Road surface condition estimation device and road surface condition estimation method
A spatiotemporal statistical model improves road surface condition estimation by using solar radiation and other variables, addressing the inaccuracy of conventional methods and enabling precise predictions for urban areas.
Patent Information
- Application Number
- JP2023071820
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-25
- Publication Date
- 2024-11-07
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Conventional methods for estimating road surface conditions, such as temperature and presence of moisture, are inaccurate due to the complexity of physical models requiring numerous parameters and coarse meteorological data, especially in urban areas with obstructed solar radiation, leading to poor prediction accuracy.
A spatiotemporal statistical model is used to estimate road surface conditions by considering the spatiotemporal structure of road surface temperature, incorporating solar radiation and other explanatory variables, with a configuration that includes linear regression, error terms, and Gaussian Predictive Process to reduce calculation costs and improve accuracy.
The method enables precise estimation of road surface conditions by accounting for non-uniformity and temporal changes, enhancing prediction accuracy and enabling targeted road maintenance and driver warnings.
Smart Images

Figure 2024157446000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to a road surface condition estimating device and a road surface condition estimating method. [Background technology]
[0002] Conventionally, technologies for estimating road surface temperature and road surface condition have been proposed to improve the driving comfort of automobiles and optimize driving in automatic driving. In general, road surface temperature and road surface condition are often estimated mainly by solving one-dimensional analytical models. For example, road surface temperature is often calculated based on the balance between the input and output of heat flux to the road surface, taking into account meteorological variables such as air temperature, physical properties such as the thermal conductivity of the road surface, and traffic volume. In addition, road surface conditions such as wetness or freezing are often predicted by calculating the presence of moisture on the road surface based on the balance of rainfall, snowfall, and the drainage capacity of the road, and combining it with road surface temperature information. For example, the system described in Patent Document 1 estimates the amount of water, snow, and ice stored at a fixed observation point, and uses a heat balance model equation that calculates the maximum amount of heat of freezing and melting, and water, snow, and ice balance equations that respectively indicate the rate of change of the amount of water, snow, and ice stored. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] JP 2009-243884 A Summary of the Invention [Problem to be solved by the invention]
[0004] The physical models used in the conventional technology are generally complex and require a large number of parameters. In addition, they may include parameters that are difficult to obtain, such as the thermal conductivity distribution and reflectance distribution of asphalt. In addition, although meteorological variables are important parameters, the size of the mesh is, for example, 5 km or more on a side, and the data is often rough compared to the road surface temperature and road surface condition to be predicted. Furthermore, meteorological variables are generally measured by meteorological observation equipment installed in open areas, and do not reflect the actual condition of the road surface, where solar radiation may be blocked by buildings or trees. Therefore, the road surface temperature and road surface condition predicted by the conventional technology are premised on an area with good visibility and few buildings, and it is difficult to predict road surfaces, such as urban areas with many buildings and many shaded areas. Therefore, there is room for improvement in the prediction accuracy.
[0005] An object of the present disclosure, made in consideration of the above circumstances, is to provide a road surface condition estimation device and a road surface condition estimation method that are capable of estimating road surface conditions with high accuracy by taking into account the spatiotemporal structure of road surface temperature. [Means for solving the problem]
[0006] (1) A road surface condition estimation device according to an embodiment of the present disclosure includes: an acquisition unit that acquires a spatio-temporal statistical model having a road surface temperature as a response variable and a parameter including at least the amount of solar radiation as an explanatory variable; an estimation unit that calculates the road surface temperature used to estimate a road surface condition by using the spatio-temporal statistical model; The spatio-temporal statistical model is a statistical model that takes into account the spatio-temporal structure of the road surface temperature. This configuration enables the road surface condition to be estimated with high accuracy.
[0007] (2) As an embodiment of the present disclosure, in (1), The spatio-temporal statistical model is constructed with a term that indicates a mean structure using linear regression and an error term. This construction allows modeling using linear regression where the time and space dependent regression coefficients are relatively simple to determine.
[0008] (3) As an embodiment of the present disclosure, in (2), The spatio-temporal statistical model uses separate covariance functions for the error terms in space and time. This configuration can reduce the calculation cost.
[0009] (4) As an embodiment of the present disclosure, in (2), The spatio-temporal statistical model is constructed such that the error terms include random effects. This configuration allows efficient estimation of parameters of the spatio-temporal statistical model.
[0010] (5) As an embodiment of the present disclosure, in (4), The spatio-temporal statistical model performs parameter centering. This configuration makes it possible to make MCMC (Markov Chain Monte Carlo) calculations more efficient.
[0011] (6) As an embodiment of the present disclosure, in (5), The spatio-temporal statistical model is configured to explicitly account for first order time dependence in the spatio-temporal data. This configuration makes it possible to make MCMC calculations more efficient while taking into account time dependence.
[0012] (7) As an embodiment of the present disclosure, in (2), The spatio-temporal statistical model uses GPP (Gaussian Predictive Process). This configuration makes it possible to suppress an increase in calculation costs.
[0013] (8) As an embodiment of the present disclosure, in (2), The spatio-temporal statistical model makes the regression coefficients spatially dependent. This configuration can reduce the effect of the error terms.
[0014] (9) As an embodiment of the present disclosure, in (2), The spatio-temporal statistical model makes the regression coefficients time dependent. This configuration can reduce the effect of the error terms. (10) As an embodiment of the present disclosure, in (2), The spatio-temporal statistical model makes the regression coefficients dependent on space and time. This configuration can reduce the effect of the error terms.
[0015] (11) A road surface condition estimation method according to an embodiment of the present disclosure, Obtaining a spatio-temporal statistical model with road surface temperature as a response variable and parameters including at least solar radiation as explanatory variables; Calculating the road surface temperature used to estimate a road surface condition using the spatio-temporal statistical model; The spatio-temporal statistical model is a statistical model that takes into account the spatio-temporal structure of the road surface temperature. This configuration enables the road surface condition to be estimated with high accuracy. Effect of the Invention
[0016] According to the present disclosure, it is possible to provide a road surface condition estimation device and a road surface condition estimation method that are capable of estimating road surface conditions with high accuracy by taking into account the spatiotemporal structure of road surface temperature. [Brief description of the drawings]
[0017] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a road surface condition estimating device according to an embodiment of the present disclosure. [Diagram 2] FIG. 2 is a diagram showing an example of the configuration of a road surface condition estimating system including the road surface condition estimating device of FIG. [Diagram 3] FIG. 3 is a diagram for explaining a model used in estimating road surface conditions. [Figure 4]FIG. 4 is a flowchart illustrating a process for estimating a road surface condition. [Diagram 5] FIG. 5 is a flowchart showing the details of the model generation process of FIG. [Figure 6] FIG. 6 is a flowchart showing the details of the road surface condition estimating process of FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0018] Hereinafter, a road surface condition estimation device and a road surface condition estimation method according to an embodiment of the present disclosure will be described with reference to the drawings. In each drawing, the same or corresponding parts are denoted by the same reference numerals. In the description of this embodiment, the description of the same or corresponding parts will be omitted or simplified as appropriate.
[0019] FIG. 1 shows a configuration example of a road surface condition estimation device 10 according to this embodiment. FIG. 2 shows a configuration example of a road surface condition estimation system including the road surface condition estimation device 10 of FIG. 1. The road surface condition estimation device 10 is a device that estimates (determines) the road surface condition of a road on which a vehicle 20 travels. The road surface condition is the state of the surface of the road and may include categories such as dry, wet, ice, and snow. The road surface condition is not limited to these four categories, and may, for example, omit some categories or may further include flooding. Furthermore, the vehicle 20 may be, for example, a passenger car, a truck, a bus, or a construction vehicle, and is not limited to a specific type of moving body.
[0020] In this embodiment, the road surface condition estimation device 10 generates a model and estimates the road surface condition of the road on which the vehicle 20 travels using the generated model. In other words, the road surface condition estimation device 10 according to this embodiment is described as a device having both the functions of a device that generates a model (model generation device) and a device that estimates road surface conditions. As another example, the road surface condition estimation device 10 may be a device that only has the function of estimating road surface conditions using a model generated in advance.
[0021] 1, the road surface condition estimation device 10 includes a communication unit 11, a storage unit 12, and a control unit 13. The control unit 13 includes an acquisition unit 131, a model generation unit 132, an estimation unit 133, and an output unit 134. The road surface condition estimation device 10 may be, for example, a computer as a hardware configuration. The components of the road surface condition estimation device 10 will be described in detail later.
[0022] 2, the road surface condition estimation device 10 may configure a road surface condition estimation system together with devices other than the road surface condition estimation device 10 (hereinafter, "external devices") connected via a network 40. The network 40 is, for example, the Internet. Furthermore, the network 40 may be configured to include, for example, a local area network (LAN) in part. The external devices may include, for example, at least one of a terminal device 50 and a road management device 60.
[0023] The road management device 60 is a device for managing roads, and acquires data on road surface conditions estimated by the road surface condition estimation device 10. The road management device 60 is, for example, composed of a computer separate from the road surface condition estimation device 10. The road administrator may plan, for example, snow removal or snow melting work based on a map showing the road surface condition of the road shown by the road management device 60. Here, even in the same area, the amount of snow accumulation may differ between a road with a lot of solar radiation and a road in the shadow of a building. In addition, for example, there is an upper limit to the number of snow removal vehicles in operation, and the spraying of an excessive amount of snow-melting agent may have adverse effects such as rust on the traveling vehicles. By estimating the road surface condition with high accuracy by the road surface condition estimation device 10, the road administrator can create an operation plan to concentrate snow removal vehicles on roads estimated to have snow accumulation and spray an appropriate amount of snow-melting agent according to the estimated amount of snow accumulation. In addition, the road administrator may warn the driver of the vehicle 20 traveling on the frozen or snow-covered road, for example, by radio.
[0024] Furthermore, when the road surface temperature or road surface condition is measured, the road management device 60 may associate these measured values with at least one of the meteorological variables, map information, and traffic information at the time of measurement, accumulate them as a database, and store them in an accessible storage device. The meteorological variables are variables about the weather that change depending on the location and time, and include, for example, temperature, humidity, solar radiation hours (sunshine hours), precipitation, etc. The map information includes, for example, road coordinates, topography, building information, etc. The topography includes, for example, altitude, etc. The traffic information includes, for example, traffic volume, etc. The meteorological variables, map information, and traffic information can be acquired via the network 40. When the road surface condition estimation device 10 generates a model, the road management device 60 may provide the road surface condition estimation device 10 with performance data (data from the past rather than the present) including the measured values accumulated in the database. In this embodiment, the road surface condition estimation device 10 uses the provided performance data as learning data for machine learning.
[0025] The terminal device 50 is a device that presents information to the driver or passengers of the vehicle 20, and acquires data on road surface conditions estimated by the road surface condition estimation device 10. The terminal device 50 may be an in-vehicle device mounted on the vehicle 20, such as a car navigation system, or may be a portable device, such as a smartphone or tablet. The terminal device 50 may have, for example, a car navigation function, and may warn the driver of the vehicle 20 to be careful when the vehicle 20 enters a road that is frozen or covered with snow.
[0026] FIG. 3 is a diagram for explaining a model used in the road surface condition estimation method executed by the road surface condition estimation device 10. In this embodiment, a machine learning model and a spatio-temporal statistical model are used. The machine learning model is generated using learning data in which the road surface condition is a response variable and parameters including at least the road surface condition non-uniformity and road surface temperature of the road are explanatory variables. The non-uniformity of the road surface condition means that the road surface condition differs depending on the location, and the road surface condition differs for each road even if the roads are located in a particularly close location. Non-uniformity can also be called heterogeneity. The spatio-temporal statistical model is a model in which the road surface temperature is a response variable and parameters including at least the amount of solar radiation are explanatory variables. Although a model corresponding to the spatio-temporal statistical model can be generated by machine learning, a statistical model is adopted in this embodiment. The spatio-temporal statistical model may be a model including, for example, a linear regression technique. As shown in FIG. 3, in this embodiment, the road surface temperature input to the machine learning model is the road surface temperature calculated using the spatio-temporal statistical model. In this specification, when the machine learning model and the spatio-temporal statistical model are not particularly distinguished from each other, they may be collectively referred to as a model.
[0027] Here, in the conventional technology, the road surface temperature and road surface condition are often estimated mainly by solving a one-dimensional analytical model. The physical models used in the conventional technology are generally complex and require a large number of parameters. In addition, they may include parameters that are difficult to obtain, such as the thermal conductivity distribution and reflectance distribution of asphalt. In addition, although meteorological variables are also important parameters in such physical models, the size of the mesh is, for example, 5 km or more on a side, and the data is often rough compared to the road surface temperature and road surface condition to be predicted. Furthermore, meteorological variables are generally measured by a meteorological observation device installed in an open area, and do not reflect the actual condition of the road surface, where solar radiation may be blocked by buildings or trees. Therefore, the road surface temperature and road surface condition predicted by the conventional technology are assumed to be in an area with good visibility and few buildings, and there is room for improvement in the prediction accuracy.
[0028] The inventors of the present invention have made extensive investigations and found that, as described above, it is possible to estimate the road surface condition with high accuracy by calculating the road surface temperature, which strongly influences the road surface condition, using a spatio-temporal statistical model, inputting the calculated road surface temperature, and estimating it using a machine learning model. The road surface condition estimation device 10 according to the present embodiment calculates the road surface temperature using a spatio-temporal statistical model. Therefore, in estimating the road surface condition, it is possible to use a road surface temperature close to the actual distribution that takes into account the amount of solar radiation, rather than a road surface temperature based only on uniform meteorological variables (e.g., air temperature and humidity) in a coarse mesh of 5 km or more. In other words, it is possible to estimate the road surface condition with high accuracy based on a detailed road surface temperature that takes into account the amount of solar radiation, rather than a uniform road surface temperature over a wide area.
[0029] Furthermore, in this embodiment, the machine learning model uses parameters including at least the non-uniformity of the road surface condition and the road surface temperature as explanatory variables. Therefore, for example, when the road surface condition is estimated for a certain location, it is not assumed that the road surface condition is uniformly the same in an area within a predetermined distance from the location, but the range of the same road surface condition can be determined with high accuracy by considering the non-uniformity of the road surface condition. As an example, when a main road and a community road are close to each other, if these roads are not connected and there is no traffic of vehicles 20, there may be a state in which the community road is snowed over and the main road with a lot of traffic is dry. In the same example, if these roads are connected and many vehicles 20 are traveling, the road surface conditions of both the community road and the main road may be dry. By having the machine learning model consider the connection state between such roads (hereinafter, "road network") as the non-uniformity of the road surface condition, it is possible to estimate the road surface condition with high accuracy compared to the conventional technology that estimates the same road condition uniformly within a predetermined distance range.
[0030] As shown in FIG. 3, the machine learning model may further include at least one of the amount of solar radiation, topography, water-poor areas, traffic volume, and weather variables as parameters serving as explanatory variables. By using these pieces of information in addition to the road network, a machine learning model that takes into account the non-uniformity of the road surface condition in more detail can be generated, and the road surface condition can be estimated with higher accuracy using the machine learning model. The amount of solar radiation may be calculated based on the distribution of solar radiation on the road and the duration of solar radiation. The distribution of solar radiation may be calculated based on the topography, building information, and the like contained in map information that can be acquired via the network 40, for example. For example, even in a close range on the same road, the amount of snowfall may be estimated to be different depending on the amount of solar radiation. The water-poor areas may be identified by extracting areas where water is likely to accumulate from a hazard map that can be acquired via the network 40, for example. For example, even in a close range on the same road, the estimation result may differ between dry and wet depending on whether the area is outside or inside the water-poor areas. Here, in generating the machine learning model, the road surface condition measurement data (measured values of the road surface condition) contained in the performance data and the values of the explanatory variables at the time of measurement are used as learning data.
[0031] Furthermore, the spatio-temporal statistical model may further include meteorological variables such as temperature and humidity as parameters that are explanatory variables, as shown in Fig. 3. By generating a spatio-temporal statistical model that takes into account temperature changes on the road surface in more detail in addition to the amount of solar radiation, the road surface temperature can be estimated with even higher accuracy using the spatio-temporal statistical model.
[0032] 1 again, the components of the road surface condition estimation device 10 will be described in detail below. The communication unit 11 includes one or more communication modules that connect to the network 40. The communication unit 11 may include a communication module compatible with mobile communication standards such as 4G (4th Generation) and 5G (5th Generation). The communication unit 11 may include a communication module compatible with a wired or wireless LAN standard.
[0033] The storage unit 12 is one or more memories. The memory may be, for example, a semiconductor memory, a magnetic memory, or an optical memory, but is not limited to these, and may be any memory. The storage unit 12 is, for example, built into the road surface condition estimation device 10, but may also be configured to be accessed from outside by the road surface condition estimation device 10 via any interface.
[0034] The storage unit 12 stores various data used in various calculations performed by the control unit 13. The storage unit 12 may also store results and intermediate data of various calculations performed by the control unit 13. In this embodiment, the storage unit 12 stores the generated model.
[0035] The control unit 13 is one or more processors. The processor may be, for example, a general-purpose processor or a dedicated processor specialized for a specific process, but is not limited to these and may be any processor. The control unit 13 controls the overall operation of the road surface condition estimation device 10.
[0036] Here, the road surface condition estimation device 10 may have the following software configuration. One or more programs used to control the operation of the road surface condition estimation device 10 are stored in the storage unit 12. When the program stored in the storage unit 12 is read by the processor of the control unit 13, it causes the control unit 13 to function as an acquisition unit 131, a model generation unit 132, an estimation unit 133, and an output unit 134.
[0037] When generating a model, the acquisition unit 131 acquires performance data for generating learning data used in machine learning. The learning data may be generated by extracting a target variable (e.g., road surface condition measurement data) and an explanatory variable (e.g., weather variables at the time of measuring the road surface condition) from the performance data. When estimating a road surface condition using a model, the acquisition unit 131 acquires a model (i.e., the above-mentioned machine learning model and spatio-temporal statistical model) stored in the storage unit 12. The acquisition unit 131 also acquires input data for a road or area that is the subject of estimation. The input data includes parameter values corresponding to explanatory variables of the model.
[0038] The model generation unit 132 generates a machine learning model and a spatio-temporal statistical model based on the performance data acquired by the acquisition unit 131. In generating the machine learning model, the model generation unit 132 executes machine learning using learning data based on the performance data. The machine learning method is not limited, and may be, for example, a neural network. In this embodiment, the model generation unit 132 first generates a spatio-temporal statistical model. Then, the estimation unit 133 generates the machine learning model by including the road surface temperature calculated by the generated spatio-temporal statistical model in the learning data.
[0039] Here, the spatio-temporal statistical model is a statistical model that takes into account the spatio-temporal structure of the road surface temperature. In this embodiment, the spatio-temporal statistical model is a model that estimates the road surface temperature that changes with time in a two-dimensional space, and is a statistical model that takes into account the spatio-temporal autocorrelation of the road surface temperature data. The structure of the spatio-temporal statistical model is not limited to a specific structure, but has a hierarchical structure described later in this embodiment. In addition, the method of estimating parameters of the spatio-temporal statistical model is not limited to a specific method, but for example, Bayesian estimation using MCMC (Markov Chain Monte Carlo method) can be used. Below, configuration examples (first to ninth models) of the spatio-temporal statistical model are shown, but the description will be given assuming that MCMC is used for parameter estimation in these configuration examples. The spatio-temporal statistical model may adopt any of the configuration examples, and it is also possible to compare several of them and adopt the optimal configuration. All of the configuration examples realize specific modeling while taking into account the spatio-temporal structure of the road surface temperature.
[0040] (First model) The first model, which is one of the configuration examples of the spatio-temporal statistical model, is shown in the following formula (1). The first model is a model that serves as the basis for the other configuration examples.
[0041]
number
[0042] Here, Y(s i , t) is the road surface temperature, which corresponds to the spatiotemporal statistical model in this embodiment. i indicates a position (site) in space, and in this embodiment, it is set to the latitude and longitude of the center of gravity of a connection part (link) in a road network. t indicates time. μ(s i ,t) is a spatiotemporal process that describes the average structure of the data. In other words, μ(s i ,t) is a part that can be calculated using the main explanatory variables and coefficients when analyzing the relationship between road surface temperature in space and time. i ,t) is the error term, which is a space-time process with zero mean. i ,t) is a vector of explanatory variables, and the subscript j indicates the j-th explanatory variable. Also, β(s i ,t) are regression coefficients, and the subscript j indicates the j-th regression coefficient. In this embodiment, the above average structure is modeled using linear regression, which can relatively easily determine regression coefficients that depend on time and space. That is, the first model is configured to include a term indicating the average structure using linear regression and an error term.
[0043] Here, in modeling the spatiotemporal process, the error term e(s i ,t) can be interpreted in various ways. For example, e(s i , t), the spatio-temporal statistical model may be constructed to have a hierarchical structure (parts of the model are further modeled).
[0044] (2nd model) The error term e(s iAssuming that nT,t) is a space-time Gaussian process with zero mean generally enables flexible modeling of space-time structures. However, when considering correlations between all sites and time, the covariance matrix of the Gaussian process becomes the inverse of the nT×nT matrix. When the number of sites and time steps is large, the covariance matrix becomes large, which increases the computational cost. The second model, which is one example of a space-time statistical model, is calculated by subtracting e(s i ,t) In the second model, the computational cost can be reduced by using a covariance function that separates space and time for the error term.
[0045]
number
[0046] Here, Cov is a function that indicates the covariance. σ 2 is a constant. s (|s i -s j |;v s ) is the parameter v s is an isotropic spatial correlation function that depends on |s i -s j | is s i ands j Similarly, ρ t (|t k -t l |;v t ) is the parameter v t is an isotropic time correlation function that depends on |t k -t l | is t k and l is the distance between.
[0047] (Third model) To model more complicated spatiotemporal structures, a model that introduces spatiotemporal random effects may be used. Specifically, an additional error term, e(s i, t) may be configured to have a hierarchical structure. For example, e(s i ,t) is expressed by the following equation (3).
[0048]
number
[0049] Here, w(s i ,t) is the random effect term. i ,t) is a term that indicates the measurement error. i ,t) and ε(s i Each of the variables w(s,t) has zero mean. Random effects are used to model spatiotemporal dependencies, i.e., they are effectively used to estimate the parameters of a spatiotemporal statistical model by giving them random values. The measurement errors follow a normal distribution and are related to location, time and w(s i We assume that the variances are different at different times, independent of time (t, t), although measurement errors may include the same variance at different times.
[0050] The third model, which is one example of a spatio-temporal statistical model, is constructed so that the Gaussian process includes a time-independent random effect, and is shown in the following equation (4).
[0051]
number
[0052] Here, we consider a model in which β does not depend on space. For measurement error, ε(s i ,t)~N(0,σ ε 2 ) The spatiotemporal structure is taken as the covariance matrix of the random effects, and the terms representing the random effects are modeled as shown in equation (5) below.
[0053]
number
[0054] Here, σ w 2 is a constant. ij s i ands j is the distance between ρ(d ij ;v) is the correlation function of the Matern family. v is a parameter. This modeling assumes that the road surface temperature data has a distance-dependent spatiotemporal structure represented by a random effect. Here, the equation for the third model can be expressed in vector notation as the following equation (6).
[0055]
number
[0056] where l is the unit row, and w t By performing marginalization on , it can be expressed as the following equation (7).
[0057]
number
[0058] The random effects expressed as vectors, w t Since integral elimination is possible for , the number of parameters can be reduced.
[0059] (4th model) The fourth model, which is one of the configuration examples of the spatio-temporal statistical model, performs parameter centering in the third model. The fourth model is shown in the following equation (8). This type of centering can make MCMC calculations more efficient.
[0060]
number
[0061] (5th model) The fifth model, which is one of the configuration examples of the spatio-temporal statistical model, explicitly considers the first-order time dependency of the spatio-temporal data in the fourth model. The fifth model is shown in the following equation (9). In this case, it is possible to make the MCMC calculations more efficient while considering the time dependency.
[0062]
number
[0063] where ρ is a coefficient. When ρ is zero, the fifth model is equivalent to the fourth model. Here, Y(s i Although it is possible to directly include a first-order time dependence in Y(s , t), the construction of the fifth model, which allows for missing values, is preferred. For example, i ,t)=ρY(s i ,t-1)+ε(s i ,t) as Y(s i If time dependence is included in the time series, a calculation problem occurs when missing values occur. Here, in the fifth model, the initial value O(s i ,0) needs to be specified. i ,0) has a mean of μ0 and a variance of σ0 2 S0, which can be determined from the Matern family of correlation functions.
[0064] (6th model) The sixth model, which is one of the configuration examples of a spatio-temporal statistical model, is a model that uses GPP (Gaussian Predictive Process). In a Gaussian process, the computational cost increases as the number of sites increases. In the sixth model, random effects of positions with a number less than the number of data sites, for example m×m knots in GPP, are considered. In addition, the random effects of the data positions are predicted using Kriging. Here, the random effects include first-order time dependence, as in the fifth model. The sixth model is expressed by the following equation (10). In this case, the increase in computational cost can be suppressed.
[0065]
number
[0066] Here, the random effect is w(s i ,t)~N(0,σ ε 2 S w* ) can be expressed as S w* is an m × m matrix. The matrix A is w -1 where C is an n × m cross-correlation matrix. m is sufficiently smaller than n, and the number of knots positions (m) is smaller than the number of sites (n) actually observed as data, so that the covariance matrix can be calculated quickly and the increase in calculation costs can be suppressed.
[0067] (7th model) The seventh model, which is one of the configuration examples of the spatio-temporal statistical model, makes the regression coefficients in the first model dependent on space. In other words, the regression coefficients are expressed by the following equation (11). By using regression coefficients that take spatial dependency into account, it is expected that the accuracy of the terms that indicate the average structure will be improved, and the impact of the error terms will be reduced.
[0068]
number
[0069] Here, β j When is zero, the spatio-temporal statistical model becomes a fixed regression coefficient model. j0 represents the overall effect of the jth explanatory variable.
[0070] (8th model) The eighth model, which is one of the configuration examples of the spatio-temporal statistical model, makes the regression coefficients in the first model dependent on time. In other words, the regression coefficients are expressed by the following equation (12). By using regression coefficients that take time dependency into account, it is expected that the accuracy of the terms that indicate the average structure will be improved, and the influence of the error terms can be reduced.
[0071]
number
[0072] Here, δ j (t) is independent of location and time. When ρ is zero, the spatio-temporal statistical model becomes a stationary regression model. The initial condition is β j (0)~N(0,σ0 2 ) and is independent of location and time.
[0073] (9th model) Model 9, which is one example of a spatio-temporal statistical model, makes the regression coefficients in Model 1 dependent on space and time. In other words, Model 9 is a combination of Models 7 and 8. By using regression coefficients that take spatio-temporal dependency into account, it is expected that the accuracy of the terms that indicate the average structure will be improved, and the impact of the error terms will be reduced.
[0074] The estimation unit 133 estimates the road surface condition using the model generated by the model generation unit 132. In this embodiment, the estimation unit 133 first calculates the road surface temperature used for estimating the road surface condition using a spatio-temporal statistical model. Then, the estimation unit 133 inputs the calculated road surface temperature into a machine learning model to estimate the road surface condition.
[0075] The output unit 134 outputs the result estimated by the estimation unit 133 to a display device or the like. For example, a display connected to the road management device 60, a display included in the terminal device 50, or the like can function as a display device that displays the estimation result.
[0076] 4 is a flowchart illustrating a process of estimating road surface conditions executed by the road surface condition estimation device 10. The acquisition unit 131 and the model generation unit 132 execute a model generation step (step S1, model generation method) of generating a model for estimating road surface conditions. The acquisition unit 131 and the estimation unit 133 execute a road surface condition estimation step (step S2, road surface condition estimation method) of estimating the road surface condition of the road that is the subject of estimation using the generated model. The road surface condition estimation step may be executed consecutively to the model generation step, or may be executed after a certain time has elapsed after the model generation step.
[0077] Fig. 5 is a flowchart showing details of the model generation step of Fig. 4. The acquisition unit 131 acquires, for example, performance data accumulated in a database of the road management device 60 (step S11). The acquisition unit 131 may also acquire map information, hazard maps, and the like via the network 40.
[0078] The learning data is generated by extracting a target variable and an explanatory variable from the performance data acquired by the acquisition unit 131 (step S12). The amount of solar radiation may be calculated as described above. At this time, the generation of the learning data and the calculation of the amount of solar radiation may be executed by the model generation unit 132.
[0079] Then, the model generation unit 132 generates a model (step S13). In this embodiment, the model generation unit 132 first generates a spatio-temporal statistical model, and then generates a machine learning model. The model generation unit 132 stores the generated model in the storage unit 12.
[0080] Fig. 6 is a flowchart showing details of the road surface condition estimation process in Fig. 4. The acquisition unit 131 acquires input data (step S21), and acquires a model from the storage unit 12 (step S22). The input data includes parameter values corresponding to explanatory variables of the model as described above.
[0081] Here, the amount of solar radiation is calculated based on the duration of solar radiation, topography, building information, etc. included in the input data. At this time, the calculation of the amount of solar radiation may be executed by the estimation unit 133. The estimation unit 133 first calculates the road surface temperature using a space-time statistical model (step S23). Then, the estimation unit 133 inputs the calculated road surface temperature into a machine learning model to estimate the road surface condition (step S24). The output unit 134 may output the result estimated by the estimation unit 133. The output estimation result may be displayed on a display device so that the road administrator or the driver of the vehicle 20 can confirm it.
[0082] As described above, the road surface condition estimation device 10 and road surface condition estimation method according to this embodiment are able to calculate a road surface temperature close to an actual distribution by using a space-time statistical model that takes into account the space-time structure of the road surface temperature, and to estimate the road surface condition with high accuracy based on such road surface temperature.
[0083] Although the embodiments of the present disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art can easily make various modifications or corrections based on the present disclosure. Therefore, it should be noted that these modifications or corrections are included in the scope of the present disclosure. For example, the functions included in each component or each step can be rearranged so as not to be logically inconsistent, and multiple components or steps can be combined into one or divided. The embodiments of the present disclosure can also be realized as a program executed by a processor included in the device and a storage medium on which the program is recorded. It should be understood that these are also included in the scope of the present disclosure.
[0084] For example, the configurations of the road surface condition estimation device 10 and the road surface condition estimation system shown in FIG. 1 and FIG. 2 are only examples, and are not limited to the configurations of FIG. 1 and FIG. 2. For example, the road surface condition estimation system may be configured by integrating the road surface condition estimation device 10 and the road management device 60. In this case, the road surface condition estimation device 10 may also execute the functions of the road management device 60 (for example, support for snow removal or snow melting work plans). In addition, when the vehicle 20 is an autonomous vehicle, the road surface condition estimation system may be configured to include a driving management device that manages the driving of the autonomous vehicle together with the road management device 60 or instead of the road management device 60. The driving management device may acquire information on the road surface condition estimated from the road surface condition estimation device 10 and determine whether or not to continue autonomous driving depending on the road surface condition of the road on which the vehicle is scheduled to travel. By estimating the road surface condition with high accuracy, it is possible to appropriately determine whether or not to continue driving, and the driving safety of the vehicle 20 can be improved. In addition, the database may be managed and stored by the road surface condition estimation device 10, not by the road management device 60.
[0085] Also, the model generation unit 132 and the estimation unit 133 may be included in different computers. For example, the model generation unit 132 may be included in another computer (an independent model generation device) that is capable of communicating with the road surface condition estimation device 10 and also has access to the memory unit 12. In this case, the road surface condition estimation device 10 may determine the road surface condition using a model that is generated by the other computer and stored in the memory unit 12. The model may be generated in the same manner as above. In other words, the other computer may acquire performance data and generate a model using learning data based on the performance data. Contribution to the United Nations-led Sustainable Development Goals (SDGs)
[0086] The SDGs have been proposed to realize a sustainable society. One embodiment of the present disclosure is considered to be a technology that can contribute to "No. 9 Build infrastructure for industry and technological innovation" and other goals. [Explanation of symbols]
[0087] 10 Road surface condition estimation device 11 Communications Department 12 Storage section 13 Control section 20 Vehicles 40 Network 50 Terminal Equipment 60 Road management equipment 131 Acquisition Department 132 Model Generation Unit 133 Estimation Department 134 Output section
Claims
1. an acquisition unit that acquires a spatio-temporal statistical model having a road surface temperature as a response variable and a parameter including at least the amount of solar radiation as an explanatory variable; an estimation unit that calculates the road surface temperature used to estimate a road surface condition by using the spatio-temporal statistical model; A road surface condition estimation device, wherein the spatio-temporal statistical model is a statistical model that takes into account the spatio-temporal structure of the road surface temperature.
2. The road surface condition estimating device according to claim 1 , wherein the spatio-temporal statistical model is configured to include a term indicating a mean structure using linear regression, and an error term.
3. The road surface condition estimating device according to claim 2 , wherein the spatio-temporal statistical model uses a covariance function for the error term that separates space and time.
4. The road surface condition estimating device according to claim 2 , wherein the spatio-temporal statistical model is configured such that the error term includes a random effect.
5. The road surface condition estimating device according to claim 4 , wherein the spatio-temporal statistical model performs parameter centering.
6. The road surface condition estimating device according to claim 5 , wherein the spatio-temporal statistical model is configured to explicitly take into account first-order time dependence in the spatio-temporal data.
7. The road surface condition estimating device according to claim 2 , wherein the spatio-temporal statistical model uses a Gaussian Predictive Process (GPP).
8. The road surface condition estimating device according to claim 2 , wherein the spatio-temporal statistical model makes regression coefficients spatially dependent.
9. The road surface condition estimating device according to claim 2 , wherein the spatio-temporal statistical model makes regression coefficients time-dependent.
10. The road surface condition estimating device according to claim 2 , wherein the spatio-temporal statistical model makes regression coefficients dependent on space and time.
11. Obtaining a spatio-temporal statistical model with road surface temperature as a response variable and parameters including at least solar radiation as explanatory variables; Calculating the road surface temperature used to estimate a road surface condition using the spatio-temporal statistical model; A road surface condition estimating method, wherein the spatio-temporal statistical model is a statistical model that takes into account a spatio-temporal structure of the road surface temperature.
Citation Information
Patent Citations
System for predicting road surface condition of fixed observation point, road surface temperature distribution and road surface condition of route
JP2009243884A
