JP7422562B2 - Road surface condition prediction method - Google Patents

Description

The present invention relates to a road surface condition prediction method for predicting road surface conditions at a future time.

Conventionally, one method of predicting future road surface conditions is to predict the road surface condition at a future time by calculating the transition probability of the estimated road surface condition obtained in advance, while sequentially correcting it based on weather information. A method has been proposed (Patent Document 1).

JP2016-80451A

It is certainly possible to improve the accuracy of predicting future road surface conditions by sequentially applying corrections based on road surface condition transition probabilities and weather information.
However, it is actually difficult and inefficient to accumulate road surface condition data frequently enough to withstand time-series analysis, such as sequential correction. In addition, the actual road surface, especially a snowy or frozen road surface, is affected by traffic volume and management operations such as spraying snow melting agents, so there is also the problem that it is difficult to improve the accuracy of predicting road surface conditions.
Therefore, an object of the present invention is to provide a road surface condition prediction method that can efficiently predict the road surface condition at a future time and improve prediction accuracy.

As an aspect of the road surface condition prediction method for solving the above-mentioned problems, vehicle information, which is information on the behavior of the vehicle during driving, is acquired by an on-board sensor installed in the vehicle, and an estimated road surface obtained based on the vehicle information is obtained. A road surface condition prediction method that predicts the road surface condition within a predetermined range including the location of the vehicle using the state, and the road surface condition is estimated from weather information within the predetermined range and the most recent time when snow melting agent was sprayed. Based on the residual salt concentration calculated according to the elapsed time and the type of snow melting agent, the road surface condition is predicted at a time in the future rather than the time when the estimated road surface condition was obtained.
According to this aspect, it is possible to efficiently predict the road surface condition at a future time and to improve prediction accuracy.
In addition to weather information and residual salinity concentration, the road surface condition can be predicted at a time in the future from the time when the estimated road surface condition was obtained, based on the traffic volume within a predetermined range, and the residual salinity concentration can be used to The rate of decrease over time may be changed depending on the amount.
In addition, the road surface condition is determined by using the estimated road surface condition obtained based on vehicle information, which is information about the behavior of the vehicle during driving, acquired by an on-board sensor installed in the vehicle, as the objective variable, The estimated road surface condition is estimated by machine learning using a computer using meteorological information within the range, the elapsed time since the most recent snow melting agent application, and the residual salt concentration calculated according to the type of snow melting agent as explanatory variables. The system now predicts the road surface condition at a future time rather than at the current time.
According to this aspect, it is possible to efficiently predict the road surface condition at a future time and to improve prediction accuracy.

Note that the above summary of the invention does not list all the necessary features of the present invention, and each structure constituting the feature group can also be an invention.

FIG. 1 is a configuration diagram of a road surface condition prediction system. It is a figure showing attachment of an acceleration sensor. FIG. 3 is a diagram showing the configuration and processing concept of a road surface condition estimation processing section. FIG. 3 is a conceptual diagram showing a process of classifying vehicle information by vehicle information processing means. FIG. 3 is a diagram showing an example of information recorded on a server. It is a figure which shows the prediction result when road surface conditions are predicted at a certain time in the future. FIG. 6 is a diagram illustrating a display example of a prediction result when the future road surface condition over a certain period of time is predicted at predetermined time intervals.

Hereinafter, the present invention will be explained in detail through embodiments of the invention, but the following embodiments do not limit the claimed invention, and all combinations of features described in the embodiments are applicable to the invention. It may not be necessary for the solution.

FIG. 1 is a configuration diagram of a road surface condition prediction system 1 according to this embodiment.
As shown in FIG. 1, the road surface condition prediction system 1 according to the present embodiment includes a plurality of vehicles Wi (i=1 to N) equipped with on-vehicle sensors capable of detecting road surface conditions, and It includes a server 20 into which information etc. are input, and a road surface condition prediction device 30 which predicts road surface conditions based on the information input into the server 20. The server 20 and the road surface condition prediction device 30 are provided in the road surface condition management center 2, for example.

Each vehicle Wi (i=1 to N) includes an acceleration sensor 11 as an on-vehicle sensor, a GPS device 12, and a road surface condition estimation device 13. Note that N is, for example, the "total number" of vehicles that have traveled in a predetermined range in which the road surface condition is predicted by the road surface condition prediction device 30, which will be described later.
The acceleration sensor 11 is provided in the air chamber 42 of the tire 40 assembled to the rim 44 of the vehicle, and the GPS device 12 and the road surface condition estimation device 13 are provided in the body of the vehicle.

As shown in FIG. 2, the acceleration sensor 11 is attached to the approximate center of the inner liner portion 41 of the tire 40, and detects vibrations (tire vibrations) input to the tread 43 of the tire 40 from the road surface R on which the vehicle is running. The acceleration sensor 11 includes a communication means (not shown) for outputting detected vibrations to the road surface condition estimating device 13. A low-power wireless communication device such as Bluetooth (registered trademark) is used as the communication means.

The GPS device 12 includes a GPS antenna and a receiver (not shown), and acquires positional information of the vehicle Wi while it is traveling, and calculates the traveling speed of the vehicle based on the acquired positional information.

The road surface condition estimating device 13 is a computer equipped with a CPU as a calculation means, a ROM and a RAM as storage means, and an input/output interface as a communication means. The road surface condition estimating device 13 includes, as an input/output interface, an Internet wireless communication device that can be wirelessly connected to the Internet, and a low-power wireless communication device that enables communication with the acceleration sensor 11. The road surface condition estimating device 13 causes the road surface condition estimating device 13 to function as each section and each means described below by the CPU executing processing according to a program stored in the storage means.

The road surface condition estimation device 13 includes a road surface condition estimation processing section 130 and a vehicle information collection section 14, and determines whether the road surface on which the vehicle is traveling is a DRY road surface, based on the time series waveform of tire vibration input from the acceleration sensor 11. It is estimated whether the road surface is a wet road surface, a SNOW road surface, or an ICE road surface.

FIG. 3 is a diagram showing the configuration and processing concept of the road surface condition estimation processing section 130.
As shown in FIG. 3A, the road surface condition estimation processing unit 130 includes a vibration waveform extraction means 131 that extracts a time-series waveform of tire vibration from the acceleration sensor 11, a windowing means 132, and a feature vector calculation means 133. , four road surface models 134 stored in the storage means, a kernel function calculation means 135, and a road surface condition determination means 136, and the time-series waveform of tire vibration extracted by the vibration waveform extraction means 131 is shown in FIG. As shown in b), a plurality of time windows are extracted by applying a window function of a predetermined time width T by the windowing means 132, and a plurality of time windows are respectively extracted from the time series waveform for each time window extracted by the feature vector calculation means 133. A feature vector Xk (aK1, ak2, . . . , akm) having vibration levels (ak1 to akm) in a specific frequency band as components is calculated, and the feature vector Xk calculated by the kernel function calculation means 135 is calculated in advance. A kernel function is calculated from the feature vector for each road surface condition, and the road surface condition determining means 136 determines whether the road surface condition is a DRY road surface, a WET road surface, a SNOW road surface, or an ICE road surface from the value of a discriminant function using the calculated kernel function. Estimate whether Note that the above-mentioned feature vector for each road surface condition is a feature vector whose components are the vibration levels of multiple specific frequency bands when the test vehicle is driven on a dry road surface, a wet road surface, a SNOW road surface, and an ICE road surface. be.

Note that the road surface condition estimation process by the road surface condition estimation processing section 130 is not limited to the above configuration and method, and can estimate the road surface condition based on the time series waveform of tire vibration obtained from the acceleration sensor 11. It's good to have.

Further, the road surface condition estimation processing section 130 may be provided within the tire 40. In this case, the road surface condition estimated by the road surface condition estimation processing section 130 may be output to the vehicle information collection section 14. Alternatively, some functions of the road surface condition estimation processing section 130 may be separated, and information on the acceleration waveform detected by the acceleration sensor 11 may be processed within the tire 40. For example, a band value used to estimate road surface conditions such as the vibration level of a specific frequency band detected from an acceleration waveform, or a function that calculates a calculated value of the band value inside the tire and leaves it on the vehicle body. The configuration may be such that processing is performed by

The vehicle information collection section 14 collects the road surface condition estimated by the road surface condition estimation processing section 130 (hereinafter referred to as estimated road surface condition) and the vehicle position information acquired by the GPS device 12, and identifies the vehicle. A vehicle ID is assigned to the vehicle and output to the Internet wireless communication device as vehicle information of the vehicle. The vehicle information also includes information on the vehicle ID and the acquisition time of the vehicle information (time information). The time information may be, for example, the extraction time of the acceleration time series waveform, the acquisition time of the position information, or the transmission time when the vehicle information is output to the Internet wireless communication device. Since these times are almost the same time, there is no problem which time can be used as time information. Vehicle information output from the vehicle information collection unit 14 to the Internet wireless communication device is input to the server 20 via the Internet line.

The server 20 is a so-called computer, and includes a CPU as a calculation means provided as hardware resources, a ROM and a RAM as a storage means 20A, an input/output interface as a communication means, and the like. The server 20 is configured to be connectable to, for example, an Internet line via an input/output interface, and is configured to be able to communicate with the aforementioned road surface condition estimation device 13 and road surface condition prediction device 30 via the Internet line. The storage unit 20A stores a program that causes the server 20 to function as a unit to be described later by executing processing by the CPU.

FIG. 4 is a conceptual diagram showing the process of classifying vehicle information by the vehicle information processing means 21. As shown in FIG.
FIG. 5 is a diagram showing an example of information recorded on the server 20.
The server 20 includes a vehicle information processing means 21 that is input from the vehicle Wi, a weather information acquisition means 22 that acquires weather information, a traffic information acquisition means 23 that acquires traffic information, and acquires road management information. road management information acquisition means 24.
The vehicle information processing means 21 classifies the vehicle information input from each vehicle Wi (i=1 to N) based on the position information included in the vehicle information.

In the following explanation, processing by the vehicle information processing means 21 will be explained using one road R. The vehicle information processing means 21 acquires position information from the input vehicle information, and divides one road R into a plurality of sections Li (i indicates the number of sections) based on the acquired position information, as shown in FIG. It is determined whether the vehicle is included in any of the sections Li that have been divided in advance, and the vehicle information is classified for each section Li. The classified vehicle information is stored in the storage means 20A as a file for each section Li, for example. Further, the vehicle information stored for each section Li is stored in chronological order so as to be accumulated as time series data, and is stored in the storage means 20A as vehicle information data. Such processing may be performed on the road R whose road surface condition is to be predicted.

The weather information acquisition means 22 acquires weather information corresponding to each section Li, that is, weather information within a predetermined range including the point where the estimated road surface condition was obtained, and links the acquired weather information to each section Li. The weather information is stored in the storage means 20A as weather data. Meteorological data refers to time-series data that includes items such as air temperature (atmospheric temperature), rainfall, snowfall, snow cover conditions (snow depth), sunlight, wind speed, etc. For example, as shown in Figure 5, temperature data, rainfall data, snowfall data, snow depth data, sunshine data, wind speed data, etc. are stored in the storage means 20A. The weather information acquisition means 22 acquires weather information by automatically acquiring, for example, the information observed at the nearest weather observation point from each point where the road surface condition is estimated from a weather information providing site on the Internet. In addition, workers at the road surface condition management center 2 may be provided with numerical values of the above-mentioned items observed by a weather meter, snow depth meter, etc. at the nearest meteorological observation point from each point where the road surface condition is estimated. You may be prompted to input.

The traffic information acquisition means 23 acquires the traffic information corresponding to each section Li, and stores the acquired traffic information in the storage means 20A in association with each section Li as traffic data. Traffic volume data refers to time series data of the number of vehicles detected by traffic volume counters installed on roads. The traffic information acquisition means 23 may automatically acquire the number of vehicles detected by a traffic counter, or the traffic information acquisition means 23 may automatically acquire the number of vehicles detected by a traffic counter. , the user may be prompted to input the number of vehicles detected by the traffic counter.

Furthermore, if the direction of movement of vehicles passing through the prediction target range is divided by a center line, such as up and down, the traffic volume information is acquired for each up and down direction. Furthermore, if the traffic information is not divided into up and down sections, it is obtained as time-series data on the number of vehicles in both directions. Note that in the following explanation, road traffic information is simply referred to as traffic information).

The road management information acquisition means 24 acquires the road management information corresponding to each section Li, associates the acquired road management information with each section Li, and stores it in the storage means 20A as road management data. The road management information referred to here refers to snow and ice road surface management work, and information regarding snow melting agent spraying work is stored. Specifically, the spraying time of the snow melting agent, the spraying location, the spraying area, the amount of spraying, and the type of the snow melting agent sprayed are recorded. The types of snow melting agents are classified into, for example, wet agents (liquid), solid agents (particulate), and the like.

The road surface condition prediction device 30 is a so-called computer, and includes a CPU as a calculation means provided as hardware resources, a ROM and a RAM as storage means, an input/output interface as a communication means, and the like. The road surface condition prediction device 30 is configured to be connectable to the Internet via an input/output interface, for example, and exchanges information with the server 20 via the Internet. The road surface condition prediction device 30 is connected to input means such as a keyboard and mouse, display means for displaying prediction results, and the like.
In the road surface condition prediction device 30, the CPU processes and executes a program stored in the storage means, thereby causing the road surface condition prediction device 30 to function as each means and each section described below, and predicts the road surface condition of the target road. Predict.

The road surface condition prediction device 30 includes, for example, an information reading means 31, an information format unification means 32, a predictive model creation means 33, a road surface condition acquisition means 34, and a road surface condition determination means 35, and includes estimated road surface condition information, Based on weather information, road traffic information, and road management information, the road surface condition of the road whose road surface condition is to be predicted is predicted at a future time.

The information reading means 31 reads vehicle information data, weather data, traffic volume data, and road management data from the server 20.
The information format unifying means 32 executes processing for unifying each data read by the information reading means 31 into the same file format. Note that the information format unifying means 32 is not necessary if the file formats in which vehicle information, weather information, traffic volume information, and road management information are recorded in the server 20 are unified.

The prediction model creation means 33 creates a road surface condition prediction model for predicting future road surface conditions based on vehicle information data, weather data, traffic volume data, and road management data. That is, the prediction model creation means 33 uses the road surface conditions estimated by vehicles that actually drove on the road in the past from the present time (at the time of prediction execution), and the weather conditions, traffic volume, and road conditions at the time when the road surface conditions were estimated. A road surface condition prediction model is created to predict future road surface conditions based on the management conditions.

In this embodiment, the predictive model creation means 33 uses the estimated road surface condition included in each vehicle information constituting the vehicle information data as a target variable in machine learning, and uses weather information, traffic volume, etc. when the estimated road surface condition is estimated. The following describes how a road surface condition prediction model is created by machine learning using information and road management information as explanatory variables in machine learning.
In this embodiment, in addition to the effects of continuous rainfall, snowfall, snow depth, wind speed, traffic volume, and snow melting agents, the effects of road surface temperature and past road surface conditions are used as explanatory variables to determine whether the future road surface condition is a dry road surface or a wet road surface. , a road surface condition prediction model is created to calculate the probability of an ICE-SNOW road surface. Note that the influence of the snow melting agent takes into consideration the work of dispersing the snow melting agent included in the road management information, and is expressed as a numerical value as described below. In addition, the influence of past road surface conditions takes into consideration the degree of influence that the most recently observed road surface condition will have on future road surface conditions, and is expressed numerically like the influence of snow melting agents.

The prediction model creation means 33 includes a road surface temperature prediction model creation section 33A, a data set creation section 33B, and a road surface condition prediction model creation section 33C.
The road surface temperature prediction model creation unit 33A creates a road surface temperature prediction model for setting the road surface temperature as an explanatory variable in creating the road surface condition prediction model. The road surface temperature prediction model creation unit 33A uses machine learning to use the actual value of the road surface temperature measured in the prediction target range as the objective variable, and the air temperature, snow depth, and traffic volume at the time the road surface temperature was measured as the explanatory variables. , create a machine learning model (hereinafter referred to as a road surface temperature prediction model) to obtain road surface temperature.
That is, so-called supervised machine learning that uses actual values measured at multiple points within the prediction target range, and the air temperature, snow depth, and traffic volume at the time when the road surface temperature was measured at each point as one data set. Train a classifier to predict road surface temperature using a based classification algorithm, e.g., Random Forest.
The actual measured value of the road surface temperature can be, for example, one measured in advance by a sensor installed on the road, or one measured in conjunction with road management work such as the above-mentioned spraying of snow melting agent.
By calculating the road surface temperature using the road surface temperature prediction model in this way, even if the road surface temperature corresponding to the position and time when the estimated road surface state is input to the server 20 cannot be obtained, the estimated road surface state can be calculated by the server 20. The road surface temperature corresponding to the position and time inputted can be set with high accuracy.
The temperature, snow depth, and traffic volume used as explanatory variables are acquired from the weather information and traffic volume information acquired by the server 20.

The dataset creation unit 33B extracts weather information, traffic volume information, and road management information from the weather data, traffic volume data, and road management data at the time when each estimated road surface condition included in the vehicle information data is obtained, and extracts each estimated road surface condition from the weather data, traffic volume data, and road management data. Create multiple datasets linked to road surface conditions.

ここで、Ｔｅｆｆｅｃｔは、路面状態の変化の速さを表す定数である。上記式１乃至式３は、ｔが大きくなると（観測時点がかなり前になると）すべて同じ値に近づくが、ｔが小さいと直前の路面状態に相当する数値に近い式が大きくなるという特性がある。そこで、本実施形態では、後述の他の説明変数の設定を考慮し、Ｔｅｆｆｅｃｔには、０．５ｈ（０．５時間）を設定した。
なお、本実施形態では、推定路面状態を乾燥（ＤＲＹ）：０、湿潤（ＷＥＴ）：２、凍結（ＳＮＯＷ－ＩＣＥ）：６に数値化するものとして説明したがこれに限定されず、各車両において推定される路面状態の区分に応じて、例えば、乾燥：０、半湿潤：１、湿潤：２、・・・・、凍結：６等のように数値化しても良い。
連続雨量には、降雨データにおける推定時刻から過去１時間内の最大値を設定する。
降雪量には、降雪データにおける推定時刻から過去１時間の合計値を設定する。
積雪深さは、降雪データにおける推定時刻から過去１時間の平均値を設定する。
風速には、風速データにおける推定時刻から過去１時間の平均値を設定する。
交通量には、交通量データにおける推定時刻から過去１時間の合計値を設定する。
融雪剤の影響とは、融雪剤の散布から道路上に残る塩分濃度を示す指標であって、散布からの経過時間や融雪剤の種別に応じて残留塩分濃度として算出される。
残留塩分濃度は、散布時の塩分濃度を１００とし、
融雪剤が湿式剤（液状）の場合は、Ｅｘｐ（－経過時間/６ｈ）-----（式４）
融雪剤が固剤（粒子状）の場合は、Ｅｘｐ（－経過時間/１８ｈ）---（式５）
で減少するものとして算出される。経過時間とは、融雪剤が散布された時刻から路面状態の推定を実施した時刻までの時間である。 In this embodiment, the dataset creation unit 33B links road surface temperature, influence of past road surface conditions, continuous rainfall amount, snowfall amount, snow depth, wind speed, traffic volume, influence of snow melting agent, etc. to each estimated road surface condition. and set them as follows. Note that the following estimated time refers to the time at which the road surface condition was estimated in the vehicle.
The road surface temperature is determined by inputting the air temperature, snow depth, and traffic volume at the estimated time into the road surface temperature prediction model created by the road surface temperature prediction model creation section 33A, and selecting the value with the highest likelihood among the obtained predicted values. Set.
The influence of past road surface conditions is determined by quantifying each estimated road surface condition as follows: dry (DRY): 0, wet (WET): 2, frozen (SNOW-ICE): 6, and the time at which the road surface condition is predicted from the estimated time. The dummy variables 1 to 3 are calculated using the following equations 1 to 3, where t is the time until t. Equations 1 to 3 are used to quantify the degree of influence of the most recently observed road surface condition on the future road surface condition when predicting the future road surface condition, and the dummy variables 1 to 3 are It is set as an influence of past road surface conditions.

Here, Teffect is a constant representing the speed of change in road surface condition. Equations 1 to 3 above all approach the same value as t becomes larger (when the observation time becomes much earlier), but when t becomes smaller, the equations that are closer to the value corresponding to the immediately preceding road surface condition have a characteristic that they become larger. . Therefore, in this embodiment, 0.5h (0.5 hours) is set for Teffect in consideration of the settings of other explanatory variables that will be described later.
In the present embodiment, the estimated road surface condition is expressed numerically as 0 for dry (DRY), 2 for wet (WET), and 6 for frozen (SNOW-ICE), but the present invention is not limited to this. Depending on the classification of the road surface condition estimated in, for example, dry: 0, semi-humid: 1, wet: 2, . . . , frozen: 6, etc. may be expressed numerically.
The continuous rainfall amount is set to the maximum value within the past hour from the estimated time in the rainfall data.
The snowfall amount is set as the total value for the past hour from the estimated time in the snowfall data.
The snow depth is set as the average value for the past hour from the estimated time in the snowfall data.
For the wind speed, an average value for the past hour from the estimated time in the wind speed data is set.
The traffic volume is set to the total value for the past hour from the estimated time in the traffic volume data.
The influence of the snow melting agent is an index indicating the salt concentration remaining on the road after the snow melting agent is sprayed, and is calculated as the residual salt concentration depending on the elapsed time since the snow melting agent and the type of snow melting agent.
The residual salt concentration is based on the salt concentration at the time of spraying as 100,
If the snow melting agent is a wet agent (liquid), Exp(-elapsed time/6h) -----(Formula 4)
If the snow melting agent is a solid agent (particulate), Exp(-elapsed time/18h)---(Formula 5)
It is calculated as decreasing by . The elapsed time is the time from the time when the snow melting agent was sprayed to the time when the road surface condition was estimated.

The road surface condition prediction model creation section 33C uses the plurality of data sets created in the dataset creation section 33B to create a classifier for predicting the road surface state using a machine learning-based classification algorithm, such as a random forest. Learn and create a road surface condition prediction model. Note that the classification algorithm is not limited to random forest, and may be other algorithms such as, for example, logistic regression or support vector machine. Preferably, a tree-based algorithm such as a random forest or a decision tree is used, which has the advantage that the mechanism for predicting the road surface condition can be easily considered.

The road surface condition acquisition means 34 predicts the road surface condition at a desired future time (referred to as after a predetermined time) based on the prediction model created by the road surface condition prediction model creation section 33C.
The explanatory variables used to create the road surface condition prediction model are used as conditions for predicting future road surface conditions. In other words, the future road surface condition is calculated based on the road surface temperature after a predetermined period of time, the influence of past road surface conditions, continuous rainfall amount, snowfall amount, snow depth, wind speed, traffic volume, and residual salt concentration calculated based on rainfall data. Set as a condition for predicting.

The road surface temperature is determined by applying the air temperature, snow depth, and traffic volume at the time of prediction to the road surface temperature prediction model created by the road surface temperature prediction model creation unit 33A, and the value with the highest likelihood among the obtained predicted values. is set.
As for the influence of the past road surface condition, dummy variables 1 to 3 are set based on the road surface condition immediately before the prediction is performed, which are calculated by the above equations 1 to 3.
Snow depth and wind speed are averaged over the past hour at the time of forecasting.
The amount of snowfall shall be the total value for the past hour at the time of forecast execution.
Continuous rainfall is the maximum value for the past hour at the time of forecasting.
The traffic volume is the average of the traffic volume up to the present hour over the past hour at the time of prediction.
The residual salt concentration is set as the residual salt concentration estimated using Equation 4 or Equation 5 above, according to the time and type from the most recent snow melting agent spraying record to the predicted implementation time included in the road management information.

The road surface condition acquisition means 34 inputs the road surface temperature, continuous rainfall amount, snowfall amount, snow depth, wind speed, traffic volume, and residual salt concentration after the predetermined time set above into the road surface condition prediction model. A predicted value indicating the road surface condition is calculated. Since the prediction model of this embodiment is configured using a random forest algorithm, a plurality of predicted values indicating any one of a DRY road surface, a WET road surface, and an ICE-SNOW road surface are calculated.

The road surface condition determination means 35 calculates the appearance probability of each road surface condition among the plurality of predicted values obtained by the road surface condition acquisition means 34, and selects the road surface condition that appears most frequently as a future predicted value as the road surface condition. Output as .

For example, a worker at the center 2 selects a road (section Li) whose road surface condition is to be predicted from a map displayed on a display device connected to the road surface condition prediction device 30 via an input device.
Next, the information reading means 31 reads vehicle information data, weather data, traffic volume data, and road management data corresponding to the selected section Li from the server 20.
Next, the information format unification means 32 converts the read vehicle information data, weather data, traffic volume data, and road management data into a predetermined file format.
Next, the road surface temperature prediction model creation section 33A creates a road surface temperature prediction model for setting the road surface temperature in the data set creation section 33B.
Next, the dataset creation unit 33B acquires the estimated road surface condition from each piece of vehicle information constituting the vehicle information data, and also obtains the maximum value of continuous rainfall in the past hour, the total value of snowfall in the past hour, and the past Based on the average snow depth over the past hour, the average wind speed over the past hour, the total traffic volume over the past hour, the elapsed time until the time the snow melting agent was sprayed, and the type of snow melting agent sprayed. The residual salt concentration calculated by Equation 4 or Equation 5 above, the temperature at the time when the estimated road surface condition was obtained, the snow depth, and the traffic volume are input into the road surface temperature prediction model, and the highest likelihood is output. Three dummy variables 1 to 3 are set based on the road surface temperature and the road surface condition immediately before the prediction is performed using the above equations 1 to 3. These numerical settings are linked to each estimated road surface condition and created as a data set.
Next, the road surface condition prediction model creation section 33C performs machine learning using the data set obtained in the data set creation section 33B to determine the road surface state at a future time as a DRY road surface, a WET road surface, or an ICE-SNOW road surface. Create a predictive model to make predictions.
Next, a desired future time to be predicted is input, and prediction conditions are set through processing by the road surface condition acquisition means 34.
Next, the prediction conditions set in the road surface condition acquisition means 34 are input into the prediction model. As a result, a plurality of predicted values indicating one of the DRY road surface, WET road surface, and ICE-SNOW road surface are calculated.
Next, the road surface condition determining means 35 calculates the probability of appearance of each road surface condition among the plurality of predicted values obtained, and outputs the road surface condition that appears most frequently as the future predicted value as the road surface condition. .
Then, the occurrence probability of the DRY road surface, WET road surface, and ICE-SNOW road surface obtained by the processing of the road surface condition determining means 35, and the road surface condition that appears most frequently are determined as future predicted values of the DRY road surface, WET road surface. , ICE-SNOW road surface may be displayed on the display means to inform the operator of the future road surface condition.

FIG. 6 is a diagram showing a prediction result when the road surface condition at a certain time in the future is predicted. FIG. 7 is a diagram showing a display example of a prediction result when the future road surface condition over a certain period of time is predicted at predetermined time intervals. According to the road surface condition prediction device 30 described above, a prediction model is created using numerical values such as road surface temperature, dummy variables 1 to 3, continuous rainfall amount, wind speed, snowfall amount, snow depth, traffic volume, residual salt concentration, etc. shown in FIG. By inputting the information into the road surface condition estimation model created in means 33, the road surface condition at a specific time is probabilistically predicted with respect to the possibility that it will be DRY, the possibility that it will be WET, and the possibility that it will be ICE-SNOW. be able to.
Then, as shown in FIG. 7, by predicting and displaying the road surface condition over a certain period of time in the future at predetermined time intervals, changes in the future road surface condition can be easily grasped.
This makes it easier, for example, to manage the amount of snow melting agent in stock required for managing road surface conditions and to formulate a snow removal work plan.

As explained above, vehicle information, which is information on the behavior of the vehicle while driving, is acquired by an on-vehicle sensor installed in the vehicle, and the estimated road surface condition obtained based on the vehicle information is used to include the location of the vehicle. A road surface condition prediction method for predicting road surface conditions within a predetermined range, the road surface condition being predicted at a time in the future from the time when the estimated road surface condition was obtained, based on weather information and residual salt concentration within the predetermined range. Accordingly, prediction accuracy can be improved.
In addition, in addition to weather information and residual salt concentration, the road surface condition at a future time is predicted by predicting the road surface condition at a time in the future from the time when the estimated road surface condition was obtained, based on the traffic volume within a predetermined range. Prediction accuracy can be further improved.
Furthermore, by setting the residual salt concentration so that the rate of decrease over time changes depending on the traffic volume, it is possible to further improve the accuracy of predicting the road surface condition at a future time.
In addition, as mentioned above, in manufacturing the road surface condition prediction model, vehicle information, which is information about the behavior of the vehicle during driving, acquired by the on-board sensor mounted on the vehicle, is used by the vehicle traveling within a predetermined range. The estimated road surface condition obtained based on the estimated road surface condition is used as the objective variable, and the meteorological information and residual salt concentration within the predetermined range from which the estimated road surface condition was obtained are used as explanatory variables. It is possible to improve the prediction efficiency of the road surface condition at the time of , and it is also possible to improve the prediction accuracy of the road surface condition.

The explanatory variables used in creating the road surface condition prediction model are not limited to the above items, but can also be selectively selected from available information based on past natural conditions, traffic volume, and road surface conditions, such as adding rainfall. It can also be used for.

1. Road surface condition prediction system, 2. Road surface condition management center,
11 acceleration sensor, 13 road surface condition estimation device, 14 vehicle information collection unit,
20 server, 20A storage means, 21 vehicle information processing means,
22 Weather information acquisition means, 23 Traffic information acquisition means, 24 Road management information acquisition means,
30 road surface condition prediction device, 31 information reading means, 32 information format unification means,
33 prediction model creation means, 33A road surface temperature prediction model creation unit,
33B Data set creation unit, 33C Road surface condition prediction model creation unit,
34 road surface condition acquisition means, 35 road surface condition determination means,
40 tire, 41 inner liner section, 42 tire air chamber, 43 tread,
130 road surface condition estimation processing section, 131 vibration waveform extraction means, 132 window hanging means,
133 Feature vector calculation means, 134 Road surface model, 135 Kernel function calculation means,
136 Road surface condition determination means.

Claims (4)

Vehicle information, which is information on the behavior of the vehicle while driving, is acquired by an on-vehicle sensor installed in the vehicle, and the estimated road surface condition obtained based on the vehicle information is used to determine the area within a predetermined range including the location of the vehicle. A road surface condition prediction method for predicting a road surface condition, the method comprising:
The estimation is based on the weather information within the predetermined range, the elapsed time from the most recent time when the snow melting agent was sprayed to the time when the road surface condition is estimated, and the residual salt concentration calculated according to the type of snow melting agent. A road surface condition prediction method characterized by predicting a road surface condition at a time in the future from the time when the road surface condition was obtained. 2. The method according to claim 1, wherein the road surface condition is predicted at a time in the future from the time when the estimated road surface condition is obtained, based on the traffic volume within the predetermined range in addition to the weather information and the residual salt concentration. The road surface condition prediction method described. 3. The road surface condition prediction method according to claim 1, wherein the residual salt concentration changes at a rate of decrease over time depending on the traffic volume within the predetermined range . The road surface condition is
The objective variable is the estimated road surface condition obtained based on vehicle information, which is information on the behavior of the vehicle during driving, obtained by an on-board sensor installed in the vehicle.
Machine learning is performed by a computer using the meteorological information within the predetermined range from which the estimated road surface condition was obtained , and the residual salt concentration calculated according to the time elapsed since the most recent snow melting agent application and the type of snow melting agent as explanatory variables. 2. The road surface condition prediction method according to claim 1, wherein the road surface condition is predicted at a time in the future from the time when the estimated road surface condition is estimated.

Images