JP7400837B2 - Energy consumption estimation program, energy consumption estimation method, and energy consumption estimation device - Google Patents

Description

The present invention relates to an energy consumption estimation program, an energy consumption estimation method, and an energy consumption estimation device.

Since an electric vehicle has a shorter cruising distance than a gasoline vehicle, there is a need for a technology that presents the cruising distance to a user by estimating power consumption (for example, Patent Document 1, Patent Document 2, etc.).

Japanese Patent Application Publication No. 2014-182035 Japanese Patent Application Publication No. 2013-50367

However, conventional power consumption estimation techniques are assumed to be applied to routes that have a history of travel, and are considered to be less effective for routes that have no history of travel.

Note that this situation is considered to be the same not only for electric vehicles but also for moving bodies that use energy other than electricity, such as gasoline.

Therefore, one aspect of the present invention aims to make it possible to estimate the energy consumption of a route with no travel history.

In one proposal, for each of a plurality of pieces of trip data including a travel start point, a travel end point, travel distance, and energy consumption, a travel route related to the trip data is determined by referring to map data that represents roads as a set of links. A set of links indicating a road is identified , and characteristic information corresponding to each link included in the identified set of links is extracted by referring to a storage unit that stores road characteristic information, and characteristic information is extracted for each link. Based on the characteristic information and the travel distance and energy consumption included in the plurality of trip data, a regression model is created with the energy consumption rate, which is energy consumption per unit travel distance, as an objective variable and the characteristic information as an explanatory variable. and calculates the energy consumption for each link included in the search range of the travel route between the specified departure point and the specified destination based on the characteristic information of the link and the regression model, A computer is caused to perform a process of estimating an optimal route between the starting point and the destination, using the energy consumption calculated for each link included in the search range as a weight for each link .

According to one aspect, it is possible to estimate the energy consumption of a route with no travel history.

1 is a diagram showing a configuration example of a power consumption estimation system 1 according to an embodiment of the present invention. 1 is a diagram showing an example of a hardware configuration of a power consumption estimating device 10 according to an embodiment of the present invention. 1 is a diagram showing an example of a functional configuration of a power consumption estimating device 10 according to an embodiment of the present invention. 3 is a diagram showing an example of the configuration of a trip data storage section 121. FIG. 12 is a flowchart for explaining an example of a processing procedure that the power consumption estimation device 10 executes when learning the power consumption rate estimation model m1. FIG. 3 is a diagram for explaining an overview of movement distance estimation processing. It is a figure which shows an example of the identification result of an estimated route. FIG. 6 is a diagram for explaining a learning process of a first configuration example of the power consumption rate estimation model m1. It is a figure which shows the example of a representation of an estimated route. FIG. 7 is a diagram for explaining a learning process of a second configuration example of the power consumption rate estimation model m1. It is a flowchart for explaining an example of the processing procedure which power consumption estimating device 10 performs when estimating power consumption when electric vehicle 20 moves on a certain movement route. FIG. 6 is a diagram for explaining a process for estimating power consumption of a planned route when using the power consumption rate estimation model m1 of the first configuration example. FIG. 7 is a diagram for explaining a process for estimating power consumption of a planned route when using a power consumption rate estimation model m1 of a second configuration example. It is a flowchart for explaining an example of a processing procedure when estimating an optimal route from the viewpoint of electricity consumption. FIG. 3 is a diagram showing map data for explaining an example of a processing procedure when estimating an optimal route from the viewpoint of electricity consumption.

Embodiments of the present invention will be described below based on the drawings. FIG. 1 is a diagram showing a configuration example of a power consumption estimation system 1 according to an embodiment of the present invention. In the power consumption estimation system 1 shown in FIG. 1, the power consumption estimation device 10 communicates with an information processing device (on-vehicle device) mounted on each of a plurality of electric vehicles 20 via a network including a mobile communication network, etc. Connected.

The power consumption estimating device 10 learns a regression model for estimating the power consumption rate based on the performance data for each trip of each electric vehicle 20, and uses the regression model to calculate the newly specified travel route. One or more computers that estimate power consumption. Note that a trip is a section from when the power of the electric vehicle 20 is turned on and movement begins until when the power is turned off and movement ends. In the case of a gasoline vehicle, a trip is the period from when the ignition is turned on to when it is turned off.

FIG. 2 is a diagram showing an example of the hardware configuration of the power consumption estimating device 10 according to the embodiment of the present invention. The power consumption estimating device 10 in FIG. 2 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a CPU 104, an interface device 105, etc., which are connected to each other via a bus B.

A program that implements processing in the power consumption estimating device 10 is provided by a recording medium 101. When the recording medium 101 on which the program is recorded is set in the drive device 100, the program is installed from the recording medium 101 to the auxiliary storage device 102 via the drive device 100. However, the program does not necessarily need to be installed from the recording medium 101, and may be downloaded from another computer via a network. The auxiliary storage device 102 stores installed programs as well as necessary files, data, and the like.

The memory device 103 reads the program from the auxiliary storage device 102 and stores it therein when there is an instruction to start the program. The CPU 104 executes functions related to the power consumption estimating device 10 according to programs stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

Note that an example of the recording medium 101 is a portable recording medium such as a CD-ROM, a DVD disk, or a USB memory. Furthermore, examples of the auxiliary storage device 102 include an HDD (Hard Disk Drive), a flash memory, and the like. Both the recording medium 101 and the auxiliary storage device 102 correspond to computer-readable recording media.

FIG. 3 is a diagram showing an example of the functional configuration of the power consumption estimating device 10 according to the embodiment of the present invention. In FIG. 3, the power consumption estimation device 10 includes a route estimation section 11, a variable value extraction section 12, a model learning section 13, a power consumption estimation section 14, and the like. Each of these units is realized by one or more programs installed in the power consumption estimating device 10 causing the CPU 104 to execute the process. The power consumption estimating device 10 also utilizes storage units such as a trip data storage unit 121, a map data storage unit 122, and a weather log data storage unit 123. Each of these storage units can be realized using, for example, the auxiliary storage device 102 or a storage device connectable to the power consumption estimating device 10 via a network.

The trip data storage unit 121 stores data for each past trip of each electric vehicle 20, including performance information regarding the trip (hereinafter referred to as "trip data").

FIG. 4 is a diagram showing an example of the configuration of the trip data storage section 121. In FIG. 4, one row corresponds to one trip data. Each trip data includes values such as vehicle ID, movement start point, movement end point, movement distance, average speed, movement start date and time, movement end date and time, power consumption, and acceleration/deceleration frequency.

The vehicle ID is identification information of each electric vehicle 20. The movement start point is position information of the position where movement (trip) is started. The movement end location is position information of the position where the movement (trip) ended. These location information may be accurate latitude and longitude, or from the perspective of personal information protection, information indicating the area corresponding to one of the meshes when the map is divided into meshes (for example, 100m squares). You can.

The travel distance is the total travel distance [km] in a trip. The average speed is the average speed [km/h] in a trip. The movement start date and time is the start date and time of movement (trip). The travel end date and time is the end date and time of the travel (trip). Power consumption is power consumption [kW] in a trip. The acceleration/deceleration frequency is the ratio of the total duration of the state in which the absolute value of the ± acceleration is equal to or greater than the threshold value, with respect to the period from the movement start date and time to the movement end date and time. Note that the acceleration/deceleration frequency is an example of a parameter indicating the tendency (habit) of the driver's driving style. A different parameter with the same meaning may be used instead of the acceleration/deceleration frequency. Further, among the parameters not shown in FIG. 4, such as the usage rate of air conditioning and the loading amount of luggage, it is possible to measure them on the electric vehicle 20 side, and the power consumption rate of the electric vehicle 20 (per unit mileage) Parameters that affect (correlate with) power consumption) [Wh/km] may be included in the trip data.

In this way, trip data is data that summarizes one trip, and does not include instantaneous position information, power consumption, etc. in time series. For example, the trip data may be transmitted from each electric vehicle 20 to the power consumption estimating device 10 at a predetermined timing. Alternatively, trip data may be acquired from each electric vehicle 20 at a dealer, and the trip data may be transmitted from a terminal at the dealer to the power consumption estimating device 10.

The route estimation unit 11 estimates one travel route for each trip data stored in the trip data storage unit 121. That is, the trip data does not include information indicating the travel route of the trip. Therefore, the route estimating unit 11 calculates the trip associated with the trip data based on the travel start point, travel end location, and travel distance included in the trip data and the map data stored in the map data storage unit 122. Estimate the travel route at . Map data is data that represents a road by a set of links and nodes, and data that includes information regarding various facilities (POI (Point Of Interest), etc.). The travel route estimation result (hereinafter referred to as "estimated route") by the route estimation unit 11 is also expressed by a set of links (a shape in which a plurality of links are connected into one), similarly to map data.

The variable value extraction unit 12 learns the power consumption rate estimation model m1 from the trip data, the map data storage unit 122, and the weather log data storage unit 123 for each estimated route (that is, for each trip data in which the travel route is estimated). Extract the data. In this embodiment, the power consumption rate estimation model m1 is a regression model that uses the power consumption rate as an objective variable and uses link (road) characteristic information, driving tendency, weather conditions, etc. as explanatory variables. Therefore, the respective actual values of the target variable and the explanatory variable are extracted as learning data. Link characteristic information (slope, etc.) can be obtained from the map data storage unit 122. Weather conditions can be obtained from the weather log data storage unit 123. That is, the weather log data storage unit 123 stores log data of weather conditions in various places (hereinafter referred to as "weather data").

The model learning unit 13 learns the power consumption rate estimation model m1 based on the group of actual values of the explanatory variables and objective variables of the power consumption rate estimation model m1 extracted by the variable value extraction unit 12.

The power consumption estimating unit 14 estimates the power consumption when traveling a specified travel route (for example, the route that the user plans to travel) using the learned power consumption rate estimation model m1.

The processing procedure executed by the power consumption estimating device 10 will be described below. FIG. 5 is a flowchart for explaining an example of a processing procedure that the power consumption estimation device 10 executes when learning the power consumption rate estimation model m1.

In step S110, the route estimation unit 11 performs a movement route estimation process for each trip data stored in the trip data storage unit 121.

FIG. 6 is a diagram for explaining an overview of the movement distance estimation process. The trip data to be processed in FIG. 6 is referred to as "target trip data."

As shown in FIG. 6, the route estimation unit 11 first performs a route search with reference to target trip data and map data (S111). Specifically, the route estimation unit 11 searches for a route between the travel start point and the travel end point of the target trip data with reference to the map data. Here, multiple routes can be searched. Next, the route estimation unit 11 compares the travel distance estimated for each searched route with the travel distance included in the target trip data, and identifies one estimated route based on the error rate of the travel distance. do.

Regarding the movement distance estimation process, two examples, a first method and a second method, will be described in more detail. However, the moving route may be estimated using other methods. Step numbers in the following description correspond to step numbers in FIG.

[First method of estimating travel distance]
In step S111, the route estimation unit 11 refers to the map data, searches for a predetermined number of routes between the travel start point and the travel end point of the target trip data in descending order of travel time, and calculates the travel distance of each route. Compute an estimate of . The estimated value of the travel distance of the route can be calculated, for example, by calculating the sum of the distances of the links that make up the route.

In step S112, the route estimating unit 11 selects a route that has the highest error rate from the travel distance of the target trip data among the estimated travel distances calculated for each of the searched routes. Specify as the estimated route. Here, the error rate is a value obtained by the following equation, where the travel distance of the target trip data is _D1 , and the estimated value of the travel distance of the searched route is _D2 .
Error rate = |D ₂ -D ₁ |÷D ₁
A high error rate means that the value of the error rate is small. Note that when there are multiple estimated values with the highest error rate, one estimated value may be selected by any method, and the route related to the estimated value may be specified as the estimated route.

[Second method of estimating travel distance]
In step 111, the route estimation unit 11 searches for the shortest route between the travel start point and the travel end point of the target trip data by referring to the map data, and calculates an estimated value of the travel distance of the shortest route. For example, a known method such as Dijkstra's method may be used to search for the shortest route.

In step S112, if the error rate between the estimated travel distance of the shortest route and the travel distance of the target trip data is less than the threshold, the route estimation unit 11 replaces the shortest route searched in step S111 with the estimated route of the target trip data. shall be. Note that when the error rate is equal to or greater than the threshold value, it is determined that there is no estimated route for the target trip data. Therefore, in this case, the target trip data is not used in subsequent processing.

FIG. 7 is a diagram illustrating an example of an estimated route identification result. FIG. 7 shows that an estimated route is specified for each trip data. However, if the second method is adopted, there may be trip data in which the estimated route is not specified. Note that in FIG. 7, each estimated route has a similar shape for convenience, but in reality, each estimated route has a different shape.

Return to FIG. 5. Following step S110, the variable value extraction unit 12 extracts the learning data of the power consumption rate estimation model m1 (each explanatory variable and objective variable of the power consumption rate estimation model m1) for each estimated route specified in step S110. value (actual value)) is executed (S120).

The objective variable of the power consumption rate estimation model m1 is the power consumption rate [Wh/km]. The power consumption rate is obtained by dividing the power consumption in the trip data by the travel distance.

The explanatory variables of the power consumption rate estimation model m1 include variables that depend on roads (links in map data) (hereinafter referred to as "link characteristic variables") and variables that do not depend on roads (hereinafter referred to as "link independent"). There are variables. The value of the link feature variable can also be said to be link feature information (or feature amount). On the other hand, the value of the variable that does not depend on the link is information that does not depend on the characteristics of the road, and can be said to be information that affects (has a correlation with) energy consumption (energy consumption rate). Examples of each are given below.

[Example of link characteristic variables]
・Gradient of the road corresponding to the link ・Type of road corresponding to the link (expressway, national highway, other, etc.)
- Width of the road corresponding to the link - Number of traffic lights around the road corresponding to the link - Number of stores around the road corresponding to the link Note that in this embodiment, the position information of the link is not used as a characteristic variable of the link. This is to prevent the power consumption rate estimation model m1 from depending on link position information. Since the power consumption rate estimation model m1 does not depend on link position information, the power consumption rate estimation model m1 can estimate the power consumption rate even on a road (link) on which there is no past travel history. I can do it. For example, by further adding information including link position information to the process described in this embodiment, it is possible to estimate the power consumption rate taking movement history into consideration.

[An example of a variable that does not depend on links]
- Distance traveled on the estimated route, frequency of acceleration/deceleration, air conditioning usage rate, load capacity of luggage, temperature, wind direction, wind speed, weather Values of characteristic variables of the link are retrieved from the map data storage unit 122 based on the position information of the link. Can be extracted (obtained). That is, for each estimated route, the variable value extraction unit 12 acquires the value of the characteristic variable of each link constituting the estimated route from the map data based on the position information of the link. Regarding the number of surrounding traffic lights and the number of surrounding stores, for example, the map data may be divided into grids, and the number of traffic lights or stores in the grid including the target link may be counted.

On the other hand, the variable value extraction unit 12 extracts (acquires) values of variables that do not depend on links from trip data, weather data, etc. for each estimated route. For example, for each estimated route, the variable value extraction unit 12 extracts (obtains) the travel start date and time of the trip data corresponding to the estimated route, and weather conditions such as the temperature, wind direction/speed, and weather at the travel start location from the weather data. Then, the extracted value is assigned to each link that makes up the estimated route. In addition, the variable value extraction unit 12 extracts (obtains) travel distance, acceleration/deceleration frequency, air conditioning usage rate, cargo loading amount, etc. for each estimated route from the trip data corresponding to the estimated route. A value is assigned to each link that constitutes the estimated route. That is, the same value is assigned to each link belonging to the same estimated route as the value of a variable that does not depend on links. It is also conceivable that there is trip data for long-distance travel. In this case, it is conceivable that weather conditions etc. may change during the movement. Therefore, for explanatory variables whose values are extracted from weather data, the values may be extracted for each link. In this case, the date and time when the electric vehicle 20 passes through each link in the estimated route can be determined by dividing the elapsed time from the travel start date and time to the travel end date and time of the trip data corresponding to the estimated route based on the distance of each link. May be requested. Weather conditions and the like based on weather data corresponding to the date and time obtained in this way and the position information of the link may be added to the link.

Following step S120, the model learning unit 13 performs a learning process for the power consumption rate estimation model m1 using the explanatory variable values and objective variable values extracted for each link of each estimated route by the variable value extraction unit 12. (S130).

In this embodiment, two configuration examples, a first configuration example and a second configuration example, are shown for the power consumption rate estimation model m1. Since the learning process of the power consumption rate estimation model m1 differs depending on the configuration example, it will be explained below for each configuration example.

[Learning process of first configuration example of power consumption rate estimation model m1]
FIG. 8 is a diagram for explaining the learning process of the first configuration example of the power consumption rate estimation model m1. In FIG. 8, the power consumption rate estimation model m1 includes a link-independent model m21, a link-dependent model m22, and a composite model m23.

The link-independent model m21 is a model that does not depend on the characteristics of links (roads). That is, the link-independent model m21 is a regression model in which variables that do not depend on the link but affect (correlate with) the power consumption rate are used as explanatory variables, and the power consumption rate is used as the objective variable. The link-dependent model m22 is a model that depends on the characteristics of links (roads). That is, the link dependent model m22 is a regression model that uses link characteristic variables as explanatory variables and uses the power consumption rate as an objective variable. The composite model m23 calculates the power consumption rate as the final output of the power consumption rate estimation model m1 by giving weights to the estimated value of the link-dependent model m22 and the estimated value of the link-independent model m21, for example. This is a model that For the optimization of each weight, for example, a known technique such as grid search can be used.

FIG. 8 shows that the values of link-independent variables extracted for the estimated route corresponding to certain trip data and assigned to each link of the estimated route are input to the link-independent model m21. There is. Further, it is shown that the value of the link characteristic variable extracted and assigned to each link of the estimated route is input to the link dependence model m22. Further, it is shown that the outputs (estimated values) from each of the link-independent model m21 and the link-dependent model m22 are input to the composite model m23. Furthermore, it is shown that the power consumption rate estimation model m1 is trained by comparing the output (estimated value) from the synthetic model m23 with the actual value of the power consumption rate derived from the trip data. There is.

When the power consumption rate estimation model m1 includes three models as shown in FIG. 8, two learning methods, a first learning method and a second learning method, are given as examples of learning methods for the three models. The first learning method is a method of learning a link-independent model m21 and a link-dependent model m22, and then learning a composite model m23 by using a trained link-independent model m21 and a trained link-dependent model m22. . The second learning method is a method of learning three models simultaneously.

First, the first learning method will be explained. In the first learning method, learning of the link-independent model m21 and the link-dependent model m22 can be performed independently (in parallel).

Regarding the link-independent model m21, the model learning unit 13 performs learning by executing the following steps (1) and (2).

(1) The objective variable for each estimated route is as follows (that is, the power consumption rate [Wh/km] for one entire trip).

但し、Ｐ_ｉは、推定経路を図９に示されるように表現した場合における、リンクｉの消費電力である。また、Ｄ_ｉは、リンクｉの距離である。ｍは、推定経路に含まれるリンクの総数である。なお、

However, P _i is the power consumption of link i when the estimated route is expressed as shown in FIG. Further, D _i is the distance of link i. m is the total number of links included in the estimated route. In addition,

は、推定経路に対応するトリップデータ（図４）の消費電力である。

is the power consumption of the trip data (FIG. 4) corresponding to the estimated route.

(2) The following data set is obtained as learning data using the values of variables independent of links extracted for each estimated route in step S120 and the objective variable in (1). Therefore, the model learning unit 13 uses the learning data to learn the link-independent model m21 by, for example, multiple regression or nonlinear multivariate regression.
Learning data for estimated path 1: value of objective variable, value of explanatory variable 1...value of explanatory variable k Learning data for estimated path 2: value of objective variable, value of explanatory variable 1...value of explanatory variable k・
・
・
Learning data for estimated route n: value of objective variable, value of explanatory variable 1...value of explanatory variable k Note that explanatory variable k is the k-th variable that does not depend on the link assigned to link m. It is a variable.

On the other hand, the link dependence model m22 is expressed by the following function F (eg, neural network).

但し、α_ｉは、リンクｉの重み、各Ｘ_ｉは、リンクｉに付与された、リンクの特徴変数である。ｊは、リンクの特徴変数の総数である。また、重みαｉは、リンクｉの電力消費率（Ｐ_ｉ÷Ｄ_ｉ）に相当する。

However, α _i is the weight of link i, and each X _i is a link characteristic variable given to link i. j is the total number of link feature variables. Further, the weight αi corresponds to the power consumption rate (P _i ÷D _i ) of the link i.

In this case, learning is performed by the model learning unit 13 executing the following procedure (1).

(1) The model learning unit 13 optimizes the learning parameters of the function F (learning parameters of the neural network) so that the following equation is satisfied.

関数Ｆの学習パラメータの値は、例えば、以下の絶対値を損失とした誤差逆伝播法等により最適化することができる。

The value of the learning parameter of the function F can be optimized by, for example, an error backpropagation method using the following absolute value as a loss.

関数Ｆの学習パラメータが最適化されることにより、任意のリンクｉのＤ_ｉ及び各変数Ｘ_ｉを関数Ｆに入力することで、リンクｉの電力消費率を回帰することができる。すなわち、関数Ｆが、リンク依存モデルｍ２２に相当する。但し、図８に示したリンク依存モデルｍ２２の１つの推定経路あたりの出力は、以下の通りである。

By optimizing the learning parameters of the function F, the power consumption rate of the link i can be regressed by inputting D _i and each variable X _i of an arbitrary link i to the function F. That is, the function F corresponds to the link dependence model m22. However, the output per estimated route of the link-dependent model m22 shown in FIG. 8 is as follows.

但し、ｍは、推定経路に含まれるリンクの総数である。

However, m is the total number of links included in the estimated route.

When the learning of the link-independent model m21 and the link-dependent model m22 is completed, the model learning unit 13 performs learning of the composite model m23. Specifically, the model learning unit 13 executes the following steps (1) to (3) for each estimated route.

(1) The model learning unit 13 inputs the values of variables that do not depend on links into the link-independent model m21, inputs the values of the feature variables of each link of the estimated route to be processed into the link-dependent model m22, and Obtain an estimate of the power consumption rate from the model.

(2) The model learning unit 13 inputs the two estimated values to the composite model m23, and obtains a value (estimated value of power consumption rate) output from the composite model m23.

(3) The model learning unit 13 determines the parameters of the synthetic model m23 based on the difference between the estimated value output from the synthetic model m23 and the actual value of the power consumption rate based on the trip data corresponding to the estimated route to be processed. (For example, the weights for each of the link-independent model m21 and the link-dependent model m22) are updated.

When the difference converges, the learning process of the composite model m23 ends.

Next, the second learning method will be explained. In the second learning method, only the estimated value output from the composite model m23 is compared with the actual value of the power consumption rate. The model learning unit 13 optimizes each model by calculating back the errors of the link-independent model m21 and the link-dependent model m22 from the error of the estimated value of the composite model m23 with respect to the actual value of the power consumption rate.

For example, a description will be given assuming that the composite model m23 is a model that outputs a weighted average of the link-independent model m21 and the link-dependent model m22, and each parameter is as follows.
Output of link independent model m21: a
Output of link dependency model m22: b
Output of composite model m23: c
Weight for link independent model m21: w1
Weight for link dependent model m22: w2
In this case, the output c of the composite model m23 is as follows.
c=a×w1+b×w2
Here, assuming that the actual value of the power consumption rate is x, the error Ec of the composite model m23 is as follows.
Ec=c−x
In the second learning method, it is assumed that the error Ec is the sum of the output error Ea of the link-independent model m21 and the error Eb of the link-dependent model m22 in the composite model m23.

Therefore, the model learning unit 13 executes the following calculations.
Ea=w1/(w1+w2)*Ec
Eb=w2/(w1+w2)*Ec
Then, the error in the link-independent model m21 and the error in the model link-dependent model m22 can be calculated from the error and weight in the composite model m23.

The model learning unit 13 repeatedly optimizes the link-independent model m21 and the link-dependent model m22 while updating the weights w1 and w2, and finds the combination of w1 and w2 that gives the smallest Ec. , the link-dependent model m22 and the composite model m23 can be optimized (learned).

[Learning process of second configuration example of power consumption rate estimation model m1]
FIG. 10 is a diagram for explaining the learning process of the second configuration example of the power consumption rate estimation model m1. In FIG. 10, the power consumption rate estimation model m1 is one model that is not divided into a plurality of models as in FIG. The power consumption rate estimation model m1 in FIG. 10 is a regression model (e.g., neural network, decision tree, multiple regression model, etc.) that uses link-independent variables and link characteristic variables as explanatory variables and uses the power consumption rate as an objective variable. It is.

In this case, the model learning unit 13 calculates, for each estimated route, the values of the variables that do not depend on the links extracted for the estimated route and the values of the characteristic variables of the links extracted for the estimated route, into the power consumption rate estimation model. Enter in m1. The model learning unit 13 compares the estimated value of the power consumption rate output from the power consumption rate estimation model m1 with the actual value of the power consumption rate that can be obtained from the trip model corresponding to the estimated route, and Based on this, the learning parameters of the power consumption rate estimation model m1 are updated. The model learning unit 13 updates the learning parameters until the comparison results converge. As a result, when inputting the values of variables that do not depend on the links of any route and the values of the characteristic variables of each link that makes up the route, the power consumption rate estimation model m1 that outputs the estimated value of the power consumption rate is trained. be done.

Next, an example of a processing procedure executed using the trained power consumption rate estimation model m1 will be described. FIG. 11 is a flowchart for explaining an example of a processing procedure executed by the power consumption estimating device 10 when estimating the power consumption when the electric vehicle 20 moves along a certain travel route.

In step S210, the power consumption estimating unit 14 receives input of positional information (for example, latitude and longitude, etc.) of the departure point (scheduled starting location) and destination (scheduled moving ending location). The starting point and destination may be specified, for example, by a user of a user terminal connected to the power consumption estimating device 10 via a network.

Subsequently, the power consumption estimating unit 14 determines a travel route from the departure point to the destination (hereinafter referred to as "planned route") (S220). For example, the power consumption estimating unit 14 searches for multiple routes from the departure point to the destination by referring to the map data storage unit 122, and the route selected by the user from among the multiple routes is It may also be determined as a scheduled route. Alternatively, the shortest route from the departure point to the destination may be determined as the scheduled route. Note that the planned route is expressed by a set of links and nodes.

Note that although an example has been shown in which the power consumption estimating unit 14 searches for a planned route, a moving route searched by another computer may be input to the power consumption estimating unit 14 as the planned route.

Subsequently, the power consumption estimation unit 14 acquires the values of the explanatory variables of the power consumption rate estimation model m1 regarding the planned route (S230). Specifically, the power consumption estimation unit 14 acquires the value of the link characteristic variable for each link included in the planned route. The power consumption estimating unit 14 also obtains values of variables independent of links for the planned route. The current date and time may be applied to variables such as weather data that require the date and time to be specified for obtaining values. Alternatively, the scheduled date and time of traveling along the scheduled route may be input to the power consumption estimation unit 14 together with the scheduled route. Basically, the value of each explanatory variable may be acquired in the same manner as the extraction of each variable by the variable value extraction unit 12. However, if the scheduled date and time of movement is in the future, weather forecast data corresponding to the future date and time may be used as the weather data.

Next, the power consumption estimating unit 14 calculates the estimated value of the power consumption [kW] of the planned route by inputting the values of the obtained explanatory variables into the trained power consumption rate estimation model m1 (S240). .

Regarding step S240, two examples will be described: a case where the power consumption rate estimation model m1 is the first configuration example (FIG. 8) and the second configuration example (FIG. 10).

FIG. 12 is a diagram for explaining the process of estimating the power consumption of the planned route when the power consumption rate estimation model m1 of the first configuration example is used.

When the power consumption estimation unit 14 inputs the value of the link-independent variable to the learned link-independent model m21, the link-independent model m21 outputs one estimated value of the power consumption rate for the planned route ( S241a).

Further, when the power consumption estimating unit 14 inputs the value of the feature variable acquired for each link of the planned route (link feature amount) to the learned link dependent model m22, the link dependent model m22 , an estimated value of the power consumption rate for each link of the planned route is output (S242a). In addition, in FIG. 12, the link-dependent model m22 is illustrated for each link, but this does not indicate that the link-independent model m21 exists for each link. Indicates that the calculation is performed by This point also applies to the composite model m23.

Following steps S241a and S242a, the power consumption estimating unit 14 outputs, for each link of the planned route, the estimated value output from the link-independent model m21 for the planned route, and the estimated value output from the link-dependent model m22 for the link. When the estimated value is input to the trained synthetic model m23, the synthetic model m23 generates an estimated value of the power consumption rate for each link of the planned route, which is obtained by combining the two estimated values. is output (S243a). Note that the same value is applied to the output from the link-independent model m21 for all links.

Subsequently, the power consumption estimating unit 14 calculates the power consumption [kW ] is calculated (S244a). Note that the distance of each link can be obtained from map data.

Subsequently, the power consumption estimation unit 14 calculates the estimated value of the power consumption of the planned route by calculating the sum of the estimated values of the power consumption calculated for each link (S245a).

On the other hand, FIG. 13 is a diagram for explaining the process of estimating the power consumption of the planned route when the power consumption rate estimation model m1 of the second configuration example is used.

When the power consumption estimating unit 14 inputs the values of variables that do not depend on links and the values of characteristic variables of the link for each link of the planned route into the learned power consumption rate estimation model m1, the power consumption rate estimation model m1 outputs the estimated value of the power consumption rate for each link (S241b).

Next, the power consumption estimation unit 14 calculates the product of the estimated value output from the power consumption rate estimation model m1 for each link of the planned route and the distance of the link, thereby estimating the consumption of the link. Calculate the estimated value of electric power [kW] (S242b)
Subsequently, the power consumption estimation unit 14 calculates the estimated value of the power consumption of the planned route by calculating the sum of the estimated values of the power consumption calculated for each link (S243b).

Next, a processing procedure for estimating an optimal route from the viewpoint of electricity consumption of the electric vehicle 20 using the learned power consumption rate estimation model m1 will be described. FIG. 14 is a flowchart for explaining an example of a processing procedure when estimating an optimal route from the viewpoint of electricity consumption. Further, FIG. 15 is a diagram showing map data for explaining an example of a processing procedure when estimating an optimal route from the viewpoint of electricity cost.

In step S310, the power consumption estimating unit 14 acquires all map data within the range designated as the travel route search range from the map data storage unit 122. As a result, information as shown in (1) of FIG. 15 is obtained. Note that the search range may be determined to cover at least the starting point and the destination, or may be set to cover the entire country of Japan, for example, if computational resources permit.

Subsequently, the power consumption estimation unit 14 obtains the values of the explanatory variables of the power consumption rate estimation model m1 (S320). The method for obtaining this value may be the same as that in step S230 of FIG. 11. However, in step S320, values of explanatory variables are acquired for all links in the search range. Note that when the value of a variable that does not depend on links, such as weather data, is unknown, the value of the variable that does not depend on links may be calculated based on past actual values, such as the average temperature in the area of the search range, for example. .

Next, the power consumption estimating unit 14 calculates the power consumption rate for each link by inputting the value of the explanatory variable to the learned power consumption rate estimation model m1 for each link (S330). .

Subsequently, the power consumption estimation unit 14 calculates power consumption for each link by multiplying the power consumption rate calculated in step S330 by the distance of the link (S340). As a result, information as shown in (2) of FIG. 15 is obtained. That is, in (2) of FIG. 15, the thickness of the line indicating each link indicates the magnitude of power consumption.

Next, the power consumption estimating unit 14 estimates an optimal route by solving a route optimization problem using the power consumption calculated for each link as the cost (weight) of each link (S350). In (3) of FIG. 15, an example of the optimal route is shown by a broken line. Note that a known algorithm such as Dijkstra's method or A* method may be used to solve the route optimization problem.

Note that although the example in which the power consumption rate estimation model m1 is used has been described above, only the link dependent model m22 in the first configuration example (FIG. 8) is used instead of the entire power consumption rate estimation model m1. Good too. In this case, in step S320, the value of the feature variable of the link may be acquired. In step S330, the power consumption rate of each link may be calculated by inputting the value of the characteristic variable of the link into the link dependent model m22.

Further, steps S310 to S340 may be executed asynchronously (for example, in advance) with step S350, or may be executed synchronously (in real time) with step S350.

However, if the weight (power consumption) of each link is calculated in advance (that is, steps S310 to S340 are executed in advance), the weight is calculated at the time when the optimal route is traveled. There is a possibility that values of variables that do not depend on road links, such as weather conditions, cannot be obtained. Therefore, if the sum of the weights of each link included in the optimal route is calculated as the power consumption of the optimal route, the estimated value of power consumption may deteriorate.

Therefore, when it is necessary to calculate the power consumption of the optimal route, the power consumption estimation unit 14 calculates the estimated value of the power consumption of the optimal route by executing steps S230 and S240 in FIG. 11 for the optimal route. You can. By doing so, it can be expected that the accuracy of the estimated value will improve.

As described above, according to the present embodiment, a regression model is learned based on the characteristic information of links included in the movement route of a past trip (movement) and the power consumption rate of the trip. At this time, the link position information is not included in the feature amount. Therefore, using the regression model, it is possible to estimate power consumption even for a route that has not been traveled in the past.

This makes it possible, for example, to present information indicating whether or not the electric vehicle 20 can reach the destination with the current battery level, and to present an appropriate route to a user who wants to save power consumption.

Furthermore, according to the present embodiment, variables that do not depend on links (features that do not depend on links) are also used as explanatory variables of the regression model. As a result, in estimating power consumption, it is possible to take into account conditions that do not depend on the link (weather conditions, etc.), and it is possible to expect an improvement in estimation accuracy.

Further, according to the present embodiment, for example, there is no need for various time-series data acquired from various sensors in real time (for example, periodically) within one trip, and regression is performed based on trip data. The model can be learned. This eliminates the need to acquire data from the electric vehicle 20 in real time, making it possible to reduce the cost of constructing a network for data acquisition.

Note that in this embodiment, an example has been described in which the energy (electric power) consumed by the electric vehicle 20 is estimated, but for other moving bodies that consume energy (for example, gasoline cars, diesel cars, etc.), This embodiment may be applied. In this case, the power consumption rate may be replaced by the energy consumption rate (energy consumption per unit movement distance (for example, gasoline consumption)) corresponding to each moving object, and the power consumption may be replaced by the energy consumption.

Note that in this embodiment, the power consumption estimating device 10 is an example of an energy consumption estimating device. The route estimator 11 is an example of a first estimator. The variable value extraction unit 12 is an example of an extraction unit. The model learning unit 13 is an example of a learning unit. The power consumption estimator 14 is an example of a second estimator.

Although the embodiments of the present invention have been described in detail above, the present invention is not limited to these specific embodiments, and various modifications can be made within the scope of the gist of the present invention as described in the claims. - Can be changed.

1 Power consumption estimation system 10 Power consumption estimation device 11 Route estimation section 12 Variable value extraction section 13 Model learning section 14 Power consumption estimation section 20 Electric vehicle 100 Drive device 101 Recording medium 102 Auxiliary storage device 103 Memory device 104 CPU
105 Interface device 121 Trip data storage unit 122 Map data storage unit 123 Weather log data storage unit B Bus m1 Power consumption rate estimation model m21 Link independent model m22 Link dependent model m23 Synthesis model

Claims (4)

For each of a plurality of trip data including a travel start point, a travel end point, a travel distance, and energy consumption, a set of links that indicates a travel route related to the trip data by referring to map data that represents a road as a set of links. identify ,
extracting characteristic information corresponding to each link included in the identified set of links with reference to a storage unit that stores road characteristic information;
Based on the characteristic information extracted for each link and the travel distance and energy consumption included in the plurality of trip data, the characteristic information is explained using an energy consumption rate, which is energy consumption per unit travel distance, as an objective variable. Learn a regression model with variables,
Calculating the energy consumption for each link included in the search range of the travel route between the specified departure point and the specified destination based on the characteristic information of the link and the regression model,
estimating an optimal route between the departure point and the destination using the energy consumption calculated for each link included in the search range as a weight for each link;
An energy consumption estimation program characterized by causing a computer to perform processing. For each of the plurality of trip data, causing a computer to execute a process of extracting information that does not depend on road characteristics and that affects energy consumption,
The learning process learns the regression model based on the information,
The calculating process calculates energy consumption based on the information.
The energy consumption estimation program according to claim 1. For each of a plurality of trip data including a travel start point, a travel end point, a travel distance, and energy consumption, a set of links that indicates a travel route related to the trip data by referring to map data that represents a road as a set of links. identify ,
extracting characteristic information corresponding to each link included in the identified set of links with reference to a storage unit that stores road characteristic information;
Based on the characteristic information extracted for each link and the travel distance and energy consumption included in the plurality of trip data, the characteristic information is explained using an energy consumption rate, which is energy consumption per unit travel distance, as an objective variable. Learn a regression model with variables,
Calculating the energy consumption for each link included in the search range of the travel route between the specified departure point and the specified destination based on the characteristic information of the link and the regression model,
estimating an optimal route between the departure point and the destination using the energy consumption calculated for each link included in the search range as a weight for each link;
A method for estimating energy consumption, characterized in that processing is performed by a computer. For each of a plurality of trip data including a travel start point, a travel end point, a travel distance, and energy consumption, a set of links that indicates a travel route related to the trip data by referring to map data that represents a road as a set of links. a specific part that specifies the
an extraction unit that refers to a storage unit that stores road characteristic information and extracts characteristic information corresponding to each link included in the identified set of links;
Based on the characteristic information extracted for each link and the travel distance and energy consumption included in the plurality of trip data, the characteristic information is explained using an energy consumption rate, which is energy consumption per unit travel distance, as an objective variable. a learning section that learns a regression model as a variable;
a calculation unit that calculates energy consumption for each link included in a travel route search range between a specified departure point and a specified destination based on characteristic information of the link and the regression model;
an estimation unit that estimates an optimal route between the departure point and the destination using the energy consumption calculated for each link included in the search range as a weight for each link;
An energy consumption estimating device characterized by having:

Images