Predicting energy usage in bev route navigation

Abstract

Aspects of the present invention relate to methods (400) for predicting energy use of a battery electric vehicle (100), BEV, to navigate a route, a method comprising obtaining (402) a first trained ma

Description

TECHNICAL FIELD The present disclosure relates to a method and apparatus for predicting energy usage in battery electric vehicles, BEVs, for route navigation. In particular, the present disclosure relates to a methods for predicting energy use of a BEV to navigate a route using one or more machine learning models trained to predict energy usage by different sub-systems of the BEV. BACKGROUND Accurate prediction of available range is a common concern for drivers of battery electric vehicles (BEVs), often identified as “range anxiety”. Accurate predictions enable drivers to plan their trips effectively, ensuring that they can reach their destinations without running out of charge. The ability to provide an accurate prediction of range at the beginning of a journey may help to provide confidence to drivers that they will be able to complete the journey and / or arrive with sufficient state of charge of the traction battery to allow for any further journey. In contrast, providing an initial prediction that is subject to significant revisions during the journey may create significant uncertainty in the mind of the driver, increasing the experience of “range anxiety” resulting in a poor experience for the driver. Modelling energy use of a vehicle may help to improve the accuracy of range predictions. However, many factors that have a significant impact on attainable range for a given state of charge (SoC) of the traction battery may be external to the vehicle. Furthermore, models may struggle to take account of all potential factors that affect energy usage due to limited memory and processing available on the vehicle. It is an aim of the present invention to address one or more of the disadvantages associated with the prior art. SUMMARY OF THE INVENTION Aspects and embodiments of the invention provide methods, as claimed in the appended claims. According to an aspect of the invention there is provided a method of predicting energy use of a battery electric vehicle, BEV, to navigate a route, using one or more machine learning models trained to predict energy usage by different sub-systems of the BEV while navigating the route and based on one or more operating parameters of the BEV. According to an aspect of the invention there is provided a method of predicting energy use of a battery electric vehicle, BEV, while traversing a route, the method comprising obtaining a first trained machine learning model, the first trained machine learning model trained based on data received from a plurality of vehicles to predict an amount of energy used by a drivetrain of a BEV based on one or more operating parameters of the BEV, obtaining an intended route to be taken by the BEV and determining, for each of a plurality of segments of the intended route, one or more operating parameters for that segment for the BEV, for each segment of the plurality of segments, inputting the one or more predicted operating parameters for that segment to the first trained machine learning model to generate a predicted amount of energy used by the drivetrain of the BEV forthat segment, and calculating a total energy use for the intended route based on the predicted amount of energy used for each segment of the plurality of segments. Advantageously, a machine learning algorithm can be trained based on real-world data received from a fleet comprising a large number of vehicles to predict energy requirements for driving a BEV. The data received from the fleet of vehicles can be segmented into comparable segments to create a very large data set at a desired level of granularity. Using a machine learning approach allows a large number of input parameters that can affect energy use to be considered and a more accurate total energy use value to be predicted. Advantageously, the method may be used to evaluate different potential routes to a desired destination and a “lowest energy cost” route selected for the current conditions / configuration of the vehicle. For example, a most efficient route between two points may change dependent on a prevailing wind direction such that a longer route that avoids driving into a headwind may become a more efficient option due to current weather conditions. In embodiments, the one or more operating parameters for a segment comprise one or more of: a vehicle speed; an indication of traffic conditions forthe segment; a road elevation orgradient; a road curvature; a road surface type; a wind direction; a wind speed; an ambient temperature and a precipitation status. Advantageously, parameters that may be used by the machine learning algorithm to predict energy use may cover a wide range of environmental and road conditions that may have a significant effect on energy use. In a specific example, weather forecasting information for a segment can be obtained allowing the algorithm may take into account wind speed and direction relative to a direction of travel of the vehicle for a segment to predict energy use in the presence of a headwind / tailwind / crosswind. In embodiments, the operating parameters further comprise one or more parameters relating to a vehicle configuration. Advantageously, by applying a machine learning algorithm trained on a large number of vehicles, the effect of differences in vehicle configuration on the energy use to drive a segment of a route can be predicted, for example the machine learning algorithm may use vehicle model or trim level as an input and predict energy use accordingly, allowing for more accurate and adaptive energy use predictions. In embodiments, the one or more parameters relating to the vehicle configuration comprise one or more of: a vehicle model; a vehicle weight; a tyre pressure; an indication that the vehicle is towing a trailer; and a wheel size. Advantageously, the method is able to accurately account for a range of possible changes in configuration of the vehicle including predicting the effect of carrying a heavy load or towing a trailer on the energy used while driving an intended route. In embodiments, the method comprises generating the plurality of segments of the intended route by subdividing the intended route into a plurality of portions having a predefined length. Advantageously, subdividing routes into segments of equal length facilitates easier comparison between segments and allowing a route of any length to be predicted by the system by accumulating the energy use over each segment. In embodiments, the method further comprises obtaining a second trained machine learning model, the second trained machine learning model trained based on data received from the plurality of vehicles to predict a temperature associated with a traction battery of a BEV based on an amount of energy supplied by the traction battery, for each segment of the plurality of segments: inputting the predicted amount of energy for that segment to the second trained machine learning model to generate a prediction of the temperature associated with the traction battery forthat segment; and based on the predicted battery temperature and the predicted amount of energy, calculating a battery energy loss value for the segment; and wherein calculating the total energy use forthe intended route is further based on the calculated battery energy loss value for each segment of the plurality of segments. Advantageously, a second machine learning model can be trained to predict the operation of a traction battery, and in particular the temperature of the traction battery based at least in part on the prediction of energy use generated by the first machine learning model. As energy losses in the traction battery may be strongly correlated with cell temperature, predicting the temperature of the traction battery for each segment can be used to predict energy losses in the battery, allowing a whole system prediction of a reduction in the amount of energy present in the traction battery when driving the intended route. In embodiments, the second machine learning model is further configured to predict the temperature associated with the traction battery based on one or more battery parameters, wherein the one or more battery parameters comprise one or more of: a battery state of health parameter; an ambient temperature; and an initial temperature of the battery. Advantageously, the second machine learning model may be trained take as input other parameters related to the traction battery when predicting the temperature associated with the traction battery to provide a more accurate prediction. The temperature associated with the traction battery while driving the intended route may also be dependent on an initial temperature of the battery, an ambient temperature of the environment and / or a state of health parameter of the battery. Advantageously, the second machine learning model may be trained to receive one or more of these parameters as input to predict the temperature associated with the battery. In embodiments the method further comprises obtaining a third trained machine learning model, the third trained machine learning model trained based on data received from the plurality of vehicles to predict energy use by a heating, ventilation, and air conditioning, HVAC, system of the BEV based on at least one of an ambient temperature and a target cabin temperature; and for each segment of the plurality of segments, generate a prediction of energy use by the HVAC system for that segment; and wherein calculating the total energy use for the intended route is further based on the prediction of energy use by the HVAC system for each segment of the plurality of segments. Heating, ventilation and air conditioning (HVAC) systems have been found to be a significant sink of electrical energy in BEVs. Advantageously, the method may include using a third machine learning model trained to predict the energy use of a HVAC system to be included in the calculation to predict total energy use for the BEV when driving the intended route. In embodiments, predicting the temperature associated with the traction battery for each segment using the second trained machine learning model is further based on the predicted energy use of the HVAC system for the segment. Advantageously, the predicted energy use of the HVAC system can be taken into account when predicting the temperature associated with the traction battery using the second machine learning model, providing for more accurate prediction of the temperature of the battery and therefore the energy losses in the traction battery. In embodiments the method comprises calculating a predicted state of charge of the traction battery at a future time or location based on the total energy use for the intended route; and providing the predicted state of charge of the traction battery for display to a user of the BEV. Advantageously, the predicted total energy use for the intended route can be used to provide an accurate prediction of a state of charge of the vehicle at the end, or other specific point, during the route and this prediction displayed to the user of the BEV. Providing an accurate prediction of state of charge to the user increases the user’s confidence in the vehicle and reduces “range-anxiety”. According toa further aspect of the invention there is provided a method of training a machine learning model to predict energy use of a battery electric vehicle, BEV, to navigate a route, the method comprising for each of a plurality of BEVs: obtaining a plurality of vehicle configuration parameters defining a configuration of the BEV, a route driven by the BEV; and dividing the obtained route driven by the BEV into a plurality of segments; for each segment of the plurality of segments: obtaining one or more operating parameters forthat segment for the BEV; obtaining a segment energy value associated with the BEV driving the segment; generating a first training data set comprising first input / output pairs based on the plurality of segments for the plurality of BEVs, each first input / output pair associating the plurality of vehicle parameters and the operating parameters for a segment with a corresponding segment energy value; inputting, to a first machine learning model, the plurality of vehicle configuration parameters and the operating parameters from a first input / output pair of the training data set to obtain a predicted segment energy value; characterizing an error between the predicted segment energy and the segment energy value of the first input / output pair; using an optimization algorithm to update weights of the machine learning model based on the characterized error. Advantageously, the disclosed method allows for training of a machine learning model to predict energy use of a drive train based on data collected from a very large number of vehicles, for example all vehicles of a selected model may report the relevant data to a cloud service for use in training the model. This allows automatic collection of large amounts of data for use in training the model which would be expected to facilitate the training of more accurate machine learning models. In embodiments, the one or more operating parameters for a segment comprise one or more of: a vehicle speed; an indication of traffic conditions forthe segment; a road elevation orgradient; a road curvature; a road surface type; a wind direction; a wind speed; an ambient temperature and a precipitation status. Advantageously, the first machine learning model can be trained using a variety of different parameters relating to the route that may be obtained, for example, from a road database or weather service, to allow for more accurate prediction of energy use for an intended route. In embodiments, the plurality of vehicle configuration parameters comprise one or more of: a vehicle model; a vehicle weight; a tyre pressure; an indication that the vehicle is towing a trailer; and a wheel size. Advantageously, the first machine learning model can be trained using a variety of different parameters relating to the vehicle configuration forthe vehicle driving the route. This may allow the first machine learning model to accurately predict energy use for different vehicle models / trims and / or allow for changes in tyre pressure, etc. The machine learning model may be used to predict a change in energy use relative to a particular configuration parameter, for example determining an amount of energy that could be saved by ensuring correct inflation of tyres. In embodiments, dividing the obtained route into a plurality of segments comprises subdividing the route into a plurality of portions having a predefined length. In embodiments, the method comprises for each of the plurality of BEVs, obtaining an indication of a temperature associated with a traction battery of the BEV for each segment; generating a second training data set comprising second input / output pairs, each second input / output pair associating an amount of energy supplied by a traction battery of the BEV with a temperature of the traction battery for a segment; inputting, to a second machine learning model, the amount of energy supplied by a traction battery from a first input / output pair of the training data set to obtain a predicted temperature of the traction battery; characterizing an error between the predicted temperature and the temperature of the traction battery corresponding to the amount of energy supplied by a traction battery of the first input / output pair; and using an optimization algorithm to update weights of the second machine learning model based on the characterized error. Advantageously, the second machine learning model can be trained based on data collected from a large number of vehicles, providing a large training data set facilitating accurate predictions by the trained machine learning model. In embodiments, the input of each second input / output pair further comprises a battery parameter, the battery parameter comprising one or more of: a battery state of health parameter; an ambient temperature; and an initial temperature of the battery. Advantageously, the second machine learning model may be trained to take as input other parameters related to the traction battery when predicting the temperature associated with the traction battery to provide a more accurate prediction. In embodiments, the method comprises for each of the plurality of BEVs obtaining an indication of a target cabin temperature and an energy use of a heating, ventilation, and air conditioning, HVAC, system of the BEV for each segment; generating a third training data set comprising third input / output pairs, each third input / output pair associating a target cabin temperature with an energy use of the HVAC system for a segment; inputting, to a third machine learning model, a target cabin temperature the from a first input / output pair of the training data set to obtain a predicted energy use of the HVAC system; characterizing an error between the predicted energy use of the HVAC system and the target cabin temperature of the first input / output pair; and using an optimization algorithm to update weights of the machine learning algorithm based on the characterized error. In embodiments, the first machine learning model, the second machine learning model and / or the third machine learning model may each comprise any one of a decision tree regression model, a random forest regression model, a gradient boosting regression model, and a neural network regression model. According to another aspect of the invention, there is provided a computer program product comprising computer program instructions that when executed on a processor implements a method as described above. Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, in the claims and / or in the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and / or features of any embodiment can be combined in anyway and / or combination, unless such features are incompatible. The applicant reserves the right to change any originally filed claim or file any new claim accordingly, including the right to amend any originally filed claim to depend from and / or incorporate any feature of any other claim although not originally claimed in that manner. BRIEF DESCRIPTION OF THE DRAWINGS One or more embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which: Figure 1 shows a system suitable for implementing embodiments of the invention; Figure 2 illustrates a computer implemented method according to an embodiment of the present invention; Figure 3 illustrates a method of predicting an energy cost for a route segment according to embodiments of the invention; Figure 4 shows a method of predicting energy use of a battery electric vehicle, BEV, to navigate a route according to embodiments of the invention; Figure 5 illustrates a system operable to receive operating data from a plurality of vehicles according to embodiments of the invention; Figure 6 illustrates a method of training a plurality of machine learning models to predict energy use of different components or sub-systems of a BEV in accordance with embodiments of the invention; Figure 7 illustrates a method to train a first machine learning model for use in the method of Figure 2 according to an embodiment of the present invention; Figure 8 illustrates an iterative process corresponding to the method of Figure 7 for training the first machine learning model according to an embodiment of the present invention; and Figure 9 illustrates a control system suitable for performing the method of Figure 4 according to embodiments of the invention. DETAILED DESCRIPTION According to embodiments of the invention, an amount of energy to be provided by a traction battery of a battery electric vehicle (BEV) to navigate a particular route can be predicted taking into account a wide range of conditions and factors using a machine learning model. In particular, the machine learning model is trained using information, for example including driving conditions, vehicle configurations, road characteristics, etc., received from a plurality of vehicles. When planning a route to navigate in a BEV, it is known to provide an indication of range, for example whether a particular destination is within range or necessitates one or more charging stops. Drivers may rely on an onboard route navigation system to provide such range indications and to plan an appropriate route, including any necessary charging stops to arrive at the destination. However, as the attainable range of the vehicle may be due to a wide range of factors, both within the vehicle and external, such predictions may need to be updated during the journey. For example, it may be determined during the journey that an earlier, or extra, charging stop is required due to energy use exceeding a predicted amount used to generate the range prediction. Many drivers find such mid-course updates unsettling and would prefer not to re-route mid-journey. Furthermore, a driver may intend to take a further journey from the destination without first charging the vehicle and be relying on a minimum state of charge (SoC) of the traction battery being available on arrival at the destination, which may not be the case if an accurate prediction cannot be provided at the start of the journey. Thus, inaccurate predictions of range provided at the beginning of a journey may lead to poor customer experience and may increase feelings of “range anxiety” felt by the driver. An approach to improve the accuracy of range predictions for BEVs has been to implement a memory-based model, for example in the form of a look up table stored in a memory of a control system of the vehicle, which may have some limited ability to update the values of the model based on operation of that vehicle, i.e. a limited “learning” capability. However, such memory-models are limited in the number of potential factors that can be considered. As the number of dimensions, or factors, of the model increases, the storage requirements 7 increase in line with the number of possible combinations of factor values, known as the “curse of dimensionality”. Machine-learning approaches may allow for models to generate an output based on a large number of input values. However, machine-learning models capable of providing a range prediction based on a wide range of factors may require significant processing resources that would not be available on a control unit of a vehicle. However, many of the factors may not be known on the vehicle, but must be retrieved from an external information source. Embodiments of the present invention provide a range prediction based on predicted energy use of the vehicle to navigate an intended route determined external to the vehicle, for example as a cloud application. While a significant proportion of the energy used by a BEV while navigating a route is used by a drivetrain, in particular a traction motor, of the vehicle, other sub-systems of the vehicle may be significant users of energy that should be accounted for when predicting energy use. Furthermore, it is known that significant energy losses may occur in a traction battery supplying the electrical power to the drivetrain and other systems, and furthermore a heating ventilation and air conditioning (HVAC) system of the BEV may use a significant amount of electrical power, for example in particularly hot or cold ambient conditions. A system in accordance with an embodiment of the present invention is described herein with reference to the accompanying Figure 1. With reference to Figure 1, a vehicle 100 communicates with one or more software applications 120, 130 that are executed in a cloud environment 110. Alternatively, the software applications may be executed on one or more servers connected to a network. Communication between the vehicle and the software applications may be via a suitable communication link such as a 4G or 5G mobile communications network. The software applications include at least one application comprising a machine learning model trained to predict an amount of energy used by a component or sub-system of a BEV based on operating parameters of the BEV. Examples of the component orsub-system include a drivetrain of the BEV, the traction battery of the BEV and a heating, ventilation and cooling (HVAC) system of the BEV. The predicted amount of energy may be provided, for example, as an absolute energy value for the segment or as an average power value for the segment. The BEV 100 may include a satellite navigation system or similar to provide routing information to a user of the vehicle. At the start of a journey, the user may input a destination into the satellite navigation system to plan a route and to obtain route information. An important aspect of such route information will be an indication of range of the BEV and whether one or more charging stops will be necessary to complete the planned journey. According to embodiments, the BEV 100 may communicate with the software application 120 providing information on the desired route, or destination, along with one or more parameters describing the operation and configuration of the BEV. Figure 2 illustrates a computer implemented method 200 according to an embodiment of the present invention that can be implemented in the system illustrated in Figure 1. Figure 3 further illustrates operation of software application 120 to implement method 200 in accordance with an embodiment of the invention. The method begins at block 202 where configuration information for the BEV 100 is collated. The configuration information may provide parameters indicating aspects of the vehicle functionality and status. For example, certain predefined parameters relating to a model type and installed accessories for the BEV may be preprovisioned into a control system of the BEV 100. Other status information may be determined by one or more systems of the vehicle when initiating a range calculation, for example measurement of tyre pressures for the BEV; determining a towing status of the BEV; etc. In block 204 a user of the BEV 100 begins planning a route by setting a desired destination, for example on a satellite navigation system of the BEV 100. Alternatively, the user may be able to plan a route on a mobile device or computer, external to the BEV, using an app or web interface. In examples, the configuration information may include one or more of a vehicle model; a vehicle weight; a tyre pressure; an indication that the vehicle is towing a trailer; and a wheel size. One or more candidate routes may be generated based on a location of the BEV and the desired destination. For each candidate route, the route may be divided into a plurality of segments of a predefined length, for example each segment may be 1km, 2km, 5km, etc. For each segment of the candidate route one or more properties of that segment may be determined. For example, for each segment one or more of a road type, road gradient, average speed, average speed standard deviation, road surface, road curvature may be determined. Furthermore, information on environmental conditions may be obtained for each segment. For example information on precipitation status; wind speed; wind direction with respect to the vehicle direction; ambient temperature, etc may be obtained for each segment of the candidate route. According to the method of Figure 2, the collated information relating to the configuration and operation of the vehicle, along with the determined properties of each segment of the candidate route are provided to a plurality of energy estimator modules 206, 208, 210 comprising neural networks trained to predict energy use of a subsystem of the BEV 100 when navigating the candidate route. In an example method a first neural network 206 is trained to predict an amount of energy used by a drivetrain of the BEV based on the operating parameters, segment properties and / or environmental information for the segment; a second neural network 208 is trained to predict an amount of energy lost in a traction battery of the BEV 100 based on the operating parameters, segment properties and / or environmental information for the segment; and a third neural network 210 is trained to predict an amount of energy used by a HVAC system of the BEV based on the operating parameters, segment properties and / or environmental information for the segment. At block 212 the predicted energy values provided by each of the first, second and third trained neural networks are accumulated for each segment of the candidate route to generate a predicted total energy use for a candidate route. The predicted total energy use for the candidate route can then be provided to the vehicle 100 in block 214, for example via the communication network, to facilitate feedback of an accurate range indication to the user of the BEV 100. In examples, predicted total energy use values may be generated for a plurality of candidate routes and a one of the plurality of candidate routes selected based on a lowest predicted energy use value, that is a more energy efficient route of the candidate routes may be selected. Such an approach allows embodiments to take into account both traffic conditions and environmental conditions, in particular wind speed and direction, to identify a route to a desired destination having a lower expected energy expenditure. Figure 3 is a flowchart illustrating a method 300 of predicting an energy cost for a route segment according to some embodiments of the invention. For example, each of the first neural network 206, second neural network 208, and third neural network 210 may be implemented as illustrated in Figure 3. In block 310 an input sequence comprising the operating parameters of the BEV 100 is provided. In embodiments, the input sequence may also comprise segment properties and / or environmental information for the segment. In block 320, the input sequence is processed by neural network 320 trained to predict energy use of a sub-system of the BEV 100 when navigating the candidate route. In block 330, the predicted energy use value is provided forthat segment. Figure 4 is a flowchart illustrating a method 400 of predicting energy use of a battery electric vehicle, BEV, such as the BEV 100 of Figure 1, to navigate a route. In block 402, a first trained neural network is obtained, the first trained neural network trained based on data received from a plurality of vehicles to predict an amount of energy used by a drivetrain of a BEV based on one or more operating parameters of the BEV. The first trained neural network may be the first trained neural network 206 of Figure 2. In block 404, an intended route to be taken by the BEV is obtained and divided into a plurality of segments. In embodiments, the intended route may be one of a plurality of candidate routes as discussed above with reference to Figure 2. For each segment of the intended route, one or more operating parameters forthat segment is determined for the BEV, such as information relating to the configuration of the vehicle, operation of the vehicle, and / or determined properties of each segment. At block 406, for each segment of the plurality of segments, the determined operating parameters for the segment are input to the first trained neural network 206 to generate a predicted amount of energy used by the drivetrain of the BEV forthat segment, for example as illustrated in Figure 3. Based on the predicted energy use for each segment, a total energy use for the intended route can then be calculated at block 408. In embodiments, the method of Figure 4 may further include obtaining a second trained neural network trained to predict a temperature associated with a traction battery of a BEV based on an amount of energy supplied by the traction battery. For each segment of the plurality of segments the predicted amount of energy to be supplied by the traction battery forthat segment may be input to the second trained neural network to generate a prediction of the temperature associated with the traction battery forthat segment and based on the predicted battery temperature and the predicted amount of energy, a battery energy loss value for the segment may be calculated. The calculated battery energy loss value for each segment can then be combined with the predicted drivetrain energy use to determine the total energy use for the intended route. The second neural network may be trained to predict the temperature associated with the traction battery based on one or more battery parameters describing a state or operating characteristic of the traction battery. For example, the one or more battery parameters may include a battery state of health parameter; an ambient temperature; and an initial temperature of the battery. Alternatively, in embodiments the second neural network may be trained to directly predict an energy loss value for the traction battery when the BEV traverses a particular segment of the route. 10 In embodiments, the method of Figure 4 may further include obtaining a third trained neural network trained to predict energy use by a heating, ventilation, and air conditioning, HVAC, system of the BEV based on at least one of an ambient temperature and a target cabin temperature. For each segment of the plurality of segments, a prediction of energy use by the HVAC system forthat segment may be generated, for example based on an indication of a cabin temperature setting of the BEV. The predicted HVAC energy use for each segment can then be combined with the predicted drivetrain energy use to determine the total energy use for the intended route. In embodiments, the prediction of traction battery temperature may further be based on the prediction of energy to be used by the HVAC system which is supplied by the traction battery. In embodiments, the predicted total energy use for the BEV to traverse a candidate, or intended, route to arrive at the desired destination can be used to predict a state of charge value for the traction battery providing a prediction of the state of charge of the battery on arrival at the desired location, or at an intermediate location such as a charging location. The predicted state of charge may be further based on a current state of charge of the traction battery, prior to beginning the journey to the desired destination. In embodiments, operating data can be collected from a plurality of battery electric vehicles to provide training data to train the neural networks 206, 208, and / or 210. For example, operating data may be received from substantially all vehicles of a particular model type, or types, and used to improve energy cost predictions for individual vehicles. Figure 5 illustrates a system according to embodiments of the invention that is operable to receive operating data from the plurality of vehicles 100. Each BEV may collect operating data including information relating to the configuration and operation of the BEV and actual energy use by the BEV when in use. Furthermore, the operating data may include route information defining routes travelled by the vehicle alongside the actual drivetrain energy use, actual HVAC energy use and / or a measured traction battery temperature. For each BEV of the plurality of BEVs a driven route may be segmented into a plurality of segments of a predefined length, for example 1km, 2km, or 5km, etc. Each segment may be associated with an actual energy use measured by the corresponding BEV, along with the information relating to the configuration and operation of the BEV as it traversed the segment. In embodiments, for each segment of the driven route one or more properties of that segment may be determined. For example, for each segment one or more of a road type, road gradient, average speed, average speed standard deviation, road surface, or road curvature may be determined. Furthermore, information on environmental conditions may be obtained for each segment. For example, information on one or more of precipitation status; wind speed; wind direction with respect to the vehicle direction; ambient temperature, etc may be obtained for each segment of the candidate route. Figure 6 is a flowchart illustrating a method of training a plurality of neural networks to predict energy use of different components or sub-systems of a BEV in accordance with embodiments of the invention. At block 602 operating data is received from each BEV of a plurality of BEVs. The obtained operating information from the plurality of vehicles may be combined into a fleet data set including a plurality of segments with associated vehicle configuration and operating parameters and actual energy values at block 604. For use in training the neural network models, the fleet data set is divided into a training data set and a validation data set. In embodiments, a first training data set and a first validation data set is generated at block 606. The first training data set and first validation data set each comprising orthogonal subsets of the plurality of segments that associates the operating parameters for the segment, including environmental information and segment properties, with an actual energy use of the drivetrain of the corresponding vehicle to traverse the segment. Similarly, in block 614 a second training data set and a second validation set is generated from the fleet data set, each comprising orthogonal subsets of the plurality of segments that associates the operating parameters for the segment, including environmental information and segment properties, with a measured temperature of the traction battery; and at block 622 a second training data set and a second validation set is generated from the fleet data set each comprising orthogonal subsets of the plurality of segments that associates the operating parameters for the segment, including environmental information and segment properties, with an actual energy use of the HVAC system of the corresponding vehicle to traverse the segment. In this way, a large amount of data relating to energy use by BEVs may be collected and training data sets corresponding to energy use by certain components or sub-systems of the BEVs automatically generated. At block 608 of Figure 6, the first neural network can be trained based on the first training data set. Figure 7 illustrates a computer implemented method 700 according to an embodiment of the present invention to train first neural network for use in the method of Figure 2. Figure 8 illustrates an iterative process 800 corresponding to the method 700 of training the first neural network engine 206 according to an embodiment of the invention. As illustrated in Figure 7, the first training data set 710 is provided comprising training example input / output pairs based on the plurality of segments forthe plurality of BEVs, each input / output pair associating the plurality of vehicle parameters and the operating parameters for a segment with a corresponding segment energy value. In block 802 of the method 800, for each of the plurality of BEVs, a plurality of vehicle configuration parameters defining a configuration of the BEV and a route driven by the BEV are obtained. In block 804, the route driven the BEV is divided into a plurality of segments, for example each segment having a same predefined length. For each segment, one or more operating parameters relating to that segment forthe BEV are obtained at block 806. Further, a segment energy value indicating an actual amount of energy used by the BEV to traverse the segment is obtained at block 808. At block 810 of Figure 8, the obtained information is used to generate the first training data set 710 comprising first input / output pairs based on the plurality of segments forthe plurality of BEVs, each first input / output pair associating the plurality of vehicle parameters and the operating parameters for a segment with a corresponding segment energy value. In block 812, for one or more of the input / output pairs, the plurality of vehicle parameters and the operating parameters for a segment are input to the first neural network 206. For each of the input values used to train the first neural network 206, an output is generated by the first neural network 206 in response, the output comprising predicted energy use by the drivetrain forthe segment. Initially during training of the neural network, the predicted value may be significantly different from the actual energy use value of the corresponding input / output pair. In block 814, the predicted energy use is compared in comparator 730 with the actual energy use value 740 of the input / output pair and an error between the predicted energy use and the expected actual energy use of the training data is characterized. In block 816, an optimization algorithm, for example based on a gradient decent technique, is used to update weights of the first neural network 206 based on the characterized error. Updated weights may be calculated by a tuning algorithm 750 that receives the output of comparator 730 and in response updates weights of the neural network of the first neural network 206 using an appropriate optimization algorithm. Returning to Figure 6, the trained first neural network can be validated against the validation data set 610 at block 612 to determine if the first neural network has been successfully trained to predict the drivetrain energy use. Similarly, second training and validation sets 616 and 618 may be generated at block 614, associating segment and BEV parameters with measured traction battery temperatures for each segment; and third training and validation sets 624 and 626 may be generated at block 622, associating segment and BEV parameters with HVAC energy use for each segment. Training of the second neural network 208 and third neural network 210 can then be performed as described above forthe first neural network 206, based on the corresponding training and validation data sets. Thus, a large volume of training data can be automatically generated and used to train neural network models to more accurately predict energy use by components or sub-systems of BEVs based on a wide range of potential factors including one or more of: a vehicle speed; an indication of traffic conditions forthe segment; a road elevation or gradient; a road curvature; a road surface type; a wind direction; a wind speed; an ambient temperature; a precipitation status; a vehicle model; a vehicle weight; a tyre pressure; an indication that the vehicle is towing a trailer; and a wheel size; a battery state of health parameter; an ambient temperature; and an initial temperature of the battery. While embodiments have been described as using one or more trained neural networks to predict energy use in a battery electric vehicle, it will be recognized that neural networks are just one example of a machine learning model that can be used to generate predictions. In other embodiments, neural networks may be replaced by alternative machine learning models, including a decision tree regression model, a random forest regression model, a gradient boosting regression model, a neural network regression model, etc. Certain methods and systems as described herein may be implemented by one or more processors that processes program code that is retrieved from a non-transitory storage medium. Figure 9 shows an example 900 of a device comprising a computer-readable storage medium 920 coupled to at least one processor 910. The computer-readable media 920 can be any media that can contain, store, or maintain programs and data for use by or in connection with an instruction execution system. Computer-readable media can comprise any one of many physical media such as, for example, electronic, magnetic, optical, electromagnetic, or semiconductor media. More specific examples of suitable machine-readable media include, but are not limited to, a hard drive, a random-access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory, or a portable disc. In Figure 9, the computer-readable storage medium comprises program code to perform a method corresponding to the embodiment shown in Figure 4, that is: obtaining 402 a first trained neural network, the first trained neural network trained based on data received from a plurality of vehicles to predict an amount of energy used by a drivetrain of a BEV based on one or more operating parameters of the BEV; obtaining 404 5 an intended route to be taken by the BEV and determining, for each of a plurality of segments of the intended route, one or more operating parameters for that segment for the BEV; for each segment of the plurality of segments, inputting 406 the one or more determined operating parameters for that segment to the first trained neural network to generate a predicted amount of energy used by the drivetrain of the BEV for that segment; and calculating 408 a total energy use for the intended route based on the predicted amount of energy used 10 for each segment of the plurality of segments. In embodiments, the computer-readable storage medium may comprise program code to perform a method corresponding to the embodiment of any of Figures 3, 6 or 8. It will be appreciated that various changes and modifications can be made to the present invention without departing from the scope of the present application. 15

Claims

1. A method of predicting energy use of a battery electric vehicle, BEV, to navigate a route, the method comprising:obtaining a first trained machine learning model, the first trained machine learning model trained based on data received from a plurality of vehicles to predict an amount of energy used by a drivetrain of a BEV based on one or more operating parameters of the battery electric vehicle ;obtaining an intended route to be taken by the battery electric vehicle and determining, for each of a plurality of segments of the intended route, one or more operating parameters forthat segment for the battery electric vehicle;for each segment of the plurality of segments, inputting the one or more predicted operating parameters for that segment to the first trained machine learning model to generate a predicted amount of energy used by the drivetrain of the battery electric vehicle forthat segment; andcalculating a total energy use for the intended route based on the predicted amount of energy used for each segment of the plurality of segments.

2. The method of claim 1, wherein the one or more operating parameters for a segment comprise one or more of: a vehicle speed; an indication of traffic conditions for the segment; a road elevation or gradient; a road curvature; a road surface type; a wind direction; a wind speed; an ambient temperature and a precipitation status.

3. The method of claim 1 or claim 2, wherein the operating parameters further comprise one or more parameters relating to a vehicle configuration.

4. The method of claim 3, wherein the one or more parameters relating to the vehicle configuration comprise one or more of: a vehicle model; a vehicle weight; a tyre pressure; an indication that the vehicle is towing a trailer; and a wheel size.

5. The method of any preceding claim, further comprising generating the plurality of segments of the intended route by subdividing the intended route into a plurality of portions having a predefined length.

6. The method of any preceding claim, the method further comprising:obtaining a second trained machine learning model, the second trained machine learning model trained based on data received from the plurality of vehicles to predict a temperature associated with a traction battery of a BEV based on an amount of energy supplied by the traction battery;for each segment of the plurality of segments:inputting the predicted amount of energy for that segment to the second trained machine learning model to generate a prediction of the temperature associated with the traction battery for that segment; andbased on the predicted battery temperature and the predicted amount of energy, calculating a battery energy loss value for the segment; andwherein calculating the total energy use for the intended route is further based on the calculated battery energy loss value for each segment of the plurality of segments.

7. The method of claim 6, wherein the second machine learning model is further configured to predict the temperature associated with the traction battery based on one or more battery parameters, wherein the one or more battery parameters comprise one or more of: a battery state of health parameter; an ambient temperature; and an initial temperature of the battery.

8. The method of any of claims 6 or 7, the method further comprising:obtaining a third trained machine learning model, the third trained machine learning model trained based on data received from the plurality of vehicles to predict energy use by a heating, ventilation, and air conditioning, HVAC, system of the BEV based on at least one of an ambient temperature and a target cabin temperature; andfor each segment of the plurality of segments, generate a prediction of energy use by the HVAC system forthat segment; andwherein calculating the total energy use for the intended route is further based on the prediction of energy use by the HVAC system for each segment of the plurality of segments.

9. The method of claim 8, wherein predicting the temperature associated with the traction battery for each segment using the second trained neural network is further based on the predicted energy use of the HVAC system for the segment.

10. The method of any preceding claim further comprising:calculating a predicted state of charge of the traction battery at a future time or location based on the total energy use for the intended route; andproviding the predicted state of charge of the traction battery for display to a user of the vehicle.

11. A method of training a machine learning model to predict energy use of a battery electric vehicle, BEV, to navigate a route, the method comprising:for each of a plurality of BEVs:obtaining a plurality of vehicle configuration parameters defining a configuration of the BEV, a route driven by the BEV; anddividing the obtained route driven by the BEV into a plurality of segments;for each segment of the plurality of segments:obtaining one or more operating parameters forthat segment for the BEV;16obtaining a segment energy value associated with the BEV driving the segment;generating a first training data set comprising first input / output pairs based on the plurality of segments for the plurality of BEVs, each first input / output pair associating the plurality of vehicle parameters and the operating parameters for a segment with a corresponding segment energy value;inputting, to a first machine learning model, the plurality of vehicle configuration parameters and the operating parameters from a first input / output pair of the first training data set to obtain a predicted segment energy value;characterizing an error between the predicted segment energy and the segment energy value of the first input / output pair;using an optimization algorithm to update the first machine learning model based on the characterized error.

12. The method of claim 11, wherein the one or more operating parameters for a segment comprise one or more of: a vehicle speed; an indication of traffic conditions for the segment; a road elevation or gradient; a road curvature; a road surface type; a wind direction; a wind speed; an ambient temperature and a precipitation status.

13. The method of claim 11 or claim 12, wherein the plurality of vehicle configuration parameters comprise one or more of: a vehicle model; a vehicle weight; a tyre pressure; an indication that the vehicle is towing a trailer; and a wheel size.

14. The method of any of claims 11 to 13, further comprising:for each of the plurality of BEVs, obtaining an indication of a temperature associated with a traction battery of the BEV for each segment;generating a second training data set comprising second input / output pairs, each second input / output pair associating an amount of energy supplied by a traction battery of the BEV with a temperature of the traction battery for a segment;inputting, to a second machine learning model, the amount of energy supplied by a traction battery from a first input / output pair of the training data set to obtain a predicted temperature of the traction battery;characterizing an error between the predicted temperature and the temperature of the traction battery corresponding to the amount of energy supplied by a traction battery of the first input / output pair; andusing an optimization algorithm to update the second machine learning model based on the characterized error.

15. The method of claim 14, wherein the input of each second input / output pair further comprises a battery parameter, the battery parameter comprising one or more of: a battery state of health parameter; an ambient temperature; and an initial temperature of the battery.

Images