KR20210030047A - Random Forest based Prediction Method and System of Road Surface Condition Using Spatio-Temporal Features - Google Patents

Abstract

The present invention relates to a random forest-based road surface condition prediction method and a system using spatiotemporal characteristics, which includes the steps of: collecting road surface condition data including a road surface condition and coordinate information of a point where the road surface condition was collected; collecting meteorological data including precipitation information and temperature information of a predetermined area; converting the coordinate information included in the road surface condition data into a predetermined index value; learning a road surface condition prediction model using the road surface condition data in which the coordinate information is converted into the predetermined index value, and learning data built using meteorological data; and predicting the road surface condition at the road surface condition prediction point using the road surface condition prediction model. The coordinate information may be converted into an index value corresponding to a grid in which the predetermined area is divided into a predetermined size. The index value may use a Morton code. The present invention provides the method and system for predicting the road surface condition with improved prediction accuracy.

Description

Random Forest based Prediction Method and System of Road Surface Condition Using Spatio-Temporal Features}

The present invention relates to a road surface condition prediction method and system, and more particularly, to a random forest-based road surface condition prediction method and system utilizing spatiotemporal characteristics.

According to statistics provided by the Traffic Accident Analysis System (2018), the number of traffic accidents in 2018 increased by 813 from 2017. Although the number of traffic accidents on sunny days decreased by 3,169 compared to 2017, the number of accidents increased from 2017 when rainfall occurred to 3,526 on rainy days and 320 on snowy days.

Precipitation not only increases the number of traffic accidents, but also affects the severity of traffic accidents. According to a study by Jung et al. (2010), wet road surfaces have lower friction than dry road surfaces. If the road surface is wet and the friction is lowered, the braking distance of the vehicle increases, and the damage can be 2 to 3 times more severe than an accident on a dry road surface (Harold, 2017). Maintenance activities are continuing to reduce the frequency and severity of traffic accidents caused by precipitation, but drivers' awareness of road hazards needs to be raised as well (Jonas et al, 2000). According to Lee et al. (2018), when it rains, the rate of sudden stops for drivers is higher than in other weathers, and this sudden change in driving behavior becomes a factor that can cause traffic accidents. Accordingly, it is necessary to make continuous efforts to thoroughly manage the road, while the driver recognizes the risk of the road and reduces sudden changes in driving behavior.

In order for a driver to recognize an increase in road risk due to precipitation, it is necessary to know the condition of the road according to the weather conditions of the road, such as the road surface being wet due to rain or the wet road surface being frozen due to subzero weather. To this end, the Meteorological Administration has installed 75 CCTV-based road weather information systems nationwide and observes the weather conditions of the roads. Such fixed equipment generates road weather information with high accuracy for an observed point, but it is difficult to predict road weather information at a distant point with high accuracy. In addition, since it provides only information on the state of precipitation such as sunny, rain, and snow, it is difficult to provide the risk of roads due to the condition of the road surface changed by the past precipitation.

Road surface condition prediction research is classified into point prediction that predicts a specific point according to the predicted space range and range prediction that has a wide spatial range. Researches that predict road conditions at specific points mainly use equipment that detects specific signals to analyze the relationship between observed information and road conditions (Shao et al., 1994) and predict road conditions. In order to predict the road surface of a point, direct information on the road surface of the point is required. The information on the road surface of the point is infrared temperature sensor (Mats et al., 2012), piezoelectric sensor (Kang., et al. 2019), RWIS (Road Weather Information System) (Tobias et al., 2018), and smartphone camera (Pan). et al., 2018). The range prediction is to predict the road surface condition of an area that was not predicted spatiotemporally by using the previously measured road surface condition, and it is mainly predicted through weather information such as AWS. Lee (2016) predicted the road surface conditions of the roads around the Jeongneung Tunnel using weather information observed in AWS operated by SK. Li (2010) predicted road conditions using weather information collected from three cities located in the Hubei province of China. Matsuzawa et al. (1996) predicted the frozen road surface using weather information observed in RWIS. In addition, when forecasting on a regional basis, it is necessary to consider the condition of the road or the surrounding environment such as surrounding structures. Yeo et al. (2018) predicted the road surface condition of the inner circuit by using input variables such as road geometry information and traffic information in addition to weather information to extract factors that affect the road surface condition in summer. Lee et al. (2018) also predicted the road surface condition of the inner circuit, but added the characteristics of the time and water capacity of the predicted area by processing weather and road geometry information in addition to the meteorological information. In addition, the range prediction also predicts the road surface condition using weather information forecasted over a wide area. Kangas el al. (2015) predicted the road surface conditions across Finland in 1-hour increments using rainfall information predicted by radar and Road Weather Sensor (RWS).

Accordingly, a technical problem to be solved by the present invention is to provide a method and system for predicting road conditions based on a random forest using spatiotemporal characteristics.

The road surface condition prediction method according to the present invention for solving the above technical problem includes the steps of collecting road surface condition data including coordinate information of a road surface condition and a point at which the road surface condition is collected, precipitation information and temperature of a predetermined area. Collecting weather data including information, converting coordinate information included in the road surface condition data into a predetermined index value, using the road surface condition data and the weather data in which coordinate information is converted to a predetermined index value And learning a road surface condition prediction model using the training data constructed by doing so, and predicting a road surface condition of a road surface condition prediction point using the road surface condition prediction model.

The road surface condition prediction system according to the present invention for solving the above technical problem includes a meteorological data collection unit that collects meteorological data including precipitation information and temperature information of a predetermined area, a point at which the road surface condition and the road surface condition are collected. A road surface condition data collection unit that collects road surface condition data including coordinate information of, a data processing unit that converts coordinate information included in the road surface condition data into a predetermined index value, and the coordinate information is converted to a predetermined index value. A learning unit that learns a road surface condition prediction model using road surface condition data and learning data constructed using the weather data, and a road surface condition predictor that predicts a road surface condition of a road surface condition prediction point using the road surface condition prediction model. Includes.

The coordinate information may be converted into an index value corresponding to a grid obtained by dividing the predetermined area by a predetermined size.

The index value may be a Morton code.

The precipitation information is processed into a value obtained by dividing the cumulative precipitation of the precipitation section obtained by backtracking based on the reference time point by the time difference from the reference time point to the time when precipitation occurs in the precipitation section, and is an explanatory variable of the road surface condition prediction model. Can be used as

The temperature information may be processed into an average, standard deviation, a first quartile, a third quartile, a minimum value, and a maximum value of the temperature values between the reference time point and the time when precipitation occurs, and may be used as an explanatory variable of the road surface condition prediction model.

The road surface condition prediction model may use a random forest-based machine learning model.

According to the present invention, it is possible to provide a road surface condition prediction method and system with improved prediction accuracy by training a random forest-based model by processing weather information and road surface condition information into characteristics including spatio-temporal information.

1 is a block diagram of a road surface condition prediction system according to an embodiment of the present invention.
2 is a diagram provided to explain processing of multiple precipitation information according to an embodiment of the present invention.
3 is a diagram provided to explain a morton code used in an embodiment of the present invention.
4 is a flowchart illustrating an operation of a road surface condition prediction system according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art may easily implement the present invention.

1 is a block diagram of a road surface condition prediction system according to an embodiment of the present invention.

Referring to FIG. 1, a road surface condition prediction system 100 according to a weather condition according to the present invention includes a road surface condition data collection unit 110, a weather data collection unit 120, a database 130, and a data processing unit 140. ), a learning unit 150 and a road surface condition prediction unit 160 may be included.

The road surface condition data collection unit 110 performs a function of collecting road surface condition data including road surface condition information, coordinate information of a road surface condition collection point, and data collection time information. The road surface condition data collection unit 110 may include a road surface sensor 111 and a GNSS receiver 115.

A GNSS signal obtained from the GNSS receiving unit 115 in the road surface condition information acquired through the road surface sensor 111 while the road information collecting vehicle (not shown) equipped with the road surface condition data collecting unit 110 travels on the road The coordinate information (latitude, longitude) obtained from the data can be collected by matching the data collection time.

According to an embodiment, some of the road surface condition data collection unit 110 may be installed on a road surface to collect the road surface condition. The road surface condition information may include a road surface condition type, a road surface temperature, a road surface water film thickness, an outside temperature, a friction value, and the like.

The road surface sensor 111 may obtain a road surface condition type of a road in a non-contact method. The type of road surface condition can be simply classified into Dry, Moist, and Wet, and more specifically, Slush, Ice or Snow with water, Ice and Snow or Frost. (Snow or Hoar Froast) can be added to differentiate. The criteria for classifying the road surface condition type may vary according to embodiments.

According to an embodiment, the road surface sensor 111 may obtain not only a road surface condition type, but also a road surface temperature, a road surface water film thickness, an outside temperature, a friction value, and the like in a non-contact method. A method of obtaining a road surface temperature, a road surface water film thickness, an outdoor temperature, a road surface condition type, a friction value, etc. using the road surface sensor 111 in a non-contact method is already well known, and thus a detailed description thereof will be omitted herein.

The road surface condition data collection unit 110 may store the collected road surface condition data in a memory (not shown) or may transmit the collected road surface condition data to a predetermined destination in real time or at predetermined periods through a communication network (not shown).

The meteorological information collection unit 120 may collect meteorological data including precipitation information and temperature information of a predetermined area from a meteorological server (not shown) operated by a meteorological agency or a private meteorological service provider. Weather information includes Precipitation, Temperature, Humidity, Wind speed, Wind direction, Sunlight, Sunshine, and Sea surface atmospheric pressure. Can include. Precipitation is the amount including rainfall and snowfall.

The meteorological information collection unit 120 may collect meteorological data provided by an automatic weather system (AWS) operated by the Meteorological Agency.

The database 130 may store road surface condition data collected by the road surface condition data collection unit 110 and meteorological data collected by the meteorological information acquisition unit 120. The database 130 may store road surface state data obtained by converting the coordinate information into a predetermined index value by the coordinate information conversion unit 140.

In particular, the database 130 may store learning data constructed in the form of mapping road surface condition data in which coordinate information is converted to a predetermined index value, and weather data processed by weather information corresponding to a point at which the road surface condition data is collected. .

The data processing unit 140 may process the collected road surface condition data and meteorological data in a form suitable for learning and store them as learning data in the database 130.

It will be described that the data processing unit 140 processes the weather information into weather data suitable for learning according to the present invention.

Road surface condition prediction requires consideration of the persistence of weather factors. When the rain stops and time passes, the road surface dries up. If the temperature is high, the road surface dries quickly, and if it is subzero, the road surface freezes. In addition, in summer, it may rain again before the road surface dries, and in winter, it may rain again on frozen roads. If there is a lot of precipitation, it takes a lot of time until the road surface is dry. Conversely, the less precipitation, the faster the road surface dries.

In order to consider the persistence of the meteorological element, it is desirable to process the collected meteorological information into data including time information so that the persistence of the meteorological element can be considered.

2 is a diagram provided to explain processing of multiple precipitation information according to an embodiment of the present invention.

Referring to FIG. 2, in order to process multiple precipitation information, precipitation information measured at an automatic weather station in the past _{is traced back to the observation point (hereinafter referred to as the reference point) (T S} ), and the section where precipitation occurs (P ₁ , P ₂ , …, P _N ). Accumulated precipitation (S ₁ , S ₂ , …, S _N ) of each precipitation section (P ₁ , P ₂ , …, P _N ) from the reference point (T _s ) to the time of precipitation (T ₁ , T ₂ , …, T _N ) is processed as the sum of the values divided by the time difference (△T ₁ , △T ₂ , …, △T _N ) (hereinafter referred to as'poetic distance'), and can be used as learning data.

The processed precipitation information P can be obtained by Equation 1 below.

[Equation 1]

N is the number of precipitation sections found from the reference point (T _{s ), S i} is the cumulative precipitation of the i-th precipitation section from the reference point (T _{s ), and △T i} is the poetic distance from the reference point to the i-th precipitation occurrence to be. In the case of a precipitation section that occurred a long time ago from the reference point, the poetic distance from the reference point to the time point of precipitation in the corresponding precipitation section increases, so the effect is less. As described above, the precipitation information can be used as an explanatory variable that considers the persistence of meteorological factors by processing and using data reflecting the poetic distance.

The temperature information is converted into basic statistical information of the average, standard deviation, 1 quartile, 3 quartile, minimum value, and maximum value of the temperature values between the precipitation start time and the reference time point. The average temperature is used as a variable to explain the degree to which the rate of change of road surface conditions due to temperature is accelerated when the total precipitation and the poetic distance are similar.

The standard deviation further describes the degree to which the temperature change between the two points freezes in the middle, or the rate of change is accelerated. The 1st and 3rd quartiles, minimum and maximum values represent the degree to which a low or high temperature is maintained between two points, and can be used as explanatory variables for the possibility of freezing the road surface or accelerating the evaporation of moisture on the road surface. .

Meanwhile, the data processing unit 140 may convert coordinate information included in the road surface condition data into a predetermined index value.

The condition of the road surface is closely related not only to weather information, but also to the surrounding environment of the observation point. The surrounding buildings and trees form shade, slowing the rate of change of the road surface. Low-lying temperatures are lower than the surroundings due to the sinking of cold air in the highlands. The humidity around the river is high, and the road surface temperature changes faster in the bridge than in other areas.

In addition, in the case of an urbanized expressway where road surface construction is fast and road management is well maintained in areas with heavy traffic, continuous management is performed to prevent freezing of the road surface. The surrounding environment of such a road has a large number of cases, and it is difficult to collect all the information and process it into characteristics. In the present invention, instead of information on the surrounding environment near the point where the road surface condition is detected, information on the surrounding environment may be learned by using coordinate information of the corresponding point.

The reasons for learning surrounding environment information using coordinate information are as follows. First of all, if the coordinate information is used, data having the same weather information but different road conditions can be classified. For example, if the weather information at a specific point in time refers to the same AWS, the value is the same, but the road surface condition may be different depending on the collected location.

This means that the surrounding environment has changed according to the collected location, and the condition of the road surface has also changed. That is, it can be said that the condition of the road surface is changed for the same weather information at the same time according to the coordinate information of the location where the road surface condition is collected. Accordingly, the road surface condition prediction model can learn that the road surface condition changes as the coordinate information of the data having the same weather information at the same time point using this coordinate information is changed. In conclusion, it means that the predictive model learns that the state of the road surface can vary depending on the location in space, which can be said to learn the surrounding environment that changes as the location in space changes.

Coordinate information to be learned in the road surface condition prediction model according to the present invention may use a value indexed by a Morton code.

3 is a diagram provided to explain a morton code used in an embodiment of the present invention.

Referring to FIG. 3, the Morton Code is an algorithm for indexing 2-dimensional coordinate information in one dimension. As shown in FIG. 3, a specific area is divided into a grid of a certain size and indexed in the order of a Z shape from the lower left grid. Define the value. If the coordinate information is included in a specific grid, the coordinate information can be expressed as an index value of the corresponding grid. A predetermined area may be divided into grids having a predetermined size of, for example, 16 meters in width and 8 meters in height, and index values may be defined for each grid in an order according to a predetermined method. The size of dividing a predetermined area into a grid may vary according to embodiments.

Using the Morton code instead of the coordinate information can prevent overfitting. First, the two-dimensional characteristics become one-dimensional while maintaining the explanatory power of the space possessed by the coordinates by using the Morton code. By reducing the dimensions, the characteristics to be considered for branches in the decision tree used in the random forest are reduced, and the resulting overfitting can be avoided. By converting the coordinate information in the form of a point into a range-type index value, the depth of the tree can be reduced to prevent overfitting. If the coordinate information is used as it is, more precise prediction is possible, but the more sophisticated the more learning data is needed to learn all regions. If enough data is not learned, overfitting can occur.

The learning unit 150 may learn a road surface condition prediction model that predicts a road surface condition according to a weather condition by using the training data built in the database 130. The learning unit 140 uses machine learning models such as logistic regression, decision tree, SVM (Support Vector Machine), k-NN (k-Nearest Neighbor), and deep learning. Thus, a road surface condition prediction model can be trained. In particular, according to an embodiment, it is preferable that the learning unit 150 use a logic-based machine learning model using a decision tree, such as a random forest.

A logic-based machine learning model that uses decision trees, such as a random forest, is best suited for conditional prediction. The decision tree calculates the most optimal condition for each node of the tree to have one label of data belonging to an end node. Each internal node branches data according to a specific condition and delivers it to each child node. Also, the parent node and the child node can branch under different conditions. When data to be predicted is input to the highest node, the data is branched by various conditions at the inner node until reaching the end node. The conditions at the internal node can be either a range of index values or a specific weather condition. Therefore, the decision tree-based machine learning model can be said to be the most suitable model for learning and predicting characteristics of weather, time, and space. The random forest can train such decision trees as many as a specified number, and thus can consider the number of various cases. The top node of each decision tree in the random forest branches through different conditions to generate trees with different conditions.

The road surface condition prediction unit 160 may predict the road surface condition of the road surface condition prediction point by using the road surface condition prediction model learned by the learning unit 150. For example, by converting the coordinate information corresponding to the current position of the vehicle traveling on the road into a Morton code, and inputting the weather data of the corresponding position into the road surface condition prediction model, the road surface condition prediction result of the current vehicle driving position is output. can do. Of course, it is also possible to output the road surface condition prediction result of each point on the route the vehicle is traveling to the destination.

4 is a flowchart illustrating an operation of a road surface condition prediction system according to an embodiment of the present invention.

Referring to FIGS. 1 to 4, road surface condition data including coordinate information of a road surface condition and a point at which the road surface condition is collected may be collected using the road surface condition data collecting unit 110 (S410).

Weather data including precipitation information and temperature information of a predetermined area may be collected through the meteorological data collection unit 120 (S420).

The data processing unit 140 may convert coordinate information included in the road surface condition data collected in step S410 into a predetermined index value (S430). In step S430, the index value may use a Morton Code.

In addition, the data processing unit 140 may process the meteorological data collected in step S420 by a predetermined method (S440). In step S440, the precipitation information may be processed as a sum of values obtained by dividing the cumulative precipitation of each precipitation section found backward from the reference point by a poetic distance. The temperature information may be processed into basic statistical information of the average, standard deviation, first quartile, third quartile, minimum value, and maximum value of temperature values between the time when precipitation occurs and the reference time point.

Next, the learning unit 150 may learn a road surface condition prediction model by using road surface condition data in which coordinate information is converted into a predetermined index value and learning data constructed using weather data processed by a predetermined method ( S450).

Steps S410 to S450 may be continuously performed in a manner of updating the road surface condition prediction model by reflecting the additional learning data generated.

Thereafter, the road surface condition prediction unit 160 inputs the index value corresponding to the coordinate information of the road surface condition prediction point and the weather data obtained by processing the weather information obtained for the point into the road surface condition prediction model learned in step S450. The road surface condition of the road surface condition prediction point may be predicted (S460).

The embodiments described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices, methods, and components described in the embodiments are, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate (FPGA). array), programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. Further, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to operate as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or, to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

As described above, although the embodiments have been described by the limited drawings, a person of ordinary skill in the art can apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order from the described method, and/or components such as systems, structures, devices, circuits, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

110: road surface condition data collection unit
120: weather data collection unit
130: database
140: data processing unit
150: Learning Department
160: road surface condition prediction unit

Claims (10)

Collecting road surface condition data including coordinate information of a road surface condition and a point at which the road surface condition was collected,
Collecting meteorological data including precipitation information and temperature information of a predetermined area,
Converting coordinate information included in the road surface condition data into a predetermined index value,
Learning a road surface condition prediction model using the road surface condition data in which coordinate information is converted into a predetermined index value and learning data constructed using the weather data, and
Predicting a road surface condition of a road surface condition prediction point using the road surface condition prediction model
Including,
A road surface condition prediction method for converting the coordinate information into an index value corresponding to a grid obtained by dividing the predetermined area by a predetermined size. In claim 1,
The index value is a road surface condition prediction method using a Morton code. In claim 1,
The precipitation information,
Road surface processed as a value obtained by dividing the cumulative precipitation of the precipitation section obtained by backtracking based on the reference time point by the time difference from the reference time point to the time when precipitation occurs in the precipitation section, and used as an explanatory variable of the road surface condition prediction model State prediction method. In claim 3,
The temperature information,
A road surface condition prediction method that is processed into the average, standard deviation, 1 quartile, 3 quartile, minimum value, and maximum value of the temperature values between the reference time point and the time when precipitation occurs, and used as explanatory variables of the road surface condition prediction model. In claim 2 or 4,
The road surface condition prediction model is a road surface condition prediction method using a random forest-based machine learning model. A meteorological data collection unit that collects meteorological data including precipitation information and temperature information of a predetermined area,
A road surface condition data collection unit that collects road surface condition data including coordinate information of a road surface condition and a point at which the road surface condition is collected,
A data processing unit that converts coordinate information included in the road surface condition data into a predetermined index value,
A learning unit for learning a road surface condition prediction model using the road surface condition data in which coordinate information is converted into a predetermined index value and learning data constructed using the meteorological data, and
A road surface condition prediction unit that predicts a road surface condition of a road surface condition prediction point using the road surface condition prediction model
Including,
A road surface condition prediction system for converting the coordinate information into an index value corresponding to a grid obtained by dividing the predetermined area by a predetermined size. In claim 6,
The index value is a road surface condition prediction system using Morton Code. In claim 6,
The precipitation information,
Road surface processed as a value obtained by dividing the cumulative precipitation of the precipitation section obtained by backtracking based on the reference time point by the time difference from the reference time point to the time when precipitation occurs in the precipitation section, and used as an explanatory variable of the road surface condition prediction model Condition prediction system. In claim 10,
The temperature information,
A road surface condition prediction system that is processed into the average, standard deviation, 1 quartile, 3 quartile, minimum value, and maximum value of the temperature values between the reference point and the time when precipitation occurs, and used as explanatory variables of the road surface condition prediction model. In claim 7 or 9,
The road surface condition prediction model is a road surface condition prediction system using a random forest-based machine learning model.

Images