JP6804934B2 - Energy consumption prediction device and energy consumption prediction method - Google Patents

Description

An embodiment of the present invention relates to an energy consumption prediction device and an energy consumption prediction method.

In recent years, battery-powered vehicles such as electric vehicles have been developed. The points that power such vehicles (such as charging stations) are small and may be geographically limited. In order to determine whether a battery-powered vehicle can reach a target point, it is necessary to predict the energy consumption to the target point. For example, the user of the vehicle acquires the predicted energy consumption from a prediction server or the like that predicts the energy consumption. The prediction server is required to have a technology for accurately predicting energy consumption according to various vehicles.

Japanese Patent No. 5855163 Japanese Unexamined Patent Publication No. 2014-163996 Japanese Patent No. 5818774

In order to solve the above problems, an energy consumption prediction device and an energy consumption prediction method capable of predicting the energy consumption of a vehicle are provided.

According to the embodiment, the energy consumption prediction device for predicting the energy consumption of the vehicle using the model includes a learning data acquisition unit, a cluster classification unit, a model generation unit, a prediction data acquisition unit, and a prediction unit. , Equipped with. The learning data acquisition unit acquires a plurality of learning data including battery information regarding the vehicle battery and motion information including energy consumption of the vehicle. The cluster classification unit classifies the learning data acquired by the learning data acquisition unit into a plurality of clusters according to the classification criteria. The model generation unit includes a determination unit, a first generation unit, and a second generation unit. The determination unit determines whether the number of learning data included in each cluster classified by the cluster classification unit is equal to or greater than the threshold value. The first generation unit determines the energy consumption of the vehicle having a plurality of parameters by regression based on the learning data for the cluster determined by the determination unit that the number of the learning data is equal to or greater than the threshold value. Generate a model to predict. The second generation unit has the cluster having a parameter obtained by integrating the same correction value into the model parameter for each cluster for the cluster in which the determination unit determines that the number of learning data is not equal to or greater than the threshold value. A model that predicts the energy consumption of the vehicle is generated by an approximate model that is close to. The prediction data acquisition unit acquires prediction data for predicting energy consumption from the vehicle. The prediction unit determines a cluster from the prediction data acquired by the prediction data acquisition unit according to the classification criteria, and from the prediction data according to the model generated by the model generation unit corresponding to the determined cluster. Predict the energy consumption of the vehicle.

FIG. 1 is a block diagram showing a configuration example of a prediction system according to the first embodiment. FIG. 2 is a block diagram showing a configuration example of the prediction server according to the first embodiment. FIG. 3 is a diagram showing a configuration example of learning data when classified into clusters according to the first embodiment. FIG. 4 is a diagram showing a configuration example of learning data when collected for each cluster according to the first embodiment. FIG. 5 is a diagram showing a configuration example of the consumption model learning unit according to the first embodiment. FIG. 6 is a diagram showing an example of a cluster according to the first embodiment. FIG. 7 is a diagram for explaining the correction of the parameters according to the first embodiment. FIG. 8 is a diagram for explaining the sparse estimation according to the first embodiment. FIG. 9 is a diagram for explaining the sparse estimation according to the first embodiment. FIG. 10 is a flowchart showing an operation example of the prediction server according to the first embodiment. FIG. 11 is a flowchart showing an operation example of the prediction server according to the first embodiment. FIG. 12 is a flowchart showing an operation example of the prediction server according to the first embodiment. FIG. 13 is a block diagram showing a configuration example of the prediction server according to the second embodiment. FIG. 14 is a diagram for explaining the classification of clusters according to the second embodiment. FIG. 15 is a diagram for explaining the classification of clusters according to the second embodiment. FIG. 16 is a flowchart showing an operation example of the prediction server according to the second embodiment. FIG. 17 is a block diagram showing a configuration example of the prediction system according to the third embodiment. FIG. 18 is a block diagram showing a configuration example of the prediction system according to the fourth embodiment.

Hereinafter, embodiments will be described with reference to the drawings.
(First Embodiment)
The prediction system according to the embodiment predicts the electric power consumed by a vehicle such as an electric vehicle traveling on a road. For example, the prediction system provides the predicted energy consumption (predicted energy consumption) to the user of the vehicle.

Vehicles for which the prediction system calculates the predicted energy consumption are self-propelled with a battery. The vehicle for which the predicted energy consumption is calculated in each of the embodiments described below is assumed to be an electric vehicle traveling on a road. The vehicle is not limited to the electric vehicle, and may be a battery-powered train, an aircraft, a ship, or the like.

FIG. 1 is a block diagram showing a configuration example of the prediction system 1 according to the embodiment.
As shown in FIG. 1, the prediction system 1 includes a prediction server 10, a weather information server 20, a traffic information server 30, a charger management server 40, an application management server 50, chargers Q1 to Qn, vehicles V1 to Vn, and a user terminal. It includes F1 to Fn and the like.

The prediction server 10 communicates with the weather information server 20, the traffic information server 30, the charger management server 40, and the application management server 50. For example, the prediction server 10 connects to the weather information server 20, the traffic information server 30, the charger management server 40, and the application management server 50 via the Internet or its own network.

Further, the charger management server 40 communicates with the chargers Q1 to Qn via the Internet or the like. Further, the application management server 50 communicates with the user terminals F1 to Fn via the Internet or the like. Further, the chargers Q1 to Qn are connected to the vehicle so that the batteries mounted on the vehicles V1 to Vn can be charged, respectively.

The prediction server 10 (energy consumption prediction device) predicts the predicted energy consumption of vehicles V1 to Vn used by the user based on information from the weather information server 20, the traffic information server 30, the charger management server 40, the application management server 50, and the like. Is calculated. The prediction server 10 transmits the calculated predicted energy consumption to the user terminals F1 to Fn of the user who possesses the vehicles V1 to Vn through the application management server 50. The prediction server 10 will be described in detail later.

The weather information server 20 transmits information about the weather (weather information) to the prediction server 10. The weather information server 20 has a weather information DB 21. The weather information DB 21 stores a plurality of weather information. The meteorological information is information such as temperature, precipitation, wind speed, wind direction or snowfall in a predetermined period. The meteorological information DB 21 stores meteorological information for each region and period. For example, when the weather information server 20 acquires data related to the weather, it generates weather information based on the data and stores it in the weather information DB 21.

The weather information server 20 transmits the weather information stored in the weather information DB 21 to the prediction server 10. For example, the weather information server 20 transmits weather information to the prediction server 10 in response to a request from the prediction server 10. Further, the weather information server 20 may periodically transmit the weather information to the prediction server 10.

The traffic information server 30 transmits traffic information related to road traffic to the prediction server 10. The traffic information server 30 has a plurality of traffic information DB 31s. The traffic information DB 31 stores traffic information. The traffic information is information indicating the speed of each vehicle traveling on a predetermined road (for example, the average speed of each vehicle). The traffic information DB 31 stores traffic information for each road. For example, when the traffic information server 30 acquires data related to traffic on a new road, it generates traffic information based on the data and stores it in the traffic information DB 31.

The traffic information server 30 transmits the traffic information stored in the traffic information DB 31 to the prediction server 10. For example, the traffic information server 30 transmits traffic information to the prediction server 10 in response to a request from the prediction server 10. Further, the traffic information server 30 may periodically transmit traffic information to the prediction server 10.

The charger management server 40 transmits consumption history information indicating the energy consumption of the vehicles V1 to Vn to the prediction server 10. The consumption history information is, for example, information indicating the energy consumption of the vehicles V1 to Vn in a predetermined period (for example, from the start to the end of the movement of the vehicles V1 to Vn).

The consumption history information includes motion information and vehicle information of vehicles V1 to Vn. The charger management server 40 has a consumption history information DB 41. The consumption history information DB 41 stores the consumption history information. For example, when the charger management server 40 acquires new data, it generates consumption history information based on the data and stores the generated consumption history information in the consumption history information DB 41.

The kinetic information is information on the kinetic energy of the vehicles V1 to Vn. The exercise information includes, for example, a travel date and time, a travel section, a travel distance of the travel section, a travel time traveled in the travel section, an energy consumption amount displayed in the travel section, and the like.
The vehicle information is information about vehicles V1 to Vn. The vehicle information includes, for example, vehicle type information indicating a vehicle type of vehicles V1 to Vn, battery information regarding a battery mounted on the vehicles V1 to Vn, and the like.

The charger management server 40 generates consumption history information based on the data from the chargers Q1 to Qn. For example, the charger management server 40 acquires movement information and vehicle information from the chargers Q1 to Qn. The charger management server 40 generates consumption history information indicating exercise information and vehicle information from the chargers Q1 to Qn, and adds the generated consumption history information to the consumption history DB 41.

The charger management server 40 transmits the consumption history information stored in the consumption history information DB 41 to the prediction server 10. For example, the charger management server 40 transmits consumption history information to the prediction server 10 in response to a request from the prediction server 10. Further, the charger management server 40 may periodically (for example, daily) transmit consumption history information to the prediction server 10.

The chargers Q1 to Qn charge the battery mounted on the vehicle. Here, it is assumed that the chargers Q1 to Qn charge the batteries mounted on the vehicles V1 to Vn, respectively. The chargers Q1 to Qn acquire motion information and vehicle type information of the vehicle from the charged vehicle. For example, the chargers Q1 to Qn acquire motion information and vehicle type information from the vehicle when connected to the vehicle.

The chargers Q1 to Qn may be installed in a charging stand or the like. Further, the chargers Q1 to Qn may be installed in the house of the user who owns the vehicles V1 to Vn.

The vehicles V1 to Vn are vehicles equipped with a battery and self-propelled by the electric power required from the battery. Vehicles V1 to Vn generate their own motion information based on a signal from a GPS sensor or the like or an internal clock or the like. Further, the vehicles V1 to Vn store their own vehicle information. Alternatively, it may be estimated using the information of the charger on the road where the movement route can be estimated such as a highway.

When the vehicles V1 to Vn are connected to the chargers Q1 to Qn, the vehicles V1 to Vn charge the battery with the electric power from the chargers Q1 to Qn. Further, the vehicles V1 to Vn transmit their own motion information and vehicle information to the chargers Q1 to Qn.

The application management server 50 distributes the predicted energy consumption to the user terminals F1 to Fn possessed by the users of the vehicles V1 to Vn. For example, the application management server 50 distributes the predicted energy consumption to the applications operating on the user terminals F1 to Fn.

Further, the application management server 50 acquires prediction data necessary for predicting the predicted energy consumption from the user terminals F1 to Fn. For example, the application management server 50 acquires prediction data generated based on an operation input by the user to the application from the application. Further, the prediction data may be acquired in real time by GPS from an application, a car navigation system, or a vehicle.

The prediction data includes traveling information indicating a section and a date and time for which the user wants to predict energy consumption, and vehicle information of the user's vehicles V1 to Vn. The traveling information is, for example, information such as a traveling date and time, a departure place, and a destination.

The application management server 50 includes a prediction data DB 51. The prediction data DB 51 stores the prediction data.
The application management server 50 transmits the prediction data stored in the prediction data DB 51 to the prediction server 10 at a predetermined timing.

The user terminals F1 to Fn are terminal devices of the user who possesses or uses the vehicles V1 to Vn. For example, the user terminals F1 to Fn may be smartphones, tablet PCs, PCs, car navigation systems, or the like. The user terminals F1 to Fn receive input of travel information and vehicle information from the user. The user terminals F1 to Fn generate prediction data composed of traveling information and vehicle information. The user terminals F1 to Fn transmit the generated prediction data to the application management server 50.
The prediction system 1 is not limited to the above-described configuration, and a necessary configuration may be added and unnecessary configurations may be deleted as appropriate.

Next, the prediction server 10 will be described.
The prediction server 10 includes a CPU, a ROM, a RAM, an NVM, a communication unit, an operation unit, a display unit, and the like as a basic configuration. For example, a CPU is a processor that realizes various processes by executing a program stored in a memory such as an NVM.

FIG. 2 is a block diagram illustrating the function of the prediction server 10.
As shown in FIG. 2, the prediction server 10 includes a learning data acquisition unit 11, a cluster classification unit 12, a classification standard DB 13, a consumption model learning unit 14, a model parameter DB 15, a prediction data acquisition unit 16, and an energy consumption prediction unit 17. , Energy consumption information DB18, road shape information DB19, and the like.

The learning data acquisition unit 11 acquires a plurality of learning data for generating a model (consumption model) for calculating the predicted energy consumption. The learning data is composed of consumption history information, weather information, road shape information, and the like. The road shape information will be described later.

The learning data acquisition unit 11 acquires consumption history information from the charger management server 40. Further, the learning data acquisition unit 11 acquires weather information from the weather information server 20. The learning data acquisition unit 11 may transmit a request for consumption history information and a request for weather information to the charger management server 40 and the weather information server 20, respectively.
Further, the learning data acquisition unit 11 acquires the road shape information from the road shape information DB 19. The learning data acquisition unit 11 may acquire road shape information from an external device.

The learning data acquisition unit 11 generates learning data by associating consumption history information, weather information, and road shape information. For example, the learning data acquisition unit 11 associates the consumption history information with the weather information corresponding to the travel date and time of the consumption history information and the road shape information corresponding to the travel section of the consumption history information for learning. Get the data.

The cluster classification unit 12 classifies a plurality of learning data into a plurality of clusters. For example, the cluster classification unit 12 classifies the learning data into a plurality of clusters according to the classification criteria stored in the classification criteria DB 13 described later. For example, the classification standard may be a standard based on vehicle information, battery information, or the like. For example, the classification standard may be a standard based on the vehicle type. The classification criteria are not limited to a specific configuration.
The cluster classification unit 12 acquires a classification standard from the classification standard DB 13. The cluster classification unit 12 classifies a plurality of learning data into a plurality of clusters according to the acquired classification criteria.

FIG. 3 shows a configuration example of learning data classified into clusters by the cluster classification unit 12.
As shown in FIG. 3, the learning data is composed of an ID, energy consumption y, mileage x1, clusters, and the like.

The ID is an identifier for identifying the learning data. For example, the ID may be an ID set in the consumption history information. Further, the cluster classification unit 12 may assign an ID to the learning data.

The energy consumption y is the electric power consumed by the vehicles V1 to Vn for traveling. For example, the energy consumption y is a value obtained from exercise information.
The mileage x1 is the distance traveled by the vehicles V1 to Vn.

The cluster is a cluster name given to the learning data by the cluster classification unit 12. Here, it is assumed that the cluster classification unit 12 assigns A, B or C as the cluster name.
The classification standard DB 13 stores the classification standard. For example, the classification standard DB 13 stores the classification standard in advance by the operation of the operator. The classification standard DB 13 may update the classification standard by the operation of the operator.

The consumption model learning unit 14 (model generation unit) generates a model (model parameter) for calculating the predicted energy consumption for each cluster. The consumption model learning unit 14 collects learning data for each cluster. The consumption model learning unit 14 generates a model of the cluster based on the learning data classified into the clusters.

FIG. 4 is a configuration example of learning data collected by the consumption model learning unit 14 for each cluster.
In the example shown in FIG. 4, the consumption model learning unit 14 collects the learning data of A in order to generate the model of A (cluster name). The consumption model learning unit 14 stores the generated model (model parameter) in the model parameter DB 15.

The model parameter DB 15 stores the model generated by the consumption model learning unit 14. The model parameter DB 15 stores a model for each cluster. The model parameter DB 15 may acquire the model generated by the consumption model learning unit 14 and update the existing model.

The prediction data acquisition unit 16 acquires prediction data from the application management server 50. For example, the prediction data acquisition unit 16 acquires prediction data together with a power prediction request from the application management server 50. Further, the prediction data acquisition unit 16 acquires traffic information from the traffic information server 30.

The energy consumption prediction unit 17 acquires a model corresponding to a cluster of prediction data from the model parameter DB 15. Further, the energy consumption prediction unit 17 acquires the weather information corresponding to the prediction data (for example, the weather information corresponding to the date and time of the prediction data) from the weather information server 20. Further, the energy consumption prediction unit 17 acquires road shape information corresponding to the prediction data (for example, road shape information corresponding to the road between the departure point and the destination of the prediction data).

The energy consumption prediction unit 17 (prediction unit) calculates the predicted energy consumption consumed by the vehicles V1 to Vn based on the prediction data and the model. The energy consumption prediction unit 17 determines a cluster of prediction data based on the prediction data. The energy consumption prediction unit 17 determines a cluster of prediction data in the same manner as the cluster classification unit 12 based on the classification standard stored in the classification standard DB 13.

The energy consumption prediction unit 17 calculates the predicted energy consumption based on the acquired model, meteorological information, road shape information, and prediction data. For example, the energy consumption prediction unit 17 calculates the predicted energy consumption by substituting the weather information, the road shape information, and the prediction data into the prediction formula including the model parameters. Further, the energy consumption prediction unit 17 may further calculate the predicted energy consumption based on the traffic information.

The energy consumption prediction unit 17 transmits energy consumption information indicating the calculated predicted energy consumption to the application management server 50. Further, the energy consumption prediction unit 17 stores the energy consumption information in the energy consumption information DB 18.

The energy consumption information DB 18 stores energy consumption information indicating the predicted energy consumption calculated by the energy consumption prediction unit 17.

The road shape information DB 19 stores the road shape information in advance. The road shape information is information regarding the potential energy of the vehicle bodies V1 to Vn. For example, road shape information indicates the distance or slope of a road. The road shape information DB 19 stores road shape information for each road or position.
The prediction server 10 may add necessary configurations and delete unnecessary configurations as appropriate.

Next, the consumption model learning unit 14 will be described.
FIG. 5 is a block diagram showing the functions of the consumption model learning unit 14.
As shown in FIG. 5, the consumption model learning unit 14 includes a model construction method classification unit 141, a model parameter learning unit 142, a minority data model parameter learning unit 143, and the like.

The model construction method classification unit 141 (determination unit) determines a construction method for creating a model of a cluster. The model construction method classification unit 141 determines the construction method based on the number of learning data classified into clusters. For example, the model construction method classification unit 141 generates a model in the model parameter learning unit 142 when the number of learning data classified into the cluster is equal to or more than a predetermined threshold value. Further, when the number of training data is less than a predetermined threshold value, the model construction method classification unit 141 generates a model in the minority data model parameter learning unit 143. The predetermined threshold is, for example, the number of explanatory variables used to estimate the predicted energy consumption. For example, the explanatory variables are individual elements (excluding energy consumption y) that make up the training data. Explanatory variables are distance or travel time and the like. The explanatory variables may be a part of individual elements constituting the learning data.

The model parameter learning unit 142 (first generation unit) calculates a raster model in which the number of learning data is larger than a predetermined threshold value. For example, the model parameter learning unit 142 calculates the regression coefficient for calculating the energy consumption according to the following equation.

In equation (1), y is the predicted energy consumption, xi is the explanatory variable, and αi is the regression coefficient. p is the number of explanatory variables, and xi and αi indicate the i-th explanatory variable.

The model parameter learning unit 142 selects the explanatory variables used in the equation (1).
For example, the model parameter learning unit 142 performs normalization and principal component analysis on the explanatory variables that can be obtained from the learning data. For example, the model parameter learning unit 142 calculates the dispersion expansion coefficient (VIF). The model parameter learning unit 142 excludes explanatory variables having a VIF of 10 or more.

In addition, the model parameter learning unit 142 combines the remaining explanatory variables by brute force to determine the combination of the explanatory variables having the smallest Akaike's Information Criterion (AIC).
For example, the model parameter learning unit 142 generates a combination of explanatory variables as follows. The model parameter learning unit 142 may set a new explanatory variable by adding, subtracting, integrating, or dividing the explanatory variables.

Combination 1: Distance, running time, distance x running time ...
Combination 2: Distance, distance x running time ...
The model parameter learning unit 142 generates a plurality of combinations of explanatory variables as described above. The model parameter learning unit 142 calculates the AIC for each combination and determines the combination in which the AIC is the minimum.

The model parameter learning unit 142 selects the combination with the smallest AIC as the explanatory variable used in the equation (1).
The model parameter learning unit 142 may calculate the Bayesian information criterion (BIC) instead of the AIC.

The model parameter learning unit 142 constructs the equation (1) using the selected explanatory variables. For example, the model parameter learning unit 142 calculates αi based on the selected explanatory variable. The model parameter learning unit 142 stores the calculated αi as a model in the model parameter DB 15.

The minority data model parameter learning unit 143 (second generation unit) calculates a cluster model in which the number of learning data is equal to or less than a predetermined threshold value. For example, the model parameter learning unit 142 calculates the regression coefficient for calculating the energy consumption according to the following equation.

In equation (2), y is the predicted energy consumption (energy consumption), xi is the explanatory variable, and βi is the regression coefficient.

The minority data model parameter learning unit 143 generates a model of a target cluster (target cluster) based on a model (approximate model) of a cluster (approximate cluster) that is close to the cluster that generates the model (transfer learning). ).

First, the minority data model parameter learning unit 143 selects an approximate model. For example, the minority data model parameter learning unit 143 acquires each model from the model parameter DB 15. The minority data model parameter learning unit 143 applies each learning data of the target cluster to each model and calculates the predicted energy consumption. The minority data model parameter learning unit 143 selects the model having the smallest error between the predicted energy consumption and the energy consumption of the training data as an approximate model.

FIG. 6 is a diagram showing a target cluster and an approximate cluster.
In FIG. 6, the horizontal axis represents an estimated value (estimated energy consumption), and the vertical axis represents an actually measured value (energy consumption of learning data). Further, in the example shown in FIG. 6, the target cluster is cluster B, and the approximate cluster is cluster A. In FIG. 6, each point shows one learning data. That is, each point indicates the predicted energy consumption obtained from the training data using the model of cluster A and the energy consumption of the training data. Further, in FIG. 6, the line 61 shows a straight line in which the measured value = the estimated value.
As shown in FIG. 6, each point belonging to cluster A (approximate cluster) indicates that the predicted energy consumption and the energy consumption are almost the same.
In addition, each point belonging to cluster B (target cluster) indicates that the predicted energy consumption and the energy consumption do not match, but they are relatively close to each other.

The minority data model parameter learning unit 143 calculates the model of the target cluster based on the model of the approximate cluster (approximate model).
For example, the minority data model parameter learning unit 143 calculates βi according to the following formula.

In equation (3), αi is the regression coefficient of the approximate model, and εi (correction value) indicates the ratio of the approximate model to the model of the target cluster.
As shown by equation (3), the minority data model parameter learning unit 143 defines βi as a constant multiple of αi.

That is, the minority data model parameter learning unit 143 calculates the predicted energy consumption y according to the following equation.

For example, the minority data model parameter learning unit 143 calculates εi according to the assumption of the following equation.

That is, the minority data model parameter learning unit 143 assumes that the coefficient multiplied by αi is constant.

The minority data model parameter learning unit 143 calculates the predicted energy consumption y as shown in the following formula.

In the formula (6), y indicates the predicted energy consumption, and y'represents the predicted energy consumption calculated from the approximate model.

y'is calculated by the following formula.

That is, the minority data model parameter learning unit 143 assumes that the model of the target cluster is a linear form using an approximate cluster. The minority data model parameter learning unit 143 calculates εe and ε by regression based on the learning data of the target cluster.

FIG. 7 is a diagram showing an image of a model of the target cluster.
In FIG. 7, line 62 represents equation (6). That is, the line 62 is a straight line having an intercept of εe and a slope of ε.
As shown in FIG. 7, the line 62 approximates each point of the cluster B (target cluster).

Further, the minority data model parameter learning unit 143 may calculate βi according to the following equation.

In equation (8), αi is the regression coefficient of the approximate model, and εi indicates the difference between the approximate model and the model of the target cluster.
As shown by the equation (8), the minority data model parameter learning unit 143 defines βi as the value obtained by adding εi to αi.
That is, the minority data model parameter learning unit 143 calculates the predicted energy consumption y according to the following equation.

The difference from the predicted energy consumption calculated from the approximate model is calculated by the following formula.

In the formula (10), y indicates the predicted energy consumption, and y'represents the predicted energy consumption calculated from the approximate model.

The minority data model parameter learning unit 143 performs sparsity estimation assuming the sparsity of the model of the target cluster. That is, the minority data model parameter learning unit 143 calculates the model on the assumption that some of εi are 0.

Next, sparse estimation will be described. Here, the method of estimating the regression coefficient will be described using the least squares method. There are two typical methods for sparse estimation, Lasso and Elastic Net.
Since Lasso is a kind of Elastic Net, the Elastic Net will be described.

Elastic Net is a problem equation (11) for minimizing the loss function J (J is the sum of the squared errors of the measured value and the estimated value) of the least squares method using a penalty term when performing the least squares method for finding β. (Equation (3) is a concave function). The penalties for Elastic Net are shown in Equation (12).

λ is a parameter that determines the strength of the penalties. For λ, the training data is usually divided, and the optimum λ is determined by a cross-validation method (generally about 10 to 100 divisions). α is a parameter that adjusts the strength of the first term and the second term of the equation (12), and is in the range of 0 <α <1 in the Elastic Net and α = 1 in the Lasso. The value of the parameter is arbitrary, but here it is divided into 10 and α = 0.5.

In addition, the minority data model parameter learning unit 143 measures εi and the error using sparse estimation and cross-validation.
FIG. 8 is a diagram for explaining a cross-validation method.
In FIG. 8, circles 1 to 5 represent learning data, respectively. The solid circle indicates the learning data used to obtain εi (regression coefficient) in regression. Further, the broken line circle indicates the learning data used for calculating the error using the obtained εi. Here, it is assumed that each trial is performed using five learning data.

In trial 71, the learning data indicated by circle 1 is used to calculate the error.
In trial 72, the learning data indicated by circle 2 is used to calculate the error.
In trial 73, the learning data indicated by circle 3 is used to calculate the error.
In trial 74, the training data indicated by circle 4 is used to calculate the error.
In trial 75, the learning data indicated by circle 5 is used to calculate the error.

For example, the minority data model parameter learning unit 143 calculates εi using the learning data of the circles 2 to 5 in the trial 71. The minority data model parameter learning unit 143 calculates the predicted energy consumption based on the learning data of the circle 1 using the calculated εi. Further, the minority data model parameter learning unit 143 acquires the difference between the energy consumption amount and the predicted energy consumption of the learning data of the circle 1 as an error.

Similarly, the minority data model parameter learning unit 143 performs the same operation in trials 72 to 75.

FIG. 9 is a table showing εi and the error.
In FIG. 9, the trial ID identifies the trial. For example, trial IDs "1" to "5" indicate trials 71 to 75, respectively.

As shown in FIG. 9, the minority data model parameter learning unit 143 calculates ε0 = 0, ε1 = 0, ... Error = 32.4% in the trial ID “1”. Similarly, the decimal data model parameter learning unit 143 calculates ε0 = 8.3728, ε1 = 0, ... Error = 13.3% in the trial ID “2”.

Further, the minority data model parameter learning unit 143 calculates a selection reference value for selecting εi. The selection reference value is calculated for each parameter. The selection reference value is a value obtained by calculating εi / error in each trial and averaging εi / error in each trial.

For example, when calculating the selection reference value of ε1, the minority data model parameter learning unit 143 calculates ε1 / error% in trials “1” to “5”. The minority data model parameter learning unit 143 calculates the selection reference value by averaging the calculated values.

The minority data model parameter learning unit 143 selects a combination of εi (non-zero εi) required for the model of the target cluster based on the selection reference value. For example, the minority data model parameter learning unit 143 generates a combination in which εi corresponding to the parameter of the largest selection reference value is required εi and the other εi are assumed to be 0.

Further, the minority data model parameter learning unit 143 generates a combination assuming that εi corresponding to the parameter of the next largest selection reference value is also required εi. That is, the minority data model parameter learning unit 143 assumes that εi corresponding to the parameter of the largest selection reference value and εi corresponding to the parameter of the next largest selection reference value are not 0, and assumes that the others are 0. To generate.

By repeating the above operation, the minority data model parameter learning unit 143 generates a combination on the assumption that εi corresponding to the parameter having a large selection reference value is not 0.

The minority data model parameter learning unit 143 calculates εi by regression in each combination and calculates the model parameter (βi). The minority data model parameter learning unit 143 calculates the AIC of each model and selects the most appropriate combination.

The minority data model parameter learning unit 143 determines that the most appropriate combination is the combination of εi (non-zero εi) required for the model of the target cluster. The minority data model parameter learning unit 143 calculates βi based on εi in the combination.

The minority data model parameter learning unit 143 stores the calculated βi in the model parameter DB 15.
The method of sparse estimation by the minority data model parameter learning unit 143 is not limited to a specific configuration.

Next, an operation example of the prediction server 10 will be described.
First, an operation example in which the prediction server 10 builds a model of learning data for each cluster will be described.
FIG. 10 is a flowchart for explaining an operation example in which the prediction server 10 builds a model for each cluster of learning data.

First, the learning data acquisition unit 11 of the prediction server 10 learns a plurality of learning data based on the consumption history information from the charger management server 40, the weather information from the weather information server 20, and the road shape information from the road shape information DB 19. Data is acquired (S11).

When the learning data acquisition unit 11 acquires the learning data, the cluster classification unit 12 classifies each learning data into a plurality of clusters according to the classification criteria stored in the classification criterion DB 13 (S12).

When the cluster classification unit 12 classifies the learning data into a plurality of clusters, the consumption model learning unit 14 substitutes 1 for i (S13). Here, i is a number that identifies the cluster. Here, it is assumed that the clusters are numbered in order from 1.

Substituting 1 for i, the consumption model learning unit 14 builds the model of the i-th cluster (S14). When the model of the i-th cluster is constructed, the consumption model learning unit 14 stores the constructed model in the model parameter DB 15 (S15).

When the constructed model is stored in the model parameter DB 15, the consumption model learning unit 14 increments i (S16). When i is incremented, it is determined whether i is larger than the number of clusters (S17).

When it is determined that i is not larger than the number of clusters (S17, NO), the consumption model learning unit 14 returns to S14.
When the consumption model learning unit 14 determines that i is larger than the number of clusters (S17, YES), the prediction server 10 terminates the operation.

Next, an operation example (corresponding to S14 in FIG. 10) in which the prediction server 10 builds a model will be described.
FIG. 11 is a flowchart for explaining an operation example in which the prediction server 10 builds a model.

First, the model construction method classification unit 141 of the prediction server 10 determines whether the number of learning data of the cluster exceeds a predetermined threshold value (S21).
When the model building method classification unit 141 determines that the number of training data of the cluster is larger than a predetermined threshold value (S21, YES), the model parameter learning unit 142 builds a model based on the learning data of the cluster (S21, YES). S22).

When the model construction method classification unit 141 determines that the number of training data of the cluster is equal to or less than a predetermined threshold value (S21, NO), the minority data model parameter learning unit 143 acquires the parameters of each model from the model parameter DB 15. (S23).

When the parameters of each model are acquired, the minority data model parameter learning unit 143 calculates the predicted energy consumption for each model (S24). When the predicted energy consumption is calculated for each model, the minority data model parameter learning unit 143 calculates an error between the energy consumption of the training data and the predicted energy consumption calculated from each model (S25).

When the error is calculated, the minority data model parameter learning unit 143 selects the model with the smallest error as an approximate model (S26). When the approximate model is selected, the minority data model parameter learning unit 143 calculates εi for correcting the approximate model (S27).
When the model parameter learning unit 142 builds a model based on the learning data of the cluster (S22), or when the minority data model parameter learning unit 143 calculates εi (S27), the prediction server 10 ends the operation. To do.

Next, an operation example in which the prediction server calculates the predicted energy consumption from the prediction data will be described.
FIG. 12 is a flowchart for explaining an operation example in which the prediction server calculates the predicted energy consumption from the prediction data.

First, the prediction data acquisition unit 16 of the prediction server 10 acquires prediction data from the application management server 50 (S31). For example, the application management server 50 receives a request requesting predicted energy consumption from the user terminals F1 to Fn. The application management server 50 transmits the prediction data together with the request to the prediction server 10.

When the prediction data acquisition unit 16 acquires the prediction data, the energy consumption prediction unit 17 determines a cluster of the prediction data (S32).

When the cluster of the prediction data is determined, the energy consumption prediction unit 17 acquires the model of the determined cluster from the model parameter DB 15 (S33). When the model is acquired, the energy consumption prediction unit 17 calculates the predicted energy consumption from the prediction data according to the acquired model (S34).

When the predicted energy consumption is calculated, the energy consumption prediction unit 17 stores the energy consumption information indicating the calculated predicted energy consumption in the energy consumption information DB 18 (S35). When the energy consumption prediction unit 17 stores the energy consumption information in the energy consumption information DB 18, the prediction server 10 ends the operation.
The energy consumption prediction unit 17 may further calculate the predicted energy consumption based on the traffic information from the traffic information server 30.

The prediction server 10 transmits the calculated predicted energy consumption to the application management server 50. The application management server 50 transmits the predicted energy consumption from the prediction server 10 to the user terminals F1 to Fn that have sent the request.
The user terminals F1 to Fn display the received predicted energy consumption on a display unit or the like.

The prediction server 10 may display the predicted energy consumption on the display unit.

The prediction server configured as described above classifies the training data into clusters. The prediction server 10 builds a model for each cluster. The prediction server calculates the predicted energy consumption from the prediction data using a model corresponding to the cluster of the prediction data. As a result, the prediction server can use a plurality of models properly and can calculate the predicted energy consumption more effectively.

In addition, the prediction server builds a model based on an approximate model for clusters in which the number of training data is less than or equal to the threshold value. As a result, the prediction server can build an appropriate model even for a cluster with a small amount of training data.

Further, the prediction server selects the necessary εi based on the selection reference value when performing sparse estimation. As a result, the prediction server can prevent overfitting.

(Second Embodiment)
Next, the prediction system 1'according to the second embodiment will be described.
The prediction system 1'according to the second embodiment is different from the prediction system 1 according to the first embodiment in that a classification standard is constructed. Therefore, other configurations are designated by the same reference numerals and detailed description thereof will be omitted.

FIG. 1 is a block diagram showing a configuration example of the prediction system 1'according to the second embodiment.
The prediction system 1'provides a prediction server 10'instead of the prediction server 10.
Since the configurations of the other parts are the same as those of the first embodiment, the description thereof will be omitted.

FIG. 13 is a block diagram showing a configuration example of the prediction server 10'.
As shown in FIG. 13, the prediction server 10'also includes a classification standard learning unit 25.

The classification standard learning unit 25 builds a classification standard based on the learning data.
For example, the classification standard learning unit 25 (first classification standard generation unit) constructs a classification standard using non-hierarchical cluster analysis.

FIG. 14 is a diagram for explaining non-hierarchical cluster analysis.
The classification standard learning unit 25 sets the conditions according to the operation from the operator. In the example shown in FIG. 14, the classification standard learning unit 25 sets the condition a and the condition b.

The condition a is that the predetermined parameter (parameter N) included in the battery information is larger than the predetermined threshold value (parameter N). That is, the condition a is parameter N> parameter N threshold.

The condition b is that the electricity cost is larger than a predetermined threshold value (electricity cost threshold value). That is, the condition b is electricity cost> electricity cost threshold. Electricity cost is the distance traveled per unit electric power. That is, the electricity cost is the mileage / energy consumption.

The classification standard learning unit 25 sets the parameter N threshold value and the electricity cost threshold value.
Next, as shown in FIG. 14, the classification standard learning unit 25 determines whether the learning data satisfies the condition a. The classification standard learning unit 25 determines whether the learning data satisfying the condition a further satisfies the condition b. The classification standard learning unit 25 classifies the learning data satisfying the condition b (that is, the learning data satisfying the conditions a and b) into the cluster A.

Further, the classification standard learning unit 25 classifies the learning data that does not satisfy the condition b (that is, the learning data that does not satisfy the condition a but satisfies the condition b) into the cluster B.
Further, the classification standard learning unit 25 classifies the learning data that does not satisfy the condition a into the cluster C.

The classification standard learning unit 25 builds a model for each cluster. The method by which the classification standard learning unit 25 constructs the model is the same as the operation of the consumption model learning unit 14.
The classification standard learning unit 25 calculates the error that occurs in each cluster according to each model. For example, the classification standard learning unit 25 calculates the difference between the predicted energy consumption calculated from the model and the energy consumption amount of the learning data as an error. For example, the classification standard learning unit 25 calculates the average of the errors calculated from each learning data of the cluster as the error generated in the cluster. The classification standard learning unit 25 may calculate the error generated in the cluster by using the cross-validation method.

The classification standard learning unit 25 calculates an evaluation value according to an evaluation function from an error generated in each cluster. For example, the merit function is the sum of the errors.
The classification standard learning unit 25 changes the parameter N threshold value and the electricity cost threshold value within a predetermined range to perform the above operation.

The classification standard learning unit 25 acquires the parameter N threshold value and the electricity cost threshold value having the smallest evaluation value as the classification standard. That is, the classification standard learning unit 25 acquires the combination of the threshold values having the smallest error as the classification standard.
The classification standard learning unit 25 stores the acquired classification standard in the classification standard DB 13.

The classification standard learning unit 25 may set one or three or more conditions. Further, the classification standard learning unit 25 may set a hierarchy of different conditions (connection of conditions). Classification Criteria The conditions for the learning unit 25 to classify the learning data are not limited to a specific configuration.

Further, the classification standard learning unit 25 (second classification standard generation unit) may construct the classification standard by using the hierarchical cluster analysis.
FIG. 15 is a diagram for explaining a hierarchical cluster analysis.
The classification standard learning unit 25 calculates the distance between the learning data based on the explanatory variables of the learning data. The classification standard learning unit 25 sets learning data having a short distance as a cluster. For example, the distance between training data indicates the Euclidean distance using some explanatory variables. However, a general distance scale such as Mahalanobis distance may be used.

When the clusters are set, the classification standard learning unit 25 calculates the error that occurs in each cluster. The error calculation method is as described above.
The classification standard learning unit 25 calculates an evaluation value according to an evaluation function from an error generated in each cluster. For example, the merit function is the sum of the errors. The evaluation function may add the number of classifications (the number of clusters) as a penalty to the total error.

The classification standard learning unit 25 sets a cluster with different classification criteria. For example, the classification standard learning unit 25 sets clusters having a short distance as one cluster. The classification standard learning unit 25 calculates the evaluation value as described above.

The classification standard learning unit 25 repeats the above operation to acquire the classification standard of the cluster having the smallest evaluation value.
The classification standard learning unit 25 stores the acquired classification standard in the classification standard DB 13.

Next, an operation example of the classification standard learning unit 25 will be described.
FIG. 16 is a flowchart for explaining an operation example of the classification standard learning unit 25. Here, it is assumed that the learning data acquisition unit 11 has acquired the learning data.

First, the classification standard learning unit 25 substitutes infinity for the prediction error (S41). Substituting infinity for the prediction error, the classification standard learning unit 25 sets the classification standard (S42). When the classification criteria are set, the classification criteria learning unit 25 classifies the learning data into clusters according to the classification criteria set in S42 (S43).

When the learning data is classified into clusters, the classification standard learning unit 25 builds a model for each cluster (S44). The error that occurs in each model is calculated according to the model of each cluster (S45). When the error generated in each model is calculated, the classification standard learning unit 25 calculates the evaluation value from each error (S46).

When the evaluation value is calculated, the classification standard learning unit 25 determines whether the evaluation value is smaller than the prediction error (S47). When it is determined that the evaluation value is smaller than the prediction error (S47, YES), the classification standard learning unit 25 substitutes the evaluation value for the prediction error (S48). When the evaluation value is substituted for the prediction error, the classification standard learning unit 25 replaces the classification standard with the smallest evaluation value with the classification standard set in S42 (S48).

When it is determined that the evaluation value is not smaller than the prediction error (S47, NO) or when the classification standard is replaced (S49), the classification standard learning unit 25 determines whether there is another classification standard that can be set (S47, NO). S50).

When it is determined that there is another classification standard that can be set (S50, YES), the classification standard learning unit 25 returns to S42. The classification standard learning unit 25 sets different classification standards in S42.

When it is determined that there is no other configurable classification standard (S50, NO), the classification standard learning unit 25 stores the classification standard with the smallest evaluation value replaced by S49 in the classification standard DB 13 (S51). When the classification standard is stored, the classification standard learning unit 25 ends the operation.

The prediction server configured as described above classifies the prediction data into clusters according to the classification criteria, and calculates the error that occurs in each cluster according to the model of each cluster. The prediction server can construct a classification standard with a small error by calculating the error while changing the classification standard. As a result, the prediction server can set up a cluster that can effectively calculate the predicted energy consumption.

(Third Embodiment)
Next, the third embodiment will be described.
The prediction system 1 ″ according to the third embodiment is different from the first embodiment in that the application management server 50 connects to the vehicles V1 to Vn. Therefore, other configurations are designated by the same reference numerals and detailed description thereof will be omitted.

FIG. 17 is a block diagram showing a configuration example of the prediction system 1 ″ according to the third embodiment.
As shown in FIG. 17, the prediction system 1'' includes a prediction server 10, a weather information server 20, a traffic information server 30, an application management server 50'', vehicles V1 to Vn, and the like.

Vehicles V1 to Vn transmit consumption history information and prediction data to the application management server 50 ″. For example, the vehicles V1 to Vn include a processing unit, a storage unit, a communication unit, and the like. Vehicles V1 to Vn generate and store consumption history information. The vehicles V1 to Vn transmit the consumption history information to the application management server 50 ″ at a predetermined timing.

Further, the vehicles V1 to Vn transmit the prediction data to the application management server 50 ″ at the timing when the predicted energy consumption is required. For example, the vehicles V1 to Vn transmit the prediction data to the application management server 50 ″ according to the user's operation. For example, the vehicles V1 to Vn transmit the prediction data to the application management server 50 ″ together with a request requesting the predicted energy consumption.

The application management server 50'' includes a consumption history information DB 41 and a prediction data DB 51.
The application management server 50'' has a function of transmitting consumption history information stored in the consumption history information DB 41 to the prediction server 10 in addition to the function of the application management server 50. For example, the application management server 50'' transmits consumption history information to the prediction server 10 in response to a request from the prediction server 10. Further, the application management server 50'' may periodically (for example, daily) send consumption history information to the prediction server 10.

The prediction system 10 ″ may include the prediction server 10 ″ according to the second embodiment.

The prediction system configured as described above acquires consumption history information from the vehicle. As a result, the prediction system can acquire learning data without using the charger management server.

(Fourth Embodiment)
Next, the fourth embodiment will be described.
The prediction system 1 ″ according to the fourth embodiment is different from the prediction system 1 according to the first embodiment in that the prediction server 10 ″ ″ transmits a model to the user terminals F1 to Fn through the application management server 50. Therefore, other points are designated by the same reference numerals and detailed description thereof will be omitted.

The prediction system 1 "" according to the fourth embodiment includes a prediction server 10 "".

FIG. 18 is a block diagram showing a configuration example of the prediction server 10'''.
The prediction server 10'''is a learning data acquisition unit 11, a cluster classification unit 12, a classification standard DB 13, a consumption model learning unit 14, a model parameter DB 15, a prediction data acquisition unit 16, an energy consumption prediction unit 17, and road shape information. It includes a DB 19 and a model determination unit 26.

The model determination unit 26 determines a cluster of prediction data based on the prediction data. The model determination unit 26 determines a cluster of prediction data in the same manner as the cluster classification unit 12 based on the classification standard stored in the classification standard DB 13.
The model determination unit 26 acquires the determined cluster model from the model parameter DB 15.

The prediction data may be the minimum data necessary for determining the cluster.

The prediction server 10'''' transmits the model acquired by the model determination unit 26 to the application management server 50. The application management server 50 transmits the model to the user terminals F1 to Fn.
The user terminals F1 to Fn calculate the predicted energy consumption according to the model received from the application management server 50. For example, the user terminals F1 to Fn apply the data input by the user to the model to calculate the predicted energy consumption.

Further, the user terminals F1 to Fn may store the model and continue to use it. For example, the user terminals F1 to Fn may calculate the predicted energy consumption by using the model stored inside without requesting the model from the application management server 50 when calculating the predicted energy consumption.

The prediction system 1 "" may have the characteristics of the second embodiment or the third embodiment.

The prediction system configured as described above transmits the model to the user terminal. The user terminal calculates the predicted energy consumption according to the received model. Therefore, the prediction server does not need to acquire prediction data from the user. As a result, the prediction system can protect the privacy of the user.

The user terminal can continuously use the model. As a result, the prediction system reduces the frequency of processing the request from the user terminal, and can reduce the load on the prediction server.

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other embodiments, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are also included in the scope of the invention described in the claims and the equivalent scope thereof.
The inventions described in the claims at the time of filing the present application are described below.
[C1]
It is an energy consumption prediction device used in a prediction system that predicts the energy consumption of a vehicle using a model.
A learning data acquisition unit that acquires a plurality of learning data including battery information regarding a vehicle battery and motion information including energy consumption of the vehicle, and
A cluster classification unit that classifies learning data acquired by the learning data acquisition unit into a plurality of clusters according to classification criteria, and a cluster classification unit.
A model generation unit that generates a model that predicts the energy consumption of the vehicle in each cluster classified from the learning data by the cluster classification unit, and a model generation unit.
A forecasting data acquisition unit that acquires forecasting data for predicting the energy consumption of a vehicle,
A prediction unit that determines a cluster of prediction data acquired by the prediction data acquisition unit and predicts the energy consumption of the vehicle from the prediction data according to the determined cluster model generated by the model generation unit. ,
Energy consumption prediction device equipped with.
[C2]
The model generator
A determination unit that determines whether the number of training data in the cluster is equal to or greater than the threshold value,
When the determination unit determines that the number is equal to or greater than the threshold value, the first generation unit that generates a model of the cluster by regression based on the learning data of the cluster, and the first generation unit.
When the determination unit determines that the number is not equal to or greater than the threshold value, it includes a second generation unit that generates a model of the cluster based on an approximation model that is a model of an approximate cluster close to the cluster.
The energy consumption prediction device according to C1.
[C3]
The model generated by the model generation unit is composed of a plurality of parameters.
The second generation unit calculates the correction value by assuming that the parameters of the model of the cluster are the values obtained by integrating each correction value with each parameter of the approximate model and that each correction value is the same.
The energy consumption prediction device according to C2.
[C4]
The second generation unit uses the parameters of the cluster model as values obtained by adding each correction value to each parameter of the approximate model, and calculates the correction value using sparse estimation.
The energy consumption prediction device according to C2.
[C5]
The second generation unit calculates the selection reference value of each parameter and determines the correction value to be 0 based on the selection reference value.
The energy consumption prediction device according to C4.
[C6]
The second generation unit calculates the correction value on the assumption that at least a part of each correction value to be added to each parameter is 0.
The energy consumption prediction device according to C5.
[C7]
The second generation unit calculates each correction value calculated by the cross-validation method and an error between the predicted energy consumption and the energy consumption of the learning data, and is based on the correction value and the error. Calculate the selection reference value,
The energy consumption prediction device according to C5 or 6 above.
[C8]
The second generation unit uses each correction value calculated by the cross-validation method as the selection reference value based on the value divided by an error.
The energy consumption prediction device according to C7.
[C9]
The second generation unit generates a plurality of combinations of correction values assumed to be 0 based on the selection reference value, calculates an error of each combination, and determines a correction value assumed to be 0 based on the error. ,
The prediction device according to any one of C5 to 8 above.
[C10]
It comprises a first classification criterion generator that sets conditions and generates the classification criteria using non-hierarchical cluster analysis according to the conditions.
The energy consumption prediction device according to any one of C1 to 9 above.
[C11]
It includes a second classification standard generation unit that generates the classification standard by using the hierarchical cluster analysis based on the training data.
The energy consumption prediction device according to any one of C1 to 9 above.
[C12]
It is an energy consumption prediction method used in a prediction system that predicts the energy consumption of a vehicle using a model.
Acquire a plurality of learning data including battery information regarding the vehicle battery and motion information including the energy consumption of the vehicle.
A cluster classification unit that classifies the plurality of learning data into a plurality of clusters according to the classification criteria, and
A model for predicting the energy consumption of the vehicle in each cluster classified from the plurality of training data is generated.
Obtain forecasting data for predicting vehicle energy consumption,
A cluster of the prediction data is determined, and the energy consumption of the vehicle is predicted from the prediction data according to the determined model of the cluster.
Energy consumption prediction method.

1 (1', 1'', 1''') ... Prediction system, 10 (10', 10''') ... Prediction server, 11 ... Learning data acquisition unit, 12 ... Cluster classification unit, 13 ... Classification criteria DB, 14 ... consumption model learning unit, 15 ... model parameter DB, 16 ... prediction data acquisition unit, 17 ... energy consumption prediction unit, 18 ... energy consumption information DB, 19 ... road shape information, 20 ... weather information server, 21 ... Meteorological information DB, 25 ... Classification standard learning department, 26 ... Model determination department, 30 ... Traffic information server, 31 ... Traffic information DB, 40 ... Charger management server, 41 ... Consumption history information DB, 50 (50'') ... App management server, 51 ... Prediction data DB, 61 and 62 ... Lines, 71 to 75 ... Trials, 141 ... Model construction method classification unit, 142 ... Model parameter learning unit, 143 ... Minority data Model parameter learning unit, Q1 to Qn ... Charger, V1 to Vn ... Vehicle, F1 to Fn ... User terminal.

Claims (13)

A consumption energy prediction apparatus that to predict the energy consumption of the vehicle by using the model,
A learning data acquisition unit that acquires a plurality of learning data including battery information related to a vehicle battery and motion information including energy consumption of the vehicle.
A cluster classification unit that classifies learning data acquired by the learning data acquisition unit into a plurality of clusters according to classification criteria, and a cluster classification unit.
For each of the clusters that are classified Ri by the cluster classification unit, and the number is equal to or greater than a threshold value is either determination unit of the learning data included, if the number of the learning data by the determination unit is equal to or greater than the threshold value For the determined cluster, the first generation unit that generates a model that predicts the energy consumption of the vehicle having a plurality of parameters by regression based on the training data, and the determination unit have the number of the training data. For clusters determined not to be equal to or greater than the threshold, a model for predicting vehicle energy consumption is generated by an approximate model close to the cluster having a parameter obtained by integrating the same correction value into the model parameters for each cluster. A model generator including 2 generators ,
A forecasting data acquisition unit that acquires forecasting data for predicting energy consumption from a vehicle,
Wherein determining the cluster according to the classification criteria from the calculation data calculation data acquisition unit has acquired, consumed according to the model the model generating unit has generated corresponding to determine the constant has been said clusters, from the calculation data of the vehicle Predictor that predicts energy and
Energy consumption prediction device equipped with. It is an energy consumption prediction device that predicts the energy consumption of a vehicle using a model.
A learning data acquisition unit that acquires a plurality of learning data including battery information regarding a vehicle battery and motion information including energy consumption of the vehicle.
A cluster classification unit that classifies learning data acquired by the learning data acquisition unit into a plurality of clusters according to classification criteria, and a cluster classification unit.
For each cluster classified by the cluster classification unit, a determination unit for determining whether the number of training data included is equal to or greater than the threshold, and the determination unit for determining that the number of training data is equal to or greater than the threshold. For the clusters, the first generation unit that generates a model that predicts the energy consumption of a vehicle having a plurality of parameters by regression based on the training data, and the determination unit that the number of the training data is the threshold value. For the clusters determined not to be the above, the model of the approximate cluster close to the cluster having the parameters obtained by adding the values calculated by using the sparse estimation to each correction value is used to obtain the parameters of the model for each cluster. A model generator including a second generator that generates a model that predicts energy consumption,
A forecasting data acquisition unit that acquires forecasting data for predicting energy consumption from a vehicle,
A cluster is determined from the prediction data acquired by the prediction data acquisition unit according to the classification criteria, and the energy consumption of the vehicle is consumed from the prediction data according to the model generated by the model generation unit corresponding to the determined cluster. Prediction unit that predicts
Energy consumption prediction device equipped with. The second generating unit calculates the selection reference value for each parameter, to determine the compensation values based on the selected reference value,
The energy consumption prediction device according to claim 2 . The second generation unit calculates the correction value on the assumption that at least a part of each correction value to be added to each parameter is 0.
The energy consumption prediction device according to claim 3 . The second generation unit calculates the selection reference value based on the error between each correction value calculated by the cross-validation method and the predicted energy consumption and the energy consumption of the learning data.
The energy consumption prediction device according to claim 3 or 4 . The selection reference value of the second generation unit is calculated based on the value obtained by dividing each correction value calculated by the cross-validation method by an error.
The energy consumption prediction device according to claim 5 . The second generating unit, the selection reference value, the combination of 0 and assume correction value generates a plurality, determining the 0 assuming correction value based on error of each combination,
The prediction device according to any one of claims 3 to 6 . It is an energy consumption prediction device that predicts the energy consumption of a vehicle using a model.
A learning data acquisition unit that acquires a plurality of learning data including battery information regarding a vehicle battery and motion information including energy consumption of the vehicle.
A first classification criterion generator that sets conditions and generates classification criteria using non-hierarchical cluster analysis according to the conditions.
A cluster classification unit that classifies learning data acquired by the learning data acquisition unit into a plurality of clusters according to the classification criteria, and a cluster classification unit.
A model generation unit that generates a model that predicts the energy consumption of the vehicle for each cluster classified from the learning data by the cluster classification unit.
A forecasting data acquisition unit that acquires forecasting data for predicting energy consumption from a vehicle,
A cluster is determined from the prediction data acquired by the prediction data acquisition unit according to the classification criteria, and the energy consumption of the vehicle is consumed from the prediction data according to the model generated by the model generation unit corresponding to the determined cluster. Prediction unit that predicts
Energy consumption prediction device equipped with. It is an energy consumption prediction device that predicts the energy consumption of a vehicle using a model.
A learning data acquisition unit that acquires a plurality of learning data including battery information regarding a vehicle battery and motion information including energy consumption of the vehicle.
A second classification standard generation unit that generates classification criteria using the hierarchical cluster analysis based on the training data, and
A cluster classification unit that classifies learning data acquired by the learning data acquisition unit into a plurality of clusters according to the classification criteria, and a cluster classification unit.
A model generation unit that generates a model that predicts the energy consumption of the vehicle for each cluster classified by the cluster classification unit, and a model generation unit.
A forecasting data acquisition unit that acquires forecasting data for predicting energy consumption from a vehicle,
A cluster is determined from the prediction data acquired by the prediction data acquisition unit according to the classification criteria, and the energy consumption of the vehicle is consumed from the prediction data according to the model generated by the model generation unit corresponding to the determined cluster. Prediction unit that predicts
Energy consumption prediction device equipped with. By using the model to a consumption energy prediction how to predict the energy consumption of the vehicle,
The learning data acquisition unit acquires a plurality of learning data including battery information regarding the vehicle battery and motion information including the energy consumption of the vehicle.
The cluster classification unit classifies the acquired plurality of learning data into a plurality of clusters according to the classification criteria .
The determination unit determines whether the number of learning data contained in each cluster classified by the cluster classification unit is equal to or greater than the threshold value.
For the clusters whose number is determined to be equal to or greater than the threshold value in the first generation unit, a model for predicting the energy consumption of a vehicle having a plurality of parameters is generated by regression based on the learning data.
For clusters whose number is determined not to be equal to or greater than the threshold value in the second generation unit, a vehicle is used by an approximate model close to the cluster having a parameter obtained by integrating the same correction value with the parameter of the model for each cluster. Generate a model that predicts the energy consumption of
The forecasting data acquisition unit acquires forecasting data for predicting energy consumption from the vehicle.
The prediction unit determines a cluster from the prediction data acquired by the prediction data acquisition unit according to the classification criteria , and is generated by the first generation unit or the second generation unit corresponding to the determined cluster. according to the model to predict the energy consumption of the vehicle from the calculation data,
Energy consumption prediction method. It is an energy consumption prediction method that predicts the energy consumption of a vehicle using a model.
The learning data acquisition unit acquires a plurality of learning data including battery information regarding the vehicle battery and motion information including the energy consumption of the vehicle.
The cluster classification unit classifies the plurality of learning data into a plurality of clusters according to the classification criteria.
The determination unit determines whether the number of learning data contained in each cluster classified by the cluster classification unit is equal to or greater than the threshold value.
For the cluster in which the number of the learning data is determined by the determination unit to be equal to or greater than the threshold value in the first generation unit, the energy consumption of the vehicle having a plurality of parameters is determined by regression based on the learning data. Generate a model to predict and
For clusters in which the determination unit determines that the number of learning data is not equal to or greater than the threshold value in the second generation unit, a parameter obtained by adding a value calculated using sparse estimation to each correction value is required. Generate a model that predicts the energy consumption of the vehicle by the model of the approximate cluster close to the cluster,
The forecasting data acquisition unit acquires forecasting data for predicting energy consumption from the vehicle.
The prediction unit determines a cluster from the acquired prediction data according to the classification criteria, and the prediction data is based on a model generated by the first generation unit or the second generation unit corresponding to the determined cluster. Predict the energy consumption of the vehicle from
Energy consumption prediction method. It is an energy consumption prediction method that predicts the energy consumption of a vehicle using a model.
The learning data acquisition unit acquires a plurality of learning data including battery information regarding the vehicle battery and motion information including the energy consumption of the vehicle.
In the first classification standard generation unit, conditions are set, and classification criteria are generated using non-hierarchical cluster analysis according to the conditions.
The cluster classification unit classifies the plurality of learning data into a plurality of clusters according to the classification criteria.
The model generation unit generates a model that predicts the energy consumption of the vehicle for each cluster classified from the learning data by the cluster classification unit.
The forecasting data acquisition unit acquires forecasting data for predicting energy consumption from the vehicle.
The prediction unit determines a cluster from the prediction data acquired by the prediction data acquisition unit, and the vehicle is derived from the prediction data according to a model generated by the model generation unit corresponding to the determined model of the cluster. Predict the energy consumption of
Energy consumption prediction method. It is an energy consumption prediction method that predicts the energy consumption of a vehicle using a model.
The learning data acquisition unit acquires a plurality of learning data including battery information regarding the vehicle battery and motion information including the energy consumption of the vehicle.
In the second classification standard generation unit, the classification standard is generated by using the hierarchical cluster analysis based on the plurality of learning data.
The cluster classification unit classifies the plurality of learning data into a plurality of clusters according to the classification criteria.
The model generation unit generates a model that predicts the energy consumption of the vehicle for each cluster classified by the cluster classification unit.
The forecasting data acquisition unit acquires forecasting data for predicting energy consumption from the vehicle.
The prediction unit determines a cluster from the prediction data acquired by the prediction data acquisition unit according to the classification criteria, and the prediction data according to the model generated by the model generation unit corresponding to the determined model of the cluster. Predict the energy consumption of the vehicle from
Energy consumption prediction method.