Systems and methods for electric vehicle smart charging

Abstract

Methods and systems for determining charging profile information to charge an energy storage system of an electric vehicle are provided. The method includes determining the charging profile information based, at least in part, on historical vehicle operation data for the electric vehicle.

Description

SYSTEMS AND METHODS FOR ELECTRIC VEHICLE SMART CHARGINGTECHNICAL FIELD

[0001] This disclosure relates to methods and apparatus for smart charging of electric vehicles.BACKGROUND

[0002] Some battery-powered devices, such as electric vehicles, manage their available energy by estimating quantities that are indicative of the current state of their energy storage system (e.g., batteries). Example quantities that are often used in assessing available energy in a battery- powered device include State of Health (SOH), State of Charge (SOC), State of Power (SOP), and State of Temperature (SOT). Charging operations for the energy storage system in an electric vehicle are conventionally manually managed by drivers who determine when and how to charge the batteries. Drivers, who are typically unfamiliar with the effects of battery charging on battery degradation, often charge the vehicle’s battery pack as soon as possible, as much as possible (e.g., up to 100% SOC), and as quickly as possible (e.g., using maximum charging power).SUMMARY

[0003] Energy storage systems (e.g., batteries or battery packs) are one of the most expensive components of an electric vehicle to replace. Accordingly, maintaining the health of the batteries to extend their useful life may be an important consideration for drivers. Battery charging operations, as they are typically performed using existing technology, may severely degrade the SOH of the battery, resulting in a shorter usable life. For instance, due to range anxiety, drivers tend to charge their vehicle’s batteries whenever charging is available and the batteries are typically charged to their full capacity (e.g., 100% SOC) during each charging session. Additionally, when an electric vehicle is connected to a charging station, the battery is typically charged with the maximum allowed power determined by the vehicle’s battery management system (BMS) and the charging station type (e.g., Level 1, Level 2, DC fast charging, etc.). The inventors have recognized and appreciated that such charging strategies may result in severe battery SOH degradation due to the high charging current rates and permanence of the battery at high SOC levels. In some embodiments, conventional charging operations may be improved bytailoring charging operations (also referred to herein as “smart charging”) based on information associated with a driver’s driving and / or charging habits. For instance, if a driver typically charges their electric vehicle overnight at their home and also at their workplace, rather than charging the batteries to 100% SOC at their workplace, using the techniques described herein, the battery may be charged to a lesser amount (e.g., 60% SOC) at their workplace, which may provide sufficient charge for the driver to return home while reducing battery SOH degradation. Additionally, when charging the batteries overnight (e.g., over a period of up to 12 hours) at home, rather than charging the batteries at the maximum allowed power for the first 2-3 hours and then resting for the remaining time (e.g., by having the car plugged into the charger but not charging), some embodiments of the present disclosure allow for slower charging over a longer period of time, which may reduce battery SOH degradation. Other example scenarios for using smart charging to reduce battery degradation in accordance with some embodiments of the present disclosure are provided in more detail below.

[0004] In some embodiments, a computer-implemented method is provided. The computer- implemented method includes determining charging profile information to charge an energy storage system of an electric vehicle, the charging profile information being based, at least in part, on historical vehicle operation data for the electric vehicle.

[0005] In one aspect, the historical vehicle operation data includes historical driving data and / or historical charging data for the electric vehicle. In another aspect, the charging profile information includes a target electric power. In another aspect, the method further includes determining based, at least in part, on the historical vehicle operation data, one or more of energy needs for the electric vehicle, an expected charging time, or a target state of charge or the energy storage system, and the charging profile information is determined based, at least in part, on the energy needs, the expected charging time, and / or the target state of charge. In another aspect, determining the energy needs for the electric vehicle comprises using a first statistical model to determine the energy needs for the electric vehicle. In another aspect, determining the expected charging time comprises using a second statistical model to determine the expected charging time. In another aspect, the method further includes discretizing the historical vehicle operation data into time segments, and associating with each of the time segments, state information for the energy storage system and an average electric power applied to the energy storage system to generate annotated time segments, wherein the energy needs for the electric vehicle and / or theexpected charging time is determined based on the annotated time segments. In another aspect, the method further includes determining, based on the state information in the annotated time segments, state transition probabilities for each of the time segments, and the energy needs for the electric vehicle and / or the expected charging time is determined based on the state transition probabilities. In another aspect, the method further includes receiving event data for a driver of the electric vehicle, and modifying the state transition probabilities based, at least in part, on the received event data, wherein the energy needs for the electric vehicle and / or the expected charging time is determined based on the modified state transition probabilities. In another aspect, the method further includes determining, based on the annotated time segments, probability density functions for each of a plurality of energy storage system variables, and the energy needs for the electric vehicle and / or the expected charging time is determined based on the probability density functions. In another aspect, the energy needs for the electric vehicle and / or the expected charging time is determined based on an integration of at least two of the probability density functions.

[0006] In another aspect, determining energy needs for the electric vehicle includes determining based, at least in part, on the historical vehicle operation data, a probability of driving the electric vehicle again on a particular day, and determining the energy needs based, at least in part, on the probability of driving the electric vehicle again on the particular day. In another aspect, determining the energy needs based, at least in part, on the probability of driving the electric vehicle again on the particular day includes determining based, at least in part, on the historical vehicle operation data, an expected energy consumption for the electric vehicle for the particular day, and determining the energy needs as the expected energy consumption for the electric vehicle for the particular day when the probability of driving the electric vehicle again is greater than a threshold value. In another aspect, determining the energy needs based, at least in part, on the probability of driving the electric vehicle again on the particular day includes determining based, at least in part, on the historical vehicle operation data, an expected energy consumption for the electric vehicle for a next day after the particular day, and determining the energy needs based, at least in part, on the expected energy consumption for the electric vehicle for the next day when the probability of driving the electric vehicle again is less than a threshold value. In another aspect, the method further includes determining based, at least in part, on the historical vehicle operation data, an expected energy consumption for the electric vehicle for theparticular day, and determining the energy needs is further based, at least in part, on the expected energy consumption for the electric vehicle for the particular day when the probability of driving the electric vehicle again is less than the threshold value.

[0007] In another aspect, the method further includes determining a target state of charge for an energy storage system of the electric vehicle based, at least in part, on the energy needs, wherein determining the charging profde information is based, at least in part, on the target state of charge. In another aspect, determining the target state of charge for the energy storage system is further based, at least in part, on a minimum state of charge. In another aspect, determining charging profile information comprises determining the charging profile information using a reinforcement learning technique or a dynamic programming technique. In another aspect, determining charging profile information using a reinforcement learning technique comprises determining charging profile information using a model trained based on the historical vehicle operation data using the reinforcement learning technique. In another aspect, training the model based on the historical vehicle operation data comprises training the model based on a reward function having a magnitude that is inversely proportional to battery pack degradation obtained as a result of supplying a particular power to the energy storage system during a charging session. In another aspect, determining charging profile information further includes determining, based on statistical information extracted from the historical vehicle operation data, a target state of charge for an energy storage system of the electric vehicle and an expected charging time for a charging session, and determining, using the reinforcement learning technique or the dynamic programming technique, a charging profile that achieves the target state of charge given the expected charging time while minimizing battery pack degradation. In another aspect, the reinforcement learning technique or the dynamic programming technique is further configured to minimize an energy cost during charging.

[0008] In another aspect, the method further includes receiving an indication that the electric vehicle is coupled to a charging station, and determining the charging profile information in response to receiving the indication that the electric vehicle is coupled to the charging station. In another aspect, the method further includes charging the energy storage system of the electric vehicle based on the charging profile information. In another aspect, determining the charging profile information is further based, at least in part, an energy cost associated with charging the energy storage system.

[0009] In some embodiments, a system is provided. The system includes at least one computer processor, and at least one computer-readable medium having stored thereon instructions which, when executed, program the at least one computer processor to perform any of the methods described herein.

[0010] In some embodiments, at least one computer-readable medium is provided. The at least one computer-readable medium having stored thereon instructions which, when executed, program at least one computer processor to perform any of the methods described herein.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] FIG. 1 schematically illustrates a charging architecture for charging an energy storage system of an electric vehicle, in accordance with some embodiments of the present disclosure.

[0012] FIG. 2 is a flow chart for charging an energy storage system of an electric vehicle, in accordance with some embodiments of the present disclosure.

[0013] FIG. 3 is a flow chart of extracting statistical information based on historical vehicle operation data, in accordance with some embodiments of the present disclosure.

[0014] FIG. 4 shows plots of example historical vehicle operation data that may be used to determine charging profile information for a charging session, in accordance with some embodiments of the present disclosure.

[0015] FIG. 5 shows plots of example discretized historical vehicle operation data that may be used to determine charging profile information for a charging session, in accordance with some embodiments of the present disclosure.

[0016] FIG. 6A illustrates a graphical representation of state transition probabilities that may be determined based on historical vehicle operation data, in accordance with some embodiments of the present disclosure.

[0017] FIGS. 6B and 6C illustrate probability density functions for average power associated with operation of a battery pack for two different electric vehicles.

[0018] FIG. 7 is a flowchart of a process for updating state transition probabilities based on event data, in accordance with some embodiments of the present disclosure.

[0019] FIG. 8 illustrates a Gaussian Mixture Model probability density function (PDF) determined based on historical vehicle operation data, in accordance with some embodiments of the present disclosure.

[0020] FIG. 9A illustrates a conditional PDF determined based on historical vehicle operation data, in accordance with some embodiments of the present disclosure.

[0021] FIG. 9B illustrates that expected charging time may be determined based on the conditional PDF shown in FIG. 9A.

[0022] FIGS. 9C and 9D illustrate example PDFs determined based on historical vehicle operation data, in accordance with some embodiments of the present disclosure.

[0023] FIG. 10A is a flow chart of an example rule-based technique for determining charging profde information for a charging session of a battery pack, in accordance with some embodiments of the present disclosure.

[0024] FIG. 10B shows results of applying a rule-based technique to simulated data, in accordance with some embodiments of the present disclosure.

[0025] FIG. 11 A schematically illustrates a reinforcement learning technique for determining charging profde information for a charging session of a battery pack, in accordance with some embodiments of the present disclosure.

[0026] FIG. 1 IB illustrates an architecture for combining rule-based control and reinforcement learning / dynamic programming control, in accordance with some embodiments of the present disclosure.

[0027] FIG. 12 schematically illustrates a computing architecture on which some embodiments of the present disclosure may be implemented.DETAILED DESCRIPTION

[0028] Devices (e.g., electric vehicles) that incorporate energy storage systems (e.g., battery packs) typically include a Battery Management System (BMS), which monitors and controls individual cells or groups of cells within the battery pack. When an electric vehicle is plugged into a charging station, the BMS in the electric vehicle and electronic circuitry in the charging station may communicate to exchange information about how to charge the energy storage system. For instance, the BMS may indicate a maximum charging power to the charging station that should be used to charge the battery pack. The BMS may determine the maximum charging power based on one or more characteristics of the battery pack (e.g., temperature, current state of charge, manufacturer, etc.). For instance, if the battery pack is cold, limiting the maximum powerused to charge the battery may prevent severe degradation of the battery, even if the charging station is able to provide a higher power output.

[0029] Due to range anxiety, drivers often charge their electric vehicle as much as possible, as soon as possible, and as fast as possible to minimize the risk that the battery in their vehicle will run out of charge before they reach the next destination at which the vehicle can be charged. However, it has been demonstrated that charging a battery pack in this way results in faster SOH degradation for the battery pack, resulting in a shorter useful battery life. As described herein, conventional charging strategies typically only take into consideration the maximum power that may be used to charge the battery pack and are not configured to determine a charging strategy that limits battery pack SOH degradation based on other factors.

[0030] The inventors have recognized and appreciated that conventional charging strategies may be improved by taking into consideration historical vehicle operation data that reflects, for example, a driver’s driving and / or charging habits. To this end, some embodiments of the present disclosure relate to techniques for using historical vehicle operation data to determine a charging strategy for a charging session (e.g., a target electric power and / or a target charging time for charging the battery pack during the charging session) that reduces battery SOH degradation.

[0031] FIG. 1 illustrates a system 100 in accordance with some embodiments of the present disclosure. System 100 includes electric vehicle 110 and charging station 120. Electric vehicle 110 may include a battery pack 112, as an energy storage system for the vehicle, and a battery management system (BMS) 114, configured to manage charge and discharge operations of the battery pack 112. Electric vehicle 110 also includes electric interface 116 that, when electrically coupled to a charging station (e.g., charging station 120) enables the battery pack 112 to be charged during a charging session. Charging station 120 may include control electronics 122 and electric interface 124 configured to be electrically coupled to an electric interface (e.g., electric interface 116) of an electric vehicle. Control electronics 122 may be configured to control the amount of power output from power source 126 and provided to electric vehicle 110 via electric interface 124 during a charging session of the electric vehicle.

[0032] In response to electrically coupling the electric interface 116 and the electric interface 124 (e.g., by plugging an electrical connector associated with electric interface 124 into a charging port associated with electric interface 116), communication may be exchanged between electric vehicle 110 and charging station 120 to configure one or more aspects of the chargingsession. For instance, BMS 114 may be configured to determine a maximum power that may be used to charge the battery, and a signal may be transmitted from electric vehicle 110 to charging station 120 indicating the maximum power. BMS 114 may be configured to determine the maximum power based on one or more characteristics of battery pack 112. For instance, the maximum power may be determined based, at least in part, on a temperature of the battery pack 112, a current state of charge (SOC) of the battery pack 112, or any other suitable information. In response to receiving the maximum power indication, control electronics 122 may be configured to control power source 126 to charge battery pack 112 of electric vehicle 110 based, at least in part, on the maximum power. For instance, charging station 120 may proceed to charge battery pack 112 at the maximum power provided that the charging station 120 is able to provide such power.

[0033] As described herein, the inventors have recognized and appreciated that always charging a battery pack of an electric vehicle at the maximum power, as determined by the BMS of the vehicle up to the maximum state of charge (e.g., 100% SOC), may result in battery SOH degradation that may be avoided if the battery pack was charged in a different manner. Such “smart charging” of the battery pack may extend its useful life. For instance, charging the battery pack at a power less than the maximum allowable power over a longer period of time and / or charging the battery pack to a state of charge (e.g., 80% SOC) less than its maximum state of charge may be sufficient for the energy needs of the vehicle prior to a next charging session and may result in less battery SOH degradation if compared to fast charging the battery pack to the maximum SOC achievable.

[0034] Some embodiments of the present disclosure relate to determining charging profile information (e.g., a target charging power and / or a target SOC) for a charging session based, at least in part, on historical vehicle operation data associated with an electric vehicle. In some embodiments, the historical vehicle operation data may describe how the vehicle has been previously operated and / or charged, and the data may be used to inform a charging strategy for future charging sessions that reduces SOH degradation. For example, as described in more detail below, some embodiments determine energy needs for a battery pack of an electric vehicle and an expected charging time for the charging session based, at least in part, on the historical vehicle operation data. The energy needs and the expected charging time can then be used to determine charging profile information to use during the charging session. In other embodiments,the historical vehicle operation data may be used to train a Reinforcement Learning agent (e.g., a machine learning model) that learns to optimally interact with different simulated charging scenarios (e.g., using a reinforcement learning technique). The trained model may then be used to determine the charging profile information to use during a charging session. In some embodiments, a combination of techniques including rule-based algorithms, dynamic programming, and / or trained reinforcement learning agents (e.g., trained using reinforcement learning techniques) may be used to determine charging profile information for a charging session.

[0035] As shown in FIG. 1, system 100 may further include network (e.g., cloud-based) resources 130. Network resources 130 may be communicatively coupled (wired or wirelessly) to electric vehicle 110 and / or charging station 120. Network resources 130 may include one or more hardware computer processors 132. In some embodiments, processor(s) 132 may be programmed to implement one or more of the methods described herein. In some embodiments, one or more of the methods described herein may be implemented using one or more components (e.g., BMS 114) of electric vehicle 110. In other embodiments, one or more of the methods described herein may be implemented using one or more components (e.g., control electronics 122) of charging station 120. In yet further embodiments, one or more of the methods described herein may be implemented using a combination of processing resources in any two or more of electric vehicle 110, charging station 120 and network resources 130.

[0036] FIG. 2 illustrates a process 200 for charging an energy storage system (e.g., a battery pack) of an electric vehicle in accordance with some embodiments of the present disclosure. Process 200 may begin in act 210, where an indication that an electric vehicle is coupled to a charging station is received. As described above, in conventional charging scenarios, when an electric vehicle is plugged into a charging station, the BMS of the electric vehicle may be configured to communicate information about the maximum power and / or the minimum power at which the charging station may charge the battery pack of the electric vehicle during the charging session. In some embodiments, an indication that an electric vehicle is coupled to a charging station may be used to determine that a new charging session is desired and charging profile information for the charging session should be determined. It should be appreciated however, that at least some of the techniques described herein (e.g., training of statistical models,etc.) may be performed prior to initiation of a charging session, and embodiments are not limited in this respect.

[0037] Process 200 may then proceed to act 212, where charging profile information for a charging session may be determined based, at least in part, on historical vehicle operation data. As described herein, rather than simply charging the battery pack of an electric vehicle to maximum capacity as fast as possible, which, if performed repeatedly, may result in substantial SOH degradation of the battery pack, some embodiments of the present disclosure relate to techniques for determining a charging profile for a charging session that takes into consideration information about the habits of the driver. For instance, the expected energy needs of the vehicle before a next charging session, an expected duration of the charging session, and / or other factors (e.g., energy cost) may be taken into consideration when determining how to charge a battery pack during a charging session, while minimizing or otherwise reducing the amount of SOH degradation of the battery pack. Further details on determining charging profile information according to various embodiments of the present disclosure are described in more detail below.

[0038] Process 200 may then proceed to act 214, where the energy storage system (e.g., battery pack) of the electric vehicle is charged based on the charging profile information determined in act 212. For instance, the charging profile information may include energy needs of 70% SOC and an expected charging session duration of 6 hours. Rather than charging the battery pack of the electric vehicle to 100% SOC as fast as possible (using maximum power) after plugging the electric vehicle into a charging station, as would typically occur in conventional charging scenarios, the battery pack of the electric vehicle may be charged using a lower power over a longer portion of the expected charging session until the battery pack reaches 70% SOC. By charging the battery pack slower and to a lower SOC, the extent of battery pack SOH degradation may be reduced relative to the typical charging scenario. This effect may be particularly evident over several charging cycles.

[0039] Some embodiments relate to techniques for collecting and processing data on habitual driving and / or charging behavior (also referred to herein as “historical vehicle operation data”) of an electric vehicle. The processed data may be used to determine charging profile information (e.g., charging power) for a charging session. In some embodiments, captured historical vehicle operation data may be discretized, and statistical information extracted from the discretized datamay be used to train one or more statistical models. The one or more statistical models may be used to determine the charging profile information.

[0040] FIG. 3 illustrates a process 300 for processing historical vehicle operation data using statistical techniques in accordance with some embodiments of the present disclosure. Process 300 begins in act 310, where historical vehicle operation data is received. For instance, the BMS of the electric vehicle may continuously or periodically collect operation data for the electric vehicle. The operation data may include, but is not limited to, state of charge of the battery pack over time, mileage of the vehicle over time, and state of the battery pack (e.g., charging, resting, discharging). FIG. 4 illustrates an example of historical vehicle operation data for an electric vehicle, in accordance with some embodiments. In the example shown in FIG. 4, the historical vehicle operation data includes state information, electric power applied to the battery pack of the vehicle and state of charge of the battery pack over time. For instance, as can be observed, when the battery pack is discharged (Pelec is positive), the SOC of the battery pack decreases, and when the battery pack is charged (Pelec is negative), the SOC of the battery pack increases.

[0041] Process 300 may then proceed to act 312, where the received historical vehicle operation data is discretized into time segments. In some embodiments, the historical vehicle operation data may be discretized into equal length time segments (e.g., 1 minute, 10 minute, 20 minute, 30 minute, or 1 hour time segments). In some embodiments, the historical vehicle operation data may be discretized into 10 minute time segments, resulting in 144 time segments per day. Discretizing the historical vehicle operation data into equal length time segments is also referred to herein as providing a “bin view” of the data. Alternatively, the historical vehicle operation data may be discretized into variable length time segments based on events, such as the state of the electric vehicle (e.g., charge, discharge, rest). Discretizing the historical vehicle operation data into equal length time segments is also referred to herein as providing a “scene view” of the data. It should be appreciated that the historical vehicle operation data may be transformed from bin view to scene view or vice versa depending on the type of statistical information desired to be extracted from the data.

[0042] Process 300 may then proceed to act 314, where state information and an average electric power value is associated with each of the time segments (also referred to herein as “time bins”). In the case of the historical vehicle operation data being represented in bin view, in someembodiments, the vehicle system state for each time segment may be described by the following data fields:Table 1: Example fields for bin view of historical vehicle operation data

[0043] In the case of the historical vehicle operation data being represented in scene view, in some embodiments, the vehicle system state for each time segment may be described by the following data fields:Table 2: Example fields for scene view of historical vehicle operation data

[0044] FIG. 5 illustrates example historical vehicle operation data that has been discretized, in accordance with some embodiments. In FIG. 5, each plot shows the vehicle system state for each of a plurality of time segments throughout a day for one of three states - discharge, charge, and rest. As can be observed from the data shown in FIG. 5, the driver’s habits as they operate the vehicle throughout each day of the week can be discerned. For instance, on weekdays (Mon - Fri), the driver has a morning discharge session (e.g., driving to work) and an evening discharge session (e.g., driving home from work). The battery pack of the vehicle may be charged after the morning discharge session and / or charged overnight after the evening discharge system. As may be expected, the vehicle system state information is different on the weekend days compared tothe weekdays. For example, on the weekend, the electric vehicle experiences longer discharge sessions later in the day, with the vehicle being in a rest state for most of the day.

[0045] Returning to process 300, after the historical vehicle operation data has been discretized into time bins, and each time bin has been associated with state information and electric power information, process 300 may proceed to act 316, wherein state transition probabilities for each of the time bins in bin view (e.g., each of the 144 10 minute time bins for each day) may be determined. As described above, in some embodiments each time bin may be associated with one of three states - discharge, charge, or rest. State transition probabilities represent the probability that the vehicle will transition from its current state to a next state (e.g., discharge —> rest; discharge — > discharge; charge — > rest; rest —> charge, etc.). In some embodiments, state transition probabilities may be represented as Markov Chains. FIG. 6A schematically illustrates state transition probabilities between the three states - charge 610, discharge 620, and rest 630. In the example shown in FIG. 6A, which represents the state transitions for a single time segment, the probability of a state transition charge —> charge is 80%, rest —> rest is 100%, discharge — discharge is 60%, charge — discharge is 20%, discharge — charge is 30%, discharge — rest is 10% and all other probabilities are 0%. In some embodiments, the state transition probabilities may be represented as a 3x3 matrix of values. For example, the state transition probabilities shown in FIG. 6A may be represented as follows:Table 3: Example 3x3 matrix of state transition probabilities

[0046] In some embodiments, a 3x3 matrix of state transition probabilities may be associated with the other fields shown in Table 1 for each of the plurality of time segments of the historical vehicle operation data. In some embodiments, the state transition probabilities determined based on the historical vehicle operation data may be modified based on event data for a driver of the electric vehicle. Modifying the state transition probabilities may, for example, enable a more accurate determination of metrics such as energy needs and / or expected charging time, which may be used to determine charging profile information in accordance with some embodiments.

[0047] FIG. 7 illustrates a process 700 for modifying state transition probabilities based, at least in part, on driver event data. Process 700 may begin in act 710, where event data associated with a driver of an electric vehicle is received. In some embodiments, the event data may be derived from calendar information for a driver. For instance, the calendar information may indicate one or more events in the driver’s schedule (e.g., a meeting in the middle of the day to which the driver is likely to drive) that may impact the state transition probabilities. Process 700 may then proceed to act 712, where driving time from the origin to the destination may be determined based on the event information. For example, if the event information indicates that the driver has a dentist appointment at a location 15 miles from their office, the driving time may be determined based on the distance to the appointment (e.g., 15 miles) and the typical driving time along the route between the driver’s office and appointment location. In some embodiments, additional information (e.g., traffic information) may be used to determine the driving time. Process 700 may then proceed to act 714, where the time bins corresponding to the likely driving time may be determined based on the event information. As described above, in some embodiments a 3x3 matrix of state transition probabilities may be associated with each of a plurality of time segments to represent the probabilities of state transitions for individual time bins. Accordingly, the event time and the estimated driving time may be used in act 714 to determine which time bins to modify the corresponding state transition probabilities. For instance, if the event data indicates that the driver has an appointment at 4:00 pm for one hour and the estimated driving time is 30 minutes, the time bins identified in act 714 may correspond to those between 3:30-4:00 pm and / or between 5:00-5:30 pm. Process 700 may then proceed to act 716, where the state transition probabilities may be updated for the time bins identified in act 714.

[0048] Continuing with the example above, time bin 93 may correspond to 3:30 pm, and may have the original state transition probabilities as follows:In response to determining based on event data for the driver that the state transition probabilities should be changed for this time bin, the modified state transition probabilities may be set as follows:

[0049] As also shown in the process of FIG. 3, after the historical vehicle operation data has been discretized into time segments in act 312, and each time segment has been associated with state information and electric power information in act 314, process 300 may proceed to act 318, wherein probability density functions (PDFs) are determined for each of a plurality of variables. In some embodiments, vehicle power consumption can be modeled as a stochastic process with a Gaussian model obtained from the historical vehicle operation data. For a Gaussian model, the PDF can be obtained by setting the mean (e.g., expected value) and variance. FIGS. 6B and 6C illustrate example data for average power consumption for two different electric vehicles. In some embodiments, for each weekday and time segment included in the historical vehicle operation data, a Gaussian model may be used to model the vehicle average power consumption by fitting the mean and variance to the data. By fitting a vehicle average power consumption model considering combinations of weekdays and time segments, the habitual power consumptions associated with driving certain routes repeated over time can be modeled.

[0050] In some embodiments, PDFs are used to make statistical considerations on the occurrence probability of certain events. The determined PDFs may take into consideration one or more of joint probabilities (P(A, B) = P(A|B) ■ P(B) and P(A, B) = P(B, A)) marginal probabilities (PQ4) =P(A, b , where bi is the probability of an event for one random variable irrespective of the outcome of another random variable) and conditional probabilities (P(A|B) = P(A B ). Non-limiting examples of PDFs that may be fit to discretized historical vehicle operation data, in accordance with some embodiments, include charge start time segment, discharge start time segment, rest start time segment, charge total time, discharge total time, rest total time,charge energy consumption, discharge energy consumption, charge start time segment + charge total time, charge total time | charge start time segment, discharge start time segment + discharge total time, discharge total time | discharge start time segment, rest start time segment + rest total time, rest total time | rest start time segment, charge start time segment + charge energy consumption, charge energy consumption | charge start time segment, discharge start time segment + discharge energy consumption, and discharge energy consumption | discharge start time segment. It should be appreciated that PDFs may be fit for any other combination of variables.

[0051] In some embodiments, PDFs are determined from Gaussian Mixture Models (GMMs) fit on the discretized historical vehicle operation data through an optimization process, where the number of Gaussians fit to the data is optimized to achieve a best fit. In this way, some embodiments fit a final Gaussian model as a weighted average of a certain (e.g., optimized) number of simple Gaussian models to provide a best fit to the discretized historical vehicle operation data. FIG. 8 schematically illustrates a PDF determined from a GMM in accordance with some embodiments. In the example PDF shown in FIG. 8 the x-axis represents 144 time bins (each corresponding to 10 minutes of data) for a single day. Within each time bin, the count from the data is shown. In the example of FIG. 8, the counts are based on the number of data points within the historical vehicle operation data in which the driver started driving in that time bin (i.e., discharge start time segment). A particular number of Gaussian models (e.g., four Gaussian models in the example of FIG. 8) are fit to model the stochastic process. For example, among the four Gaussian models are a first Gaussian model 810 and a second Gaussian model 812. The final PDF 820 shown in FIG. 8 may be determined as a weighted combination of the four individually-fit Gaussian models (including Gaussian models 810 and 812). In some embodiments, the particular number of Gaussian models used to model the stochastic process may be determined using an optimization process.

[0052] The computation of the GMM for conditional probabilities is nontrivial and may be estimated in some embodiments using Bayes Theorem or appropriate known formulas. FIG. 9A illustrates a conditional PDF of charge duration (in time bins) given charge operation starting at a certain time bin. As shown in FIG. 9B, such a PDF may be used to estimate an expected (e.g. most probable) charging time for a charging session according to the following formula:E ChTime) = timebin ■ PDF CHTime\ChStartBin), where E ChTime) is the expected value function £(d) = ' ai■ P(ai'). As shown in FIG. 9A, one can observe that if the driver starts charging the vehicle at around time bin 11 (6:00 pm), most likely with a high probability, the expected charging time will be about 80 time bins (~13 hours). It should be appreciated that PDFs may be used to derive other expected metrics based on the historical vehicle operation data. FIGS. 9C and 9D illustrate further examples of PDFs that may be determined in accordance with the techniques described herein. FIGS. 9C and 9D illustrate PDFs representing the average amount of energy consumed by a battery pack of an electric vehicle based on the discharge start bin for different driving profiles.

[0053] Returning to process 300 in FIG. 3, after the state transition probabilities are determined in act 316 and the PDFs are determined in act 318, process 300 may proceed to act 320, where one or more statistical models are trained based on the statistical information determined in acts 316 and 318. For instance, a first statistical model for determining the energy needs of an electric vehicle (e.g., a vehicle average power consumption model) and a second statistical model for determining expected charging time for a charging session may be trained based, at least in part, on the determined statistical information. Process 300 may then proceed to act 322, where the trained statistical model(s) are output for use in determining charging profile information for a charging session, examples of which are described in more detail below.

[0054] In some embodiments, charging profile information for a charging session is determined using a rule-based approach. For instance, an expected charging time and an expected energy consumption may be used to determine a “just in time” type of charging strategy, where the energy charged in the battery pack reflects the energy needs for expected future vehicle operations. FIG. 10A illustrates a process 1030 for determining charging profile information for a charging session (e.g., a target charging power), in accordance with some embodiments. Process 1030 may begin in act 1040, where the probability of driving again today may be determined. For instance, in some embodiments, the probability of driving again today may be determined based, at least in part, on the state transition probabilities and / or the PDFs determined for one or more time bins.

[0055] Process 1030 may then proceed to act 1042, where it is determined whether the probability of driving again today is greater than a threshold value. When it is determined in act 1042 that the probability of driving again today is greater than the threshold value, process 1030may proceed to act 1044, where the energy needs for the electric vehicle is set equal to the expected energy consumption for today. When it is determined in act 1042 that the probability of driving again today is less than the threshold value, process 1030 may proceed to act 1046, where the energy needs for the electric vehicle is set equal to the expected energy consumption for today and / or the next day. For instance if the charging session start time is in the evening at a time when the driver typically does not drive again for a particular day (e.g., determined based on a statistical analysis of historical vehicle operation data), the energy needs for the vehicle may be determined as the expected energy consumption for the next day. As described above, in some embodiments, the expected energy consumption for a particular day (e.g., a current day, a next day, etc.) may be determined using statistical information extracted from historical vehicle operation data.

[0056] After determining the energy needs for the electric vehicle in act 1044 or act 1046, process 1030 may proceed to act 1048, where a target state of charge for the battery pack of the electric vehicle is determined. As discussed herein, rather than fully charging the battery pack to 100% SOC during every charging session which may result in unnecessary SOH degradation, some embodiments of the present disclosure may determine a charging strategy that charges the battery pack to less than 100% SOC, with the target SOC being aligned with the expected energy needs of the battery until the next charging session. For instance, if the charging session is occurring in the middle of the day at a driver’s workplace, the energy needs may correspond to the energy needed to drive the vehicle home (possibly with some buffer).

[0057] The inventors have recognized and appreciated that setting the target SOC within a range having a lower limit (e.g., 10% SOC) and an upper limit (e.g., 90% SOC), provided that at least the upper limit satisfies the vehicle’s expected energy needs may result in less SOH degradation of the battery pack. Accordingly, in some embodiments, a lower limit (e.g., 10% SOC) representing the minimum SOC 1050 may be taken into consideration when determining the target SOC in act 1048. Process 1030 may then proceed to act 1052, where charging profile information (e.g., target electric power) may be determined based, at least in part, on the target SOC determined in act 1048. In some embodiments, the target electric power may additionally be determined in act 1052 based, at least in part, on an expected charging time for the charging session to further reduce potential SOH degradation by charging the battery pack only as fast as is necessary to achieve the target SOC given the expected charging time. In some embodiments,the target electric power may further be determined based, at least in part, on a level of confidence associated with the statistical model outputs (e.g., for energy needs and expected charging time). For instance, when the level of confidence is low, the target electric power may be increased to improve the likelihood that the electric vehicle will be sufficiently charged to satisfy the energy needs. By contrast, when the level of confidence is high, the target electric power may stay the same or decrease from an initial determination to further prevent unnecessary SOH degradation of the battery pack. It should be appreciated that the level of confidence need not to be binary, but instead may include more than two values (including being a continuum) such that different actions may be taken to appropriately adjust the target electric power based on the corresponding level of confidence.

[0058] FIG. 10B illustrates an example of results obtained by applying rule-based control logic as described herein for a period of one year (January 2020 - January 2021) to synthetically generated data for a driver. As shown in FIG. 10B, the habitual driving pattern for the driver includes driving to work in the morning and returning home in the evening following a defined routine each day. By computing an estimation of how much the vehicle will be available for charge when coupled to a charging station (e.g. most probable charging time) and future energy consumption (e.g., most probable energy consumption), the rule-based control logic is able to keep the battery at low SOC levels (e.g., between 30%-40% SOC) and charge the vehicle at reduced charging current rates allowing for reduced SOH degradation over the course of the year.

[0059] In some embodiments, charging profile information for a charging session is determined using a machine-learning approach. In particular, some embodiments use reinforcement learning to determine the charging profile information. Reinforcement learning is a type of machine learning paradigm where an agent learns to make decisions by interacting with an environment. During training, the agent receives feedback in the form of rewards or punishments as it takes actions within the environment, and the objective of the training is to learn a strategy or policy that maximizes the cumulative reward over time.

[0060] FIG. 11 A schematically illustrates a reinforcement learning process 1100 that may be used to train an agent that, when trained, may determine charging profile information, in accordance with some embodiments of the present disclosure. As shown in FIG. 11 A, an agent 1110 (e.g., a control algorithm for deciding how much power should be supplied at each instantto the battery pack to allow for normal vehicle operation) may be provided with information on an environment 1120 (e.g., a vehicle model for driving according to a certain utilization pattern as described, for example, using state transition probabilities and / or power consumption distributions, such as PDFs determined from historical vehicle operation data and a battery pack model that provides degradation information based, for example, on battery SOC and charging rate sensitivities). During training, the agent 1110 may take various actions At(e.g., providing different amounts of power to the battery pack during charging sessions) within the environment 1120 described by the historical vehicle operation data, with the result being a reward Rt assigned based on a reward function. The actions At that the agent can make are based on a policy, which is a strategy for mapping states St to actions At, where a state St is a representation of the current situation or configuration of the environment 1120 that the agent 1110 observes at time t. In the context of battery charging, the policy may map a certain battery state to the electric power supplied to charge the battery.

[0061] In some embodiments, the reward Rt may be based on a function with a magnitude inversely proportional to the battery pack degradation obtained as a result of performing a chosen action At. For example, the reward function Rt may be represented by the following mathematical relationship: Rt= 1 — k(SOHt_^ — SOHf2. Additionally, different values can be assigned to the reward function based on certain actions to be discouraged or preferred. For example, the reward function may be negative if the SOC, as a result of the action, is outside of a certain valid range (e.g., 10%-90%) to discourage the agent from allowing too low or two high of an SOC. As another example, the reward function may be negative if the agent tries to charge the battery when the vehicle is either in the discharge state (e.g., during driving) or the resting state (e.g., not coupled to a charging station). As the agent 1110 takes various actions At and receives various rewards Rt, the policy used by the agent to map states St to actions At may be updated such that the agent learns behaviors that reinforce the objective.

[0062] Although the example objective function described above minimizes SOH degradation, it should be appreciated that other objective functions may alternatively be used, and / or different reward functions may be implemented to minimize those objective functions. For instance, some embodiments may employ reward functions that take into account energy cost for a charging session. The price of energy for a charging session may be determined by the amount of energy provided by the charging station and the price of energy (e.g., per kWh) determined by theenergy provider. The amount of energy provided by the charging station may be dependent on the target energy requested and the efficiency of the energy transfer, which may be a function of the battery SOC and the electric charging power. The price of the energy (e.g., per kWh) may depend on the demand and offer of energy at a certain time band during the day. In some embodiments, models of such energy cost phenomena may be accounted for in the definition of the reward function used to train an agent by updating its policy during reinforcement learning.

[0063] In some embodiments, a combination of rule-based techniques, machine learning techniques (e.g., reinforcement learning techniques), or dynamic programming may be used to determine charging profile information. For example, in some embodiments, a rule-based approach may be used to determine a target SOC for a charging session and an expected duration of the charging session, whereas the control strategy used to achieve the target SOC given the expected charging session duration, while minimizing various factors such as battery pack degradation and / or cost of energy supplied may be determined using reinforcement learning or dynamic programming techniques. By performing at least some of the computation using rulebased techniques, the overall amount of computation needed to determine the charging profile information may be reduced relative to using approaches that rely solely on machine learning techniques such as reinforcement learning.

[0064] FIG. 1 IB schematically illustrates a logical architecture 1130 for combining rule-based control and reinforcement learning (RL)Zdynamic programming, in accordance with some embodiments of the present disclosure. Architecture 1130 includes data importer 1132 configured to receive vehicle operation data and discretize the data into bins to generate bin data 1134. Architecture 1130 also includes distribution extractor 1136, which may determine statistical information (e.g., Markov chains, PDFs) from the bin data 1134. The statistical information may be provided as input to rule-based controller 1138 configured to determine charging information (e.g., current SOC, target SOC, expected charging time) using rule-based logic, examples of which are described herein. The charging information output from rule-based controller 1138 may be provided as input to optimal controller 1140 configured to determine a charging strategy for a charging session using reinforcement learning and / or dynamic programming techniques, examples of which are described herein. As shown in FIG. 1 IB, optimal controller 1140 may be configured to determine the charging strategy based on information from an energy cost model 1142 and a cost function definition 1144. For example,the energy cost model 1142 may be an energy cost model for predicting an energy cost typical for a certain area or energy provider. Examples of cost function definition 1144 include, but are not limited to, a cost function definition for SOH and / or a cost function definition for energy cost.

[0065] Using architecture 1130 in accordance with some embodiments may results in some advantages relative to using either rule-based or reinforcement learning techniques in isolation. For example, combining statistical (rule-based) models and optimal control (Reinforcement Learning or Dynamic Programming) may result in a simpler optimal problem to solve (e.g., determining a charging operation instead of a charging schedule for the whole week). As another example, one or more federated agents that can learn in a centralized or distributed way (e.g., charging process given SOC and expected charging time common to all vehicles in the same area with similar energy costs and same battery type and model) may be trained. As another example, use of architecture 1130 may allow for integration with event data (e.g., driver calendar information) through rule-based logic and may allow for considering charging cost and energy trading through the optimal controller.

[0066] The machine learning models described herein may be trained using various learning techniques. In some embodiments, a centralized learning approach may be used to train the model(s). In such a centralized learning approach, the training data may be gathered and processed at a centralized location (e.g., a single server or a powerful computing infrastructure). In this approach, all training data is available at a central entity, which trains the machine learning model. In some embodiments, a federated learning approach may be used to train the model(s). In such a federated learning approach, the machine learning model may be trained across multiple edge devices or local servers holding data samples. Instead of sending raw data to a central server (e.g., in the centralized learning approach), model updates in the federated learning approach may be computed locally, and only the model parameters may be transmitted centrally. Such a federate learning approach may allow for privacy preservation, reduced communication overhead, and the ability to learn from distributed datasets without exposing sensitive information. In some embodiments, a gossip learning approach may be used to train the model(s). In such a gossip learning approach, each node in a network collaborates by exchanging model updates with a subset of its neighbors. The exchanged information "gossips" through the network, allowing each node to aggregate insights from multiple sources. Gossip learning canenhance scalability, fault tolerance, and adaptability in distributed systems, while reducing the need for centralized coordination.

[0067] FIG. 12 shows, schematically, an illustrative computer 1000 on which any aspect of the present disclosure may be implemented. In the example of FIG. 12, the computer 1000 includes a processing unit 1001 having one or more computer hardware processors and one or more articles of manufacture comprising at least one non-transitory computer-readable medium (e.g., a memory 1002 that may include, for example, volatile and / or non-volatile memory). The memory 1002 may store one or more instructions to program the processing unit 1001 to perform any of the functionalities described herein. The computer 1000 may also include other types of non- transitory computer-readable media, such as a storage 1005 (e.g., one or more disk drives) in addition to the memory 1002. The storage 1005 may also store one or more application programs and / or resources used by application programs (e.g., software libraries), which may be loaded into the memory 1002. Thus, the memory 1002 and / or the storage 1005 may serve as one or more non-transitory computer-readable media storing instructions for execution by the processing unit 1001.

[0068] The computer 1000 may have one or more input devices and / or output devices, such as devices 1006 and 1007 illustrated in FIG. 12. These devices may be used, for instance, to present a user interface. Examples of output devices that may be used to provide a user interface include printers, display screens, and other devices for visual output, speakers and other devices for audible output, braille displays and other devices for haptic output, etc. Examples of input devices that may be used for a user interface include keyboards, pointing devices (e.g., mice, touch pads, and digitizing tablets), microphones, etc. For instance, the input devices 1007 may include a microphone for capturing audio signals, and the output devices 1006 may include a display screen for visually rendering, and / or a speaker for audibly rendering, recognized text.

[0069] In the example of FIG. 12, the computer 1000 also includes one or more network interfaces (e.g., a network interface 1010) to enable communication via various networks (e.g., a network 1020). Examples of networks include local area networks (e.g., an enterprise network), wide area networks (e.g., the Internet), etc. Such networks may be based on any suitable technology operating according to any suitable protocol, and may include wireless networks and / or wired networks (e.g., fiber optic networks).

[0070] Having thus described several aspects of at least one embodiment, it is to be appreciated that various alterations, modifications, and improvements will readily occur to those skilled in the art. Such alterations, modifications, and improvements are intended to be within the spirit and scope of the present disclosure. Accordingly, the foregoing descriptions and drawings are by way of example only.

[0071] The above-described embodiments of the present disclosure can be implemented in any of numerous ways. For example, the embodiments may be implemented using hardware, software, or a combination thereof. When implemented in software, the software code may be executed on any suitable processor or collection of processors, whether provided in a single computer, or distributed among multiple computers.

[0072] Also, the various methods or processes outlined herein may be coded as software that is executable on one or more processors running any one of a variety of operating systems or platforms. Such software may be written using any of a number of suitable programming languages and / or programming tools, including scripting languages and / or scripting tools. In some instances, such software may be compiled as executable machine language code or intermediate code that is executed on a framework or virtual machine. Additionally, or alternatively, such software may be interpreted.

[0073] The techniques described herein may be embodied as a non-transitory computer- readable medium (or multiple such computer-readable media) (e.g., a computer memory, one or more floppy discs, compact discs, optical discs, magnetic tapes, flash memories, circuit configurations in Field Programmable Gate Arrays or other semiconductor devices, or other non- transitory, tangible computer-readable medium) encoded with one or more programs that, when executed on one or more processors, perform methods that implement the various embodiments of the present disclosure described above. The computer-readable medium or media may be portable, such that the program or programs stored thereon may be loaded onto one or more different computers or other processors to implement various aspects of the present disclosure as described above.

[0074] The terms “program” or “software” are used herein to refer to any type of computer code or set of computer-executable instructions that may be employed to program one or more processors to implement various aspects of the present disclosure as described above. Moreover, it should be appreciated that according to one aspect of this embodiment, one or more computerprograms that, when executed, perform methods of the present disclosure need not reside on a single computer or processor, but may be distributed in a modular fashion amongst a number of different computers or processors to implement various aspects of the present disclosure.

[0075] Computer-executable instructions may be in many forms, such as program modules, executed by one or more computers or other devices. Program modules may include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Functionalities of the program modules may be combined or distributed as desired in various embodiments.

[0076] Also, data structures may be stored in computer-readable media in any suitable form. For simplicity of illustration, data structures may be shown to have fields that are related through location in the data structure. Such relationships may likewise be achieved by assigning storage for the fields to locations in a computer-readable medium, so that the locations convey how the fields are related. However, any suitable mechanism may be used to relate information in fields of a data structure, including through the use of pointers, tags, and / or other mechanisms that establish how the data elements are related.

[0077] Various features and aspects of the present disclosure may be used alone, in any combination of two or more, or in a variety of arrangements not specifically described in the foregoing, and are therefore not limited to the details and arrangement of components set forth in the foregoing description or illustrated in the drawings. For example, aspects described in one embodiment may be combined in any manner with aspects described in other embodiments.

[0078] Also, the techniques described herein may be embodied as methods, of which examples have been provided. The acts performed as part of a method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed in an order different from illustrated, which may include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments.

[0079] Use of ordinal terms such as “first,” “second,” “third,” etc. in the claims to modify a claim element does not by itself connote any priority, precedence, or order of one claim element over another or the temporal order in which acts of a method are performed, but are used merely as labels to distinguish one claim element having a certain name from another element having the same name (but for use of the ordinal term) to distinguish the claim elements.

[0080] Also, the phraseology and terminology used herein is for the purpose of description and should not be regarded as limiting. The use of “including,” “comprising,” “having,” “containing,” “involving,” “based on,” “according to,” “encoding,” and variations thereof herein, is meant to encompass the items listed thereafter and equivalents thereof as well as additional items.

Claims

CLAIMS1. A computer-implemented method, comprising: determining charging profile information to charge an energy storage system of an electric vehicle, the charging profile information being based, at least in part, on historical vehicle operation data for the electric vehicle.

2. The method of claim 1, wherein the historical vehicle operation data includes historical driving data and / or historical charging data for the electric vehicle.

3. The method of claim 1, wherein the charging profile information includes a target electric power.

4. The method of claim 1, further comprising: determining based, at least in part, on the historical vehicle operation data, one or more of energy needs for the electric vehicle, an expected charging time, or a target state of charge or the energy storage system, wherein the charging profile information is determined based, at least in part, on the energy needs, the expected charging time, and / or the target state of charge.

5. The method of claim 4, wherein determining the energy needs for the electric vehicle comprises using a first statistical model to determine the energy needs for the electric vehicle.

6. The method of claim 5, wherein determining the expected charging time comprises using a second statistical model to determine the expected charging time.

7. The method of claim 6, further comprising: discretizing the historical vehicle operation data into time segments; and associating with each of the time segments, state information for the energy storage system and an average electric power applied to the energy storage system to generate annotated time segments,wherein the energy needs for the electric vehicle and / or the expected charging time is determined based on the annotated time segments.

8. The method of claim 7, further comprising: determining, based on the state information in the annotated time segments, state transition probabilities for each of the time segments, wherein the energy needs for the electric vehicle and / or the expected charging time is determined based on the state transition probabilities.

9. The method of claim 8, further comprising: receiving event data for a driver of the electric vehicle; and modifying the state transition probabilities based, at least in part, on the received event data, wherein the energy needs for the electric vehicle and / or the expected charging time is determined based on the modified state transition probabilities.

10. The method of claim 7, further comprising: determining, based on the annotated time segments, probability density functions for each of a plurality of energy storage system variables, wherein the energy needs for the electric vehicle and / or the expected charging time is determined based on the probability density functions.

11. The method of claim 10, wherein the energy needs for the electric vehicle and / or the expected charging time is determined based on an integration of at least two of the probability density functions.

12. The method of claim 4, wherein determining energy needs for the electric vehicle comprises: determining based, at least in part, on the historical vehicle operation data, a probability of driving the electric vehicle again on a particular day; anddetermining the energy needs based, at least in part, on the probability of driving the electric vehicle again on the particular day.

13. The method of claim 12, wherein determining the energy needs based, at least in part, on the probability of driving the electric vehicle again on the particular day comprises: determining based, at least in part, on the historical vehicle operation data, an expected energy consumption for the electric vehicle for the particular day; and determining the energy needs as the expected energy consumption for the electric vehicle for the particular day when the probability of driving the electric vehicle again is greater than a threshold value.

14. The method of claim 12, wherein determining the energy needs based, at least in part, on the probability of driving the electric vehicle again on the particular day comprises: determining based, at least in part, on the historical vehicle operation data, an expected energy consumption for the electric vehicle for a next day after the particular day; and determining the energy needs based, at least in part, on the expected energy consumption for the electric vehicle for the next day when the probability of driving the electric vehicle again is less than a threshold value.

15. The method of claim 14, further comprising: determining based, at least in part, on the historical vehicle operation data, an expected energy consumption for the electric vehicle for the particular day; wherein determining the energy needs is further based, at least in part, on the expected energy consumption for the electric vehicle for the particular day when the probability of driving the electric vehicle again is less than the threshold value.

16. The method of claim 12, further comprising determining a target state of charge for an energy storage system of the electric vehicle based, at least in part, on the energy needs, wherein determining the charging profile information is based, at least in part, on the target state of charge.

17. The method of claim 16, wherein determining the target state of charge for the energy storage system is further based, at least in part, on a minimum state of charge.

18. The method of claim 1, wherein determining charging profile information comprises determining the charging profile information using a reinforcement learning technique or a dynamic programming technique.

19. The method of claim 18, wherein determining charging profile information using a reinforcement learning technique comprises determining charging profile information using a model trained based on the historical vehicle operation data using the reinforcement learning technique.

20. The method of claim 19, wherein training the model based on the historical vehicle operation data comprises training the model based on a reward function having a magnitude that is inversely proportional to battery pack degradation obtained as a result of supplying a particular power to the energy storage system during a charging session.

21. The method of claim 18, wherein determining charging profile information further comprises: determining, based on statistical information extracted from the historical vehicle operation data, a target state of charge for an energy storage system of the electric vehicle and an expected charging time for a charging session; and determining, using the reinforcement learning technique or the dynamic programming technique, a charging profile that achieves the target state of charge given the expected charging time while minimizing battery pack degradation.

22. The method of claim 21, wherein the reinforcement learning technique or the dynamic programming technique is further configured to minimize an energy cost during charging.

23. The method of claim 1, further comprising: receiving an indication that the electric vehicle is coupled to a charging station; anddetermining the charging profile information in response to receiving the indication that the electric vehicle is coupled to the charging station.

24. The method of claim 23, further comprising: charging the energy storage system of the electric vehicle based on the charging profile information.

25. The method of claim 1, wherein determining the charging profile information is further based, at least in part, an energy cost associated with charging the energy storage system.

26. A system comprising: at least one computer processor; and at least one computer-readable medium having stored thereon instructions which, when executed, program the at least one computer processor to perform the method of any of claims 1- 25.

27. At least one computer-readable medium having stored thereon instructions which, when executed, program at least one computer processor to perform the method of any of claims 1-25.

Images