface, and the other part passes through the water layer to reach the road surface. The solar shortwave radiation heat flux absorbed by the water layer surface is simplified to be absorbed entirely by the upper surface of the water layer, recorded as Q ss_w The short-wave radiation that penetrates the water layer and is not absorbed by the upper layer reaches the road surface. The part absorbed by the road surface is recorded as Q ss_p .

[0035] Q ss_w =(1-α w )×Q ss =(1-α w )(1-α p )q s

[0036] Q ss_p =α w ×Q ss =α w (1-α p )q s

[0037] In the formula, α w is the water layer projection rate.

[0038] S1.1.2 The specific contents of the convection part of the environment-runway heat balance equation are as follows:

[0039] Convective heat flux includes the convective heat flux Q generated by wind conv and the convective heat flux Q generated by the aircraft wake plane The two parts are calculated as follows:

[0040] Q conv =h c (T p T a )

[0041]

[0042] In the formula, h c is the convective heat transfer coefficient, in W / (m 2 ·K); related to wind speed, v w is the wind speed in m / s; α is a coefficient related to the surface condition, which takes a value of 10.0 when there is water film on the surface and a value of 5.6 when there is ice layer on the surface.

[0043]

[0044] In the formula, m f is the mass of aviation fuel consumed by the aircraft, in kg; e f is the energy density of aviation fuel, in J / kg; l is the taxiing distance of the aircraft, in m; w is the wingspan of the aircraft, in m; h is the height of the aircraft, in m; ρ a is the air density, in kg / m 3 ; C a is the specific heat capacity of air, with the unit of J / (kg·k).

[0045] S1.1.3 The specific expression of the heat conduction part of the environment-runway heat balance equation is as follows:

[0046] Q cond =mQ ss +(1-m)Q ss_w -Q out -Q conv +Q plane

[0047] Q cond is the heat flux transferred from the surface downward by the heat transfer inside the structure, and m is the freezing state variable. When the surface is frozen, m = 1, and when it is not frozen, m = 0.

[0048] S1.2. The specific contents of the heat transfer part of the runway multi-layer structure of the theoretical model of the pavement temperature field and water ice condition are as follows:

[0049] The schematic diagram of the multi-layer structure of the runway is shown in Figure 2 As shown in the figure, the runway is considered as a multi-layered body, where x is the depth coordinate and h is the depth coordinate. i 、T i , k i , α i They represent the thickness, temperature function, thermal conductivity, and thermal diffusivity of the i-th layer respectively. The heat transfer of the multi-layered structure of the runway includes three parts: heat conduction within each layer, continuous heat transfer between layers, and continuous heat transfer at the boundary.

[0050] S1.2.1. The heat conduction inside each layer of the multi-layered structure of the runway satisfies Fourier's law of heat conduction, which is expressed as follows:

[0051]

[0052] S1.2.2. The continuous heat transfer between the layers of the multi-layered structure of the runway is used as the boundary condition of the theoretical model of the pavement temperature field and water ice condition. When constructing the theoretical model, it is assumed that the contact between the various structural layers of the runway is good. The expression is as follows:

[0053]

[0054] T i (H i+1 ,t)=T i+1 (H i+1 ,t)

[0055] S1.2.3. The continuous heat transfer at the boundary of the multi-layered structure of the runway is used as the boundary condition of the theoretical model of the pavement temperature field and water ice conditions. Considering the heat exchange between the environment and the upper surface and the stable geothermal gradient, the continuous heat transfer boundary conditions include three parts: thermal balance between the upper surface and the environment, constant initial temperature, and constant temperature at infinity.

[0056] S1.2.3.1. The thermal balance between the upper surface of the runway multi-layer structure boundary and the environment. The upper surface of the runway multi-layer structure in contact with the environment includes the upper surface of the water layer and the upper surface of the runway surface. The boundary condition expressions on the two interfaces are as follows:

[0057]

[0058] In the formula, k w is the thermal conductivity of water; T w is the temperature function of the water layer; d is the thickness of the water film, in m.

[0059] S1.2.3.2 The expression for the constant initial temperature of the multi-layered structure of the runway is as follows:

[0060] T i (x,0)=T0

[0061] S1.2.3.3 The expression for the constant temperature at infinity for a multi-layered structure of a runway is as follows:

[0062] T n (∞,t)=T0

[0063] S1.3 The specific contents of the phase transition of the water-ice-salt multiphase system of the theoretical model of the pavement temperature field and water ice conditions are as follows:

[0064] During the ice formation process on the runway surface, salt ions migrate into the unfrozen water, causing the salt concentration to increase, which in turn causes the freezing point to change and the latent heat of solidification to change. The calculation formula is as follows:

[0065] T f =237.15-ΔT f =i·K f ·c

[0066]

[0067] L' f =L f-C p ΔT′ f

[0068] Where, T f is the current freezing point, in K; ΔT f is the initial freezing point depression; i is the Van't Hoff factor, which indicates the effective number of particles produced by each solute molecule or ion in the solution; K f is the molar freezing point depression constant of the solvent, in K kg / mol, which depends on the physical properties of the solvent; c is the molar concentration of the solution, in mol / kg; ΔT f ' is the freezing point depression at the current salt concentration; Z ice is the position of the interface between the ice layer and the water layer, which is a function that changes with time; d is the thickness of the water film, in meters; L f is the latent heat of solidification of pure water, 334KJ·kg -1 , L f ' is the latent heat of solidification at the current salt concentration; C p is the specific heat capacity of water, which is 4.18KJ / (kg·K).

[0069] At the ice-water interface, the latent heat flux Q of the phase change during the ice freezing process wi It can be calculated by the rate of change of ice thickness over time:

[0070]

[0071] In the formula, ρ w is the density of water, which is 1×10 3 kg·m -3 .

[0072] The above calculation of the process of increasing salt concentration, decreasing freezing point and reducing latent heat of phase change in the water-ice-salt multiphase system during the freezing process enables a more refined expression and calculation of the heat of phase change in the water-ice-salt system.

[0073] The interface between the ice layer and the water layer still needs to meet the interlayer continuous heat transfer boundary condition, and the continuity of heat transfer and temperature continuity must be ensured. The boundary condition expression at the junction is as follows:

[0074]

[0075] According to the above Q wi The two calculation formulas are:

[0076]

[0077] Get the following formula to calculate the current freezing point T f :

[0078] T w(Z ice ,t)=T i (Z ice ,t)=T f

[0079] In the formula, k i With k w is the thermal conductivity of ice and water, T i With T w is a function of the temperature of the ice layer and the water layer.

[0080] S2: Use measured data to invert and optimize the thermal parameters of the theoretical model of pavement temperature field and water ice conditions established in S1, iteratively optimize the model parameters, and obtain a dynamic theoretical model that adapts to actual application scenarios.

[0081] In this embodiment, a physical information neural network (PINN) is used to simultaneously solve the forward and inverse problems to iteratively optimize the model parameters. The specific contents of the physical information neural network used for thermal parameter inversion and optimization are as follows:

[0082] The network structure established is as follows Figure 3 As shown in the figure, the network takes time (t) and space (x) coordinates as input variables, and temperature (T) and ice thickness (Z) as input variables. ice ) as output. A separate sub-network is constructed for each layer of the runway structure. Each sub-network contains an input layer, seven hidden layers, and an output layer. Each hidden layer consists of 64 neurons. Linear transformation calculations are applied between hidden layers. Each hidden layer receives the output of 64 neurons in the previous layer, and then passes through the activation function ReLU to output the same number of neurons. The output layer is a linear layer that receives the output of 64 neurons from the last hidden layer.

[0083] The loss function of the designed PINN network consists of two parts, the data-driven error term (data loss) and the physical information residual term (physical loss): the data-driven error term, that is, the mean square error (T data -T) 2 , through the independently designed runway surface condition model icing chamber, the temperature field and water ice condition data of the icing process under different environmental conditions are obtained as actual observation data; the physical information residual term, that is, the output variable must satisfy the physical model constructed in part S1. The PINN network can realize automatic differentiation of the output variable, calculate the output variable temperature (T) and ice thickness (Z ice ) with respect to the time (t) and space (x) derivatives of the input variables, including the first-order derivative of temperature with respect to time The first spatial derivative of temperature and the second derivative and the first derivative of ice thickness with respect to time According to the physical model constructed in part S1, equations including these differentials are established, which are the physical information residuals.

[0084] The specific operations of thermal parameter inversion and theoretical model optimization of the theoretical model through the physical information neural network established above are as follows:

[0085] The flow chart of parameter inversion and model optimization method is as follows: Figure 4 As shown. Based on the physical information neural network, the forward and inverse problems are solved simultaneously. The thermal parameters of each layer in the multi-layer structure of the runway in the model, including thermal conductivity (k), specific heat capacity (c) and density (ρ), are set as variable parameters of the network. The empirical value range of each parameter is given, and together with the parameters of the neural network itself (weights and biases), it is used as network parameters. The Adam optimization algorithm is used for iterative optimization. When the total loss (data loss + physical loss) is minimized, the material thermal parameters in the network are the parameters that best meet the actual application scenario. At this time, the theoretical model of the pavement temperature field and water ice conditions is optimal. The PINN network used for solving is defined as f(t,x; θ, k, α), θ is the neural network's own parameter, k and α are the material thermal parameters, and are used as variables to be optimized. The loss function expression is as follows:

[0086]

[0087] Where N b 、N f N0 and N1 represent the residual training set size of boundary points, equation configuration points, and initial points respectively; c w , c1 represents the thermal parameter k of the material w , c1 is the empirical value to determine the constant value; N t is the point set of the test point locations in the experimental cabin; T represents the temperature of the corresponding test point location of the actual experimental data.

[0088] S3: Combined with the real-time monitoring data of the smart runway, the time series data and the optimized thermal model are used for fusion prediction, and a runway surface water ice condition prediction model based on the physics-guided long short-term memory network (PI-LSTM) is established. The historical monitoring data is input to obtain the dynamic evolution of the runway surface water ice condition in the future.

[0089] The real-time monitoring data of the smart runway includes two types of perception data: meteorological environment and water ice conditions. The meteorological environment perception data includes air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed, and rainfall intensity; the water ice condition perception data includes water film thickness, ice thickness, pavement temperature, and freezing point temperature. The data collection time interval is every 30 minutes, and the various types of perception data are time-matched to form three sets of time series data sets for meteorological environment, pavement temperature field, and surface water ice conditions.

[0090] The road surface water ice condition prediction model architecture based on the physics-guided long short-term memory network (PI-LSTM) is as follows: Figure 5 As shown in the figure, the time series part is three sets of time series data sets formed by the real-time monitoring data of the intelligent runway, and the specific content of the long short-term memory network structure is as follows:

[0091] The time series size of the network input is 12, and the output time series size is 6, that is, the input of the model is the air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed, water film thickness, and ice thickness 6 hours before the current moment, and the air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed 3 hours after the current moment, and the model output is the water film thickness and ice thickness 3 hours after the current moment.

[0092] The network consists of four LSTM layers, two Dropout layers, and five fully connected layers. Each of the four LSTM layers contains 64 LSTM units. A single standard LSTM unit is Figure 6 As shown in the figure, the output of each layer is used as the input of the next layer; two Dropout layers are used to reduce overfitting, with a ratio of 0.3, and are located between the second and third LSTM layers and between the third and fourth fully connected layers respectively; five fully connected layers are located at the end, and the first three fully connected layers contain 512, 128 and 64 neurons respectively, using the ReLU activation function, and the last two fully connected layers do not set the activation function. The fully connected layer is used to integrate the features learned from the LSTM layer and connect the output layer. The output layer has 12 neurons, representing the water film thickness and ice thickness values ​​with a time series size of 6. Adam was selected as the optimizer, and the learning rate was set to 0.001.

[0093] The total loss includes data loss and physical information loss. The mean square error (MSE) is used as the data loss function, while the physical information loss consists of two parts: one is a hard constraint, which is directly derived from the physical model equation to ensure that the output of LSTM strictly complies with the physical relationship between environment, temperature and water ice conditions. The model established by S1 and optimized by S2 reflects the meteorological environment sequence within a certain period of time [E t+1 ,…,E t+N ] and the road surface temperature series [T t+1 ,…,T t+N ] and water ice state series [S t+1 ,…,S t+N ], the meteorological environment includes air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed, and rainfall intensity, and the water ice condition includes water film thickness and ice thickness. The road surface temperature and water ice condition output results obtained by the prediction model must also satisfy the mapping relationship with the meteorological environment, thereby obtaining the mapping loss LOSS f The expression is as follows:

[0094]

[0095] In the formula, G represents the model established by S1 and S2; E i Represents the meteorological environment value at time i, which is a vector containing six elements, namely air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed, and rainfall intensity at time i; Represents the model calculated value of the pavement temperature field at time i; The model calculated value representing the water ice condition at time i is a vector containing two elements, namely the water film thickness and ice thickness at time i.

[0096] The other part is soft constraints, including temperature continuity and freezing point and other time series characteristics. These characteristics guide the prediction of the model based on physical laws to ensure the consistency of the output results in physical phenomena. temp Expression and freezing point freezing loss LOSS freeze The expression is as follows:

[0097]

[0098]

[0099] Total loss LOSS total The expression is as follows:

[0100]

Claims

1. A data-mechanism fusion method for predicting water ice conditions on airport runway surfaces, characterized in that: include: S1. Establish theoretical models of pavement temperature field and water ice conditions; S2. Invert the thermal parameters of the theoretical model based on the measured data, and optimize the model parameters through multiple iterations to obtain a dynamic theoretical model that adapts to actual application scenarios; S3. Utilize the fused time series data and the dynamic theoretical model to predict the water ice condition on the airport runway surface and provide dynamic evolution information of the icing process.

2. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 1 is characterized in that The theoretical model for establishing the pavement temperature field and water ice conditions includes three parts: thermal balance of the environment-runway system, heat transfer of the multi-layered structure of the runway, and phase change of the water-ice-salt multiphase system.

3. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 2 is characterized by: The environment-runway thermal balance includes radiation, convection, and heat conduction. The radiation part includes the solar shortwave radiation heat flux Q ss and long-wave radiation heat transfer heat flux Q out ,in, Q ss =(1-α p )q s In the formula, q s is the solar radiation intensity, α p pavement reflectivity; In the formula, Q la is the atmospheric long-wave radiation heat flux, Q lp is the reverse long-wave radiation heat flux of the road surface, ε p is the emissivity of the road surface, σ is the Stefan-Boltzmann constant, T p is the road surface temperature, T a is the air temperature; The convection part includes: the convection heat flux includes the convection heat flux Q generated by the wind conv and the convective heat flux Q generated by the aircraft wake plane The two parts are calculated as follows: Q conv =h c (T p -T a ) In the formula, h c is the convective heat transfer coefficient, v w is the wind speed, α is a coefficient related to the surface conditions, In the formula, m f is the mass of aviation fuel consumed by the aircraft, e f is the energy density of aviation fuel, l is the taxiing distance of the aircraft, w is the wingspan of the aircraft, h is the height of the aircraft, ρ a is the air density, C a is the specific heat capacity of air; The heat conducting part comprises: Q cond =mQ ss +(1-m)Q ss_w -Q out -Q conv In the formula, Q cond is the heat flux transferred downward from the surface, and m is the freezing state variable.

4. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 2 is characterized by: The heat transfer of the multi-layered structure of the runway includes heat conduction within each layer, continuous heat transfer between layers, and continuous heat transfer at the boundary, wherein: The heat conduction inside each layer is calculated by the following formula: The interlayer continuous heat transfer is calculated by the following formula: T i (H i+1 ,t)=T i+1 (H i+1 ,t) The boundary continuous heat transfer includes three parts: thermal balance between the upper surface and the environment, constant initial temperature, and constant temperature at infinity. The upper surface includes the upper surface of the water layer and the upper surface of the road surface. The boundary condition expressions on the two interfaces are as follows: In the formula, k w is the thermal conductivity of water; T w is the temperature function of the water layer; d is the thickness of the water film, in m; The initial temperature constant is calculated by the following formula: T i (x,0)=T0 The temperature at infinity is constant by the following formula: T n (∞,t)=T0.

5. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 2 is characterized in that The phase transition of the water-ice-salt multiphase system is represented by the following formula: T f =237.15-ΔT f =i·K f ·m L' f =L f -C p ·ΔT′ f In the formula, ΔT f is the initial freezing point depression; i is the Van't Hoff factor, K f is the molar freezing point depression constant of the solvent, m is the molar concentration of the solution, ΔT' f is the freezing point depression at the current salt concentration, Z ice is the function of the position of the interface between the ice layer and the water layer changing with time, d is the thickness of the water film, L f is the latent heat of solidification of pure water, L' f is the latent heat of solidification at the current salt concentration; C p is the specific heat capacity of water; At the ice-water interface, the latent heat flux Q of the phase change during the ice freezing process wi The change rate of ice thickness over time is calculated as follows: In the formula, ρ w is the density of water; The interface between the ice layer and the water layer still needs to meet the interlayer continuous heat transfer boundary condition, and the continuity of heat transfer and temperature continuity must be ensured. The boundary condition expression at the junction is as follows: T w (Z ice ,t)=T i (Z ice ,t)=T f In the formula, k i With k w is the thermal conductivity of ice and water, T i With T w is a function of the temperature of the ice layer and the water layer.

6. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 1 is characterized by: Step S2 uses a physical information neural network to solve the forward and reverse problems synchronously to iteratively optimize the model parameters. The physical information neural network established uses time and space coordinates as input variables and temperature and ice thickness as output. A separate subnetwork is constructed for each layer of the runway structure. Each subnetwork includes an input layer, seven hidden layers, and an output layer. Each hidden layer consists of 64 neurons. Linear transformation calculations are applied between hidden layers. Each hidden layer receives the output of 64 neurons in the previous layer, and then passes through the activation function ReLU to output the same number of neurons. The output layer is a linear layer that receives the output of 64 neurons from the last hidden layer. The loss function of the physical information neural network includes two parts, a data-driven error term and a physical information residual term: the data-driven error term, that is, the mean square error between the predicted value and the actual observed data, is obtained through the independently designed runway surface condition model icing chamber, and the temperature field and water ice condition data of the icing process under different environmental conditions are used as the actual observed data; the physical information residual term, that is, the output variable must satisfy the theoretical model of the pavement temperature field and water ice condition constructed by S1.

7. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 6 is characterized by: Based on the physical information neural network, the forward and reverse problems are solved simultaneously. The thermal parameters of each layer in the multi-layer structure of the runway in the model, including thermal conductivity, specific heat capacity and density, are set as variable parameters of the network. The empirical value range of each parameter is given, and it is used together with the parameters of the neural network itself as network parameters. The Adam optimization algorithm is used for iterative optimization. When the total loss is minimized, the material thermal parameters in the network are the parameters that best meet the actual application scenario. At this time, the theoretical model of the pavement temperature field and water ice conditions is optimal. The PINN network used for solving is defined as f(t,x; θ,k,α), θ is the neural network parameter itself, k and α are material thermal parameters, as variables to be optimized, and the loss function expression is as follows: Where N b 、N f N0 and N1 represent the residual training set size of boundary points, equation configuration points, and initial points respectively; c w , c1 represents the thermal parameter k of the material w , c1 is the empirical value to determine the constant value; N t is the point set of the test point locations in the experimental cabin; T represents the temperature of the corresponding test point location of the actual experimental data.

8. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 1 is characterized by: Combined with the real-time monitoring data of the smart runway, the time series data and the optimized thermal model are used for fusion prediction, and a runway surface water ice condition prediction model based on a physics-guided long short-term memory network is established. The historical monitoring data is input to obtain the dynamic evolution of the runway surface water ice condition in the future.

9. The data-mechanism fusion airport runway surface water ice condition prediction method according to claim 8 is characterized by: The real-time monitoring data of the smart runway includes two types of perception data: meteorological environment and water ice conditions. The meteorological environment perception data include air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed, and rainfall intensity. The water ice condition perception data include water film thickness, ice thickness, pavement temperature, and freezing point temperature. The data collection time interval is every 30 minutes, and time matching is performed on various types of perception data to form three sets of time series data sets for meteorological environment, pavement temperature field, and surface water ice conditions.

10. The data-mechanism fusion method for predicting water ice conditions on airport runway surfaces according to claim 8, characterized in that: The time series part of the physical-guided long short-term memory network is three groups of time series data sets formed by the real-time monitoring data of the intelligent runway, wherein the long short-term memory network structure includes a network input time series size of 12 and an output time series size of 6, that is, the input of the model is the air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and wind speed, water film thickness, ice thickness 6 hours before the current moment and the air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and wind speed 3 hours after the current moment, and the model output is the water film thickness and ice thickness 3 hours after the current moment; The total loss of the loss function of the LSTM network based on physical guidance includes data loss and physical information loss. The physical information loss includes two parts: hard constraint loss and soft constraint loss. The hard constraint loss is directly derived from the physical model equation, so that the network output strictly abides by the physical relationship between the environment, temperature and water ice conditions. The soft constraint loss includes time series characteristics such as temperature continuity and freezing point freezing. The prediction of the model is guided based on physical laws to ensure the consistency of the output results in physical phenomena. The total loss LOSS total The expression is as follows: Where G represents the theoretical model of pavement temperature field and water ice condition; E i Represents the meteorological environment value at time i, which is a vector containing six elements, namely air temperature, air humidity, solar radiation, atmospheric pressure, wind direction and speed, and rainfall intensity at time i; represents the model calculated value of the pavement temperature field at time i; The model-calculated value representing the water ice condition at time i is a vector containing two elements, namely the water film thickness and ice thickness at time i; λ is the equivalent total thermal conductivity.

Images