Method and system for determining a distance-to-empty charge in an electric vehicle
Methods and Systems are provided for predicting a distance until the batter in an electric vehicle is discharged. Vehicle and geophysical data are received and formatted in n epochs over a time period T. For each epoch, an epoch data structure including a total energy consumption rate, a state of charge rate, and a distance rate is generated. The epochs are processed using a vector autoregression model to generate a predicted epoch data structure. A predicted state of charge is calculated using a state of charge model. A predicted distance traveled in the next epoch is calculated. A predicted charge to empty is calculated and checked to determine if the predicted charge to empty has reached an empty level. The process is repeated until the predicted charge to empty reaches an empty level. The distance to empty is then reported to the driver.
1 . A system for predicting a driving range of an electric vehicle power source comprising:
an on-board diagnostic (OBD) device configured to interface with an electric vehicle's engine control unit (ECU) via an OBD port to receive real-time vehicle data from a vehicle data management system, and stored electric vehicle data from an electric vehicle data storage system, where the vehicle data includes a timestamp, and for each timestamp, a current state of charge and a total energy consumption data set comprising at least one of: motor speed and torque data, input current and throttle data, or input current and voltage data, and where the vehicle data is collected over time for storage in the electric vehicle data storage system;
a distance-to-empty predictor comprising executable instructions stored as computer programs in a memory system configured to control electric vehicle battery management operations; and
a processor configured to execute the executable instructions of the computer programs of the distance-to-empty predictor to automatically control display of range information to a vehicle operator, where when executed, the distance-to-empty predictor:
receives a sample set of vehicle data collected over a sample time period, T, prior to a current time from the electric vehicle storage system via the OBD device interfacing with the ECU, and receives a current temperature measurement from a geophysical position sensor integrated with the OBD device;
automatically initializes in the memory system a predicted charge-to-empty to the state of charge at the current time, and a distance-to-empty to zero;
automatically constructs in the memory system a first set of n epoch data structures, D(Tn), corresponding to n epochs, tn, in the sample time period, T, with an epoch duration, T/n, each epoch data structure comprising a total energy calculated by the processor using specific vehicle sensor data during each of the n epochs, which is calculated using the total energy consumption data set in the vehicle data, and where the epoch speed is calculated by the processor based on a distance data obtained from the vehicle's odometer or GPS coordinates traveled over each epoch;
automatically generate in the memory system a predicted epoch data structure, D(Tn+1), including a predicted total energy consumed rate and a predicted epoch speed corresponding to a next epoch, tn+1, using the set of n epoch data structures, D(Tn), as inputs to a statistical model pre-trained with historical electric vehicle performance data to model a relationship between the total energy consumed rates and epoch speeds over a next epoch tn+1;
automatically generate using the processor a predicted state of charge rate for the next epoch, tn+1, by inputting the temperature measurement and the predicted total energy consumption rate to a state of charge rate prediction model trained on electric vehicle battery discharge characteristics;
automatically calculate using the processor a state of charge consumption by multiplying the predicted state of charge rate for the next epoch, tn+1, by the epoch duration;
automatically calculate using the processor a predicted next epoch distance based on the predicted epoch speed and the epoch duration;
automatically update in the memory system the distance-to-empty by adding the predicted next epoch distance to a current value of the distance-to-empty;
automatically update in the memory system the predicted charge-to-empty by subtracting the state of charge consumption from a current value of the predicted charge-to-empty;
automatically control a vehicle user interface display to report to a user interface the distance-to-empty when the predicted charge-to-empty is less than or equal to an EMPTY_VALUE; and
when the predicted charge-to-empty is not less than or equal to EMPTY_VALUE:
automatically construct in the memory system a next set of n data structures including the predicted data structure,
D(Tn+1), and the most recent n−1 epoch data structures in the previous set of n epoch data structures and repeat generating a predicted epoch data structure for a next epoch, generating the predicted state of charge rate for a next epoch, estimating the state of charge consumption for the next epoch, calculating the predicted next epoch distance, and updating the distance-to-empty and predicted charge-to-empty until the predicted charge-to-empty is less than or equal to EMPTY_VALUE.
2 . The system of claim 1 , where the total energy consumption data set received as vehicle data includes a torque and a motor speed, and the calculation of the total energy consumption during each of the n epochs includes calculating a total energy consumption at each timestamp of the vehicle data received using:
TEC(i)=speed(i)×torque(i), where i=a timestamp index of the vehicle data; and where
the calculation of the total energy consumed rate equal to the total energy consumed during each of the n epochs, t n , comprises:
TEC
(
t
n
)
=
∑
i
=
1
t
TEC
(
i
)
×
timedelta
,
where timedelta=time elapsed between i and i−1, and t is the number of timestamps in the sample time interval, T.
3 . The system of claim 1 , where the total energy consumption data set received as vehicle data includes a input current and a throttle, and the calculation of the total energy consumption during each of the n epochs includes calculating a total energy consumption at each timestamp of the vehicle data received using:
TEC(i)=input current(i)×throttle(i), where i=a timestamp index of the vehicle data; and
where the calculation of the total energy consumed rate equal to the total energy consumed during each of the n epochs, t n , comprises:
TEC
(
t
n
)
=
∑
i
=
1
t
TEC
(
i
)
×
timedelta
,
where timedelta=time elapsed between i and i−1, and t is the number of timestamps in the sample time interval, T.
4 . The system of claim 1 , where the total energy consumption data set received as vehicle data includes an input current and an input voltage, and the calculation of the total energy consumption during each of the n epochs includes calculating a total energy consumption at each timestamp of the vehicle data received using:
TEC(i)=input current(i)×input voltage(i), where i=a timestamp index of the vehicle data; and
where the calculation of the total energy consumed rate equal to the total energy consumed during each of the n epochs, t n , comprises:
TEC
(
t
n
)
=
∑
i
=
1
t
TEC
(
i
)
×
timedelta
,
where timedelta=time elapsed between i and i−1, and t is the number of timestamps in the sample time interval, T.
5 . The system of claim 1 where the statistical model pre-trained to model the relationship between the total energy consumed rates and epoch speeds over the next epoch includes a vector autoregression model.
6 . The system of claim 1 where the distance to empty predictor is configured to, after generating the predicted data structure:
filter the predicted data set, D(T n+1 ), using an outlier rejection model.
7 . The system of claim 6 where the outlier rejection model includes a Multivariate Conditional Distribution Outlier Rejection (MCDOR) model.
8 . The system of claim 1 where the distance to empty predictor is configured to, in generating the predicted state of charge rate, the state of charge rate prediction model includes a deep learning model.
9 . The system of claim 8 where the deep learning model is a feed-forward neural network.
10 . The system of claim 1 where the vehicle data includes odometer readings at the timestamp, and the distance to empty predictor is further configured to:
use the odometer readings to determine the distance traveled between each timestamp of the vehicle data;
adding the distance traveled between each timestamp to determine the distance traveled over each epoch; and
use the geolocation data in determining the temperature value at the current location.
11 . The system of claim 1 where the distance to empty predictor is further configured to:
use geolocation data to determine the distance traveled between each timestamp of the geolocation data;
adding the distance traveled between each timestamp to determine the distance traveled over each epoch.
12 . A method of predicting a driving range of an electric vehicle comprising:
receiving, via an on-board diagnostic (OBD) device interfacing with an electric vehicle's engine control unit (ECU), a sample set of vehicle data from a vehicle data management system collected over a sample time period, T, prior to a current time from a electric vehicle storage system, and a current temperature measurement from a geophysical position sensor;
automatically initializing, by a processor in a memory system, a predicted charge-to-empty to the state of charge at the current time, and a distance-to-empty to zero;
automatically constructing, by the processor in the memory system, a first set of n data structures, D(T n ), corresponding to n epochs t n , in the sample time period, T, with an epoch duration, T/n, each data structure comprising a total energy consumed rate and an epoch speed, where the total energy consumed rate is calculated using specific vehicle sensor data during each of the n epochs, which is calculated using the total energy consumption data set in the vehicle data, and where the epoch speed is based on a distance traveled over each epoch;
automatically generating, by the processor, a predicted data structure, D(T n+1 ), including a predicted total energy rate and a predicted epoch speed corresponding to a next epoch, t n+1 , using the set of n data structures, D(T n ), as inputs to a statistical model pre-trained with historical electric vehicle performance data to model a relationship between the total energy consumed rates and epoch speeds over a next epoch t n+1 ;
automatically generating, by the processor, a predicted state of charge rate (soc rate) for the next epoch, t n+1 , by inputting the temperature measurement and the predicted total energy consumption rate to a state of charge rate prediction model trained on electric vehicle battery discharge characteristics;
automatically computing, by the processor, a state of charge consumption by obtaining a predicted state of charge rate for the next epoch, t n+1 , and multiplying the epoch duration by the predicted state of charge rate;
automatically calculating, by the processor, a predicted next epoch distance based on the adjusted predicted epoch speed and the epoch duration;
updating, by the processor, the distance-to-empty by adding the predicted next epoch distance to a current value of the distance-to-empty;
automatically updating, by the processor in the memory system, the predicted charge-to-empty by subtracting the state of charge consumption from a current value of the predicted charge-to-empty;
automatically controlling, by the processor, a vehicle user interface display to report the distance-to-empty when the predicted charge-to-empty is less than or equal to an EMPTY_VALUE; and when the predicted charge-to-empty is not less than or equal to EMPTY_VALUE:
automatically constructing, by the processor in the memory system, a next set of n data structures including the predicted data structure, D(T n+1 ), and the most recent n−1 data structures in the previous set of n data structures and repeat generating a predicted data structure for a next epoch, generating the predicted state of charge rate for a next epoch, estimating the state of charge consumption for the next epoch, calculating the predicted next epoch distance, and updating the distance-to-empty and predicted charge-to-empty until the predicted charge-to-empty is less than or equal to EMPTY_VALUE.
13 . The method of claim 12 , where the total energy consumption data set received as vehicle data includes a torque and a motor speed, and the calculation of the total energy consumption during each of the n epochs includes calculating a total energy consumption at each timestamp of the vehicle data received using:
TEC(i)=speed(i)×torque(i), where i=a timestamp index of the vehicle data; and where the calculation of the total energy consumed rate equal to the total energy consumed during each of the n epochs, t n , comprises:
TEC
(
t
n
)
=
∑
i
=
1
t
TEC
(
i
)
×
timedelta
,
where timedelta=time elapsed between i and i−1, and t is the number of timestamps in the sample time interval, T.
14 . The method of claim 12 , where the total energy consumption data set received as vehicle data includes a input current and a throttle, and the calculation of the total energy consumption during each of the n epochs includes calculating a total energy consumption at each timestamp of the vehicle data received using:
TEC(i)=input current(i)×throttle(i), where i=a timestamp index of the vehicle data; and
where the calculation of the total energy consumed rate equal to the total energy consumed during each of the n epochs, t n , comprises:
TEC
(
t
n
)
=
∑
i
=
1
t
TEC
(
i
)
×
timedelta
,
where timedelta=time elapsed between i and i−1, and t is the number of timestamps in the sample time interval, T.
15 . The method of claim 12 , where the total energy consumption data set received as vehicle data includes an input current and an input voltage, and the calculation of the total energy consumption during each of the n epochs includes calculating a total energy consumption at each timestamp of the vehicle data received using:
TEC(i)=input current(i)×input voltage(i), where i=a timestamp index of the vehicle data; and
where the calculation of the total energy consumed rate equal to the total energy consumed during each of the n epochs, t n , comprises:
TEC
(
t
n
)
=
∑
i
=
1
t
TEC
(
i
)
×
timedelta
,
where timedelta=time elapsed between i and i−1, and t is the number of timestamps in the sample time interval, T.
16 . The method of claim 12 where in the step of generating the predicted data structure, the statistical model pre-trained to model the relationship between the total energy consumed rates and epoch speeds over the next epoch includes a vector autoregression model.
17 . The method of claim 12 where the distance to empty predictor is configured to perform the step of:
filtering the predicted data set, D(T n+1 ), using an outlier rejection model, after generating the predicted data structure.
18 . The method of claim 17 where the outlier rejection model includes a Multivariate Conditional Distribution Outlier Rejection (MCDOR) model.
19 . The method of claim 12 where, in generating the predicted state of charge rate, the state of charge rate prediction model includes a deep learning model.
20 . The method of claim 19 where the deep learning model is a feed-forward neural network.
21 . The method of claim 12 where the vehicle data includes odometer readings at the timestamp, the method comprising:
using the odometer readings to determine the distance traveled between each timestamp of the vehicle data;
adding the distance traveled between each timestamp to determine the distance traveled over each epoch; and
using the geolocation data in determining the temperature value at the current location.
22 . The method of claim 12 further comprising:
using geolocation data to determine the distance traveled between each timestamp of the geolocation data; and
adding the distance traveled between each timestamp to determine the distance traveled over each epoch.