A method and control system for predicting temperatures in an electronic drive system of a vehicle

Abstract

A method / system for predicting temperatures over time 200 in an electric drive system of a vehicle, the electric drive system comprising a rotor, a stator and a cooling system. The computer-implemente

Description

The present disclosure relates to a method and control system for predicting temperatures in an electric drive system of a vehicle. Aspects of the invention relate to a method, a control system, a system, computer readable instructions and a vehicle. BACKGROUND It is known to provide an electric drive system for a vehicle, which has an electric motor along with a cooling system. One type of known electric motor comprises a stator and a rotor as part of a permanent magnet synchronous motor. An air gap is maintained between the rotor and the stator with the rotor being located within the confines of the stator. The stator is a stationary element of the electric motor which may comprise a plurality of slots within which electrical windings are located. The rotor is a rotating element of the electric motor allowing a transfer of electrical energy input into the motor to a mechanical output, such as the rotation of a driveshaft of a vehicle. During its operation, parts of the electric drive system generate heat, due to various losses such as hysteresis and eddy current losses. This causes a gradual increase in the temperature. The cooling system provides a path for coolant to extract heat from the parts of the electric motor, to maintain the electric drive system within a working range. Monitoring temperature of components in an electric drive system is a crucial monitor of the performance and reliability of the electric drive system. 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 a computer-implemented method of predicting temperatures overtime in an electric drive system of a vehicle, a control system for controlling an electric drive system of a vehicle, a system comprising the control system, computer readable instructions, and a vehicle as claimed in the appended claims. According to an aspect of the present invention there is provided a computer-implemented method of predicting temperatures over time in an electric drive system of a vehicle, the electric drive system comprising a rotor, a stator and a cooling system, the method comprising: receiving a plurality of electric drive system data inputs; inputting the plurality of electric drive system data inputs into a trained multi-horizon time forecasting model comprising a deep neural network; obtaining, from the multi-horizon time forecasting model, at least one of a first predicted time series predicting a temperature of a rotor component, and a second predicted time series predicting a temperature of a winding hotspot of the stator; determining, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor component, and if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit; outputting a signal to one or more controller of the vehicle if the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and / or if the winding hotspot limit is predicted to be exceeded according to the second predicted time series. Advantageously, the computer-implemented method provides accurate prediction of time series data relating to the temperature of the electric motor, particularly of the rotor, stator and / or other components thereof. This enables predictive maintenance or early fault detection to thereby increase reliability. Optionally, the plurality of electric drive system data inputs comprise any of: a rotation speed of the rotor as a function of time; a torque output of the electric drive system as a function of time; a rotor coolant flow rate of the cooling system as a function of time; a stator coolant flow rate of the cooling system as a function of time; a rotor coolant inlet temperature of the cooling system as a function of time; a stator coolant inlet temperature of the cooling system as a function of time. Advantageously, important temperatures relating to the performance of the electric drive system, i.e. the temperature of the rotor component and / or the winding hotspot temperature, can be predicted without requiring sensor data of the temperature of the rotor component and the winding hotspot temperature. The temperature predictions can be made with minimal information about the electric drive system environment. Optionally, the method additionally comprises obtaining both of the first predicted time series and the second predicted time series from the trained multi-horizon time forecasting model, wherein the method comprises outputting the signal to the one or more controller if one or both of the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and the winding hotspot limit is predicted to be exceeded according to the second predicted time series. Advantageously, by obtaining predicted time series for the temperature of both the rotor component and the winding hotspot, this incorporates redundancy into the monitoring of the electric drive system, as only one of the rotor component temperature limit and winding hotspot limit needs to be exceeded for the signal to be sent to the one or more controller, thereby increasing reliability. Optionally, the method additionally comprises: increasing the rotor coolant flow rate and / or the stator coolant flow rate in response to receiving, at the one or more controller of the vehicle, the signal that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. In some examples, the rotor coolant flow rate is increased specifically for when the rotor component temperature limit is predicted to be exceeded, and the stator coolant flow rate is increased specifically for when the winding hotspot limit is predicted to be exceeded. In other examples, both of the rotor coolant flow rate and the stator coolant flow rate are increased when either of the rotor component temperature limit is predicted to be exceeded and the winding hotspot limit is predicted to be exceeded. Advantageously therefore the operation of the electric drive system can be altered in response to receiving the signal to attempt to prevent the temperature limit(s) from being exceeded. Optionally, the method additionally comprises: limiting the maximum current provided to the coils of the stator and / or limiting the maximum torque output of the electric drive system in response to receiving, at the one or more controller of the vehicle, the signal that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. Advantageously the operation of the electric drive system can therefore be limited in response to receiving the signal. This may therefore prevent the temperature limit(s) from being exceeded, which may have led to increased wear and / or failure of components of the electric drive system. The limiting of the maximum current provided to the coils of the stator and / or limiting the maximum torque output can be done after the rotor coolant flow rate and / or the stator coolant flow rate have been increased. For example the electric drive system can first attempt to prevent the temperature limit(s) from being exceeded by increasing the coolant flow rates, and if it is still predicted that one or both of the temperature limit(s) will be exceeded, the maximum current provided to the coils of the stator and / or the maximum torque output of the electric drive system can be limited. In other examples, increasing the rotor coolant flow rate and / or the stator coolant flow rate in response to receiving the signal is done in conjunction with limiting the maximum current provided to the coils of the stator and / or limiting the maximum torque output of the electric drive system in response to receiving the signal. Optionally, the method additionally comprises: obtaining sensor measurements of the temperature of the rotor component; determining, when the first predicted time series has been obtained, errors between the measurements of the rotor component to corresponding data points of the first predicted time series to obtain a first error time series. Optionally, the method additionally comprises: obtaining sensor measurements of the temperature of the winding hotspot of the stator; determining, when the second predicted time series has been obtained, errors between the measurements of the winding hotspot of the stator to corresponding data points of the second predicted time series to obtain a second error time series. Advantageously, obtaining either or both of the first or second error time series by the sensor measurements of the temperature of the rotor component and / or winding hotspot of the stator enables both predictive and real-time or “live” temperatures of the rotor component and / or winding hotspot to be used to monitor operation of the electric drive system. This can enable both predictive maintenance and / or early fault identification as well as live fault detection, further increasing redundancy in the electric drive system and improving reliability. Optionally, the method additionally comprises: determining whether one or more data points of the first error time series exceed a first error limit; determining whether one or more data points of the second error time series exceed a second error limit; determining, if either of the first error limit and the second error limit have been exceeded, that a fault in the electric drive system has occurred; outputting an indication to the one or more controller of the vehicle that the fault has occurred. Advantageously, diagnosis of faults can therefore be determined by comparing the live sensor measurements of the temperature(s) to the predicted temperature(s), which can lead to necessary action being taken sooner, and avoid the requirement of a person servicing the vehicle detecting the fault at a later point in time. Optionally, the method additionally comprises: outputting, if it is determined that the first error limit has been exceeded and the second error limit has not been exceeded, an indication that the fault is due to either a low flow of oil to a shaft of the rotor or that there is a high inlet oil temperature to the shaft of the rotor; outputting, if it is determined that the first error limit has not been exceeded and the second error limit has been exceeded, an indication that the fault is due either to a low flow of oil to the stator or there is a high inlet oil temperature to the stator; outputting, if it is determined that the first error limit has been exceeded and the second error limit has been exceeded, an indication that the fault is due to a low oil level in a sump of the electric drive system or that a heat exchanger of the vehicle has low efficiency. Advantageously several different types of faults can therefore be determined, which can increase efficiency in time for maintenance of the vehicle. Optionally the method additionally comprises outputting a notification to an infotainment system of the vehicle, the notification indicating that the fault has occurred and / or indicating the signal that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. Advantageously the user of the vehicle can therefore be informed about the temperature limits and / or faults that have occurred. This may be done in conjunction with a change in operation of the electric drive system as referred to below. Optionally the method additionally comprises: limiting the maximum current provided to the coils of the stator and / or limiting the maximum torque output of the electric drive system in response to receiving, at the one or more controller of the vehicle, the indication that the fault has occurred. Advantageously the vehicle therefore may change the operation of the electric drive system in order to prevent further damage to the electric drive system, thereby reducing maintenance requirements. Optionally, the method further comprises transmitting, to an electronic device via wireless communication: the indication that the fault has occurred and / or the indication that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded; the first predicted time series; and the second predicted time series; wherein the electronic device is arranged to provide a Guided User Interface for displaying on a display of the electronic device: the indication that the fault has occurred and / or the warning; the first predicted time series; and the second predicted time series. Advantageously a user of the electronic device can be informed of the indication(s) and / or fault(s). the user of the electronic device does therefore not have to be within and / or operating the vehicle to learn of the indication(s) and / or fault(s). As well as being beneficial for the normal operator (i.e. driver and / or owner) of the vehicle, it also benefits servicing people to determine what type of maintenance / repair could need performing on the vehicle without needing to enter the vehicle, and may even be provided before a vehicle arrives at the servicing location, which can increase efficiency of sourcing parts, booking appointment slots, etc.. Optionally, the trained multi-horizon time forecasting model is a trained Temporal Fusion Transformer model. Advantageously using a Temporal Fusion Transformer model provides dynamic temperature prediction with greater accuracy and by the use of readily accessible data from the electric drive system. Optionally, the rotor component is one of: a rotor magnet; a rotor bar. Advantageously the method can be used for different types of electric drive motors. According to another aspect of the invention, there is provided Computer readable instructions which, when executed by a computer, are arranged to perform a method according to any of the preceding paragraphs. According to yet another aspect of the invention, there is provided a control system for controlling an electric drive system of a vehicle, the electric drive system comprising a rotor, a stator and a cooling system the control system comprising one or more controller, the control system comprising one or more processors collectively configured to: receive a plurality of electric drive system data inputs; input the plurality of electric drive system data inputs into a trained multi-horizon time forecasting model comprising a deep neural network; obtain, from the multi-horizon time forecasting model, at least one of a first predicted time series predicting a temperature of a rotor component and a second predicted time series predicting a temperature of a winding hotspot of the stator; determine, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor component; determine, if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit; output a signal to one or more controller of the vehicle if the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and / or the winding hotspot limit is predicted to be exceeded according to the second predicted time series. The control system can comprise one or more controllers collectively comprising at least one electronic processor having an electrical input for receiving an input signal; and at least one memory device electrically coupled to the at least one electronic processor and having instructions stored therein; and wherein the at least one electronic processor is configured to access the at least one memory device and execute the instructions thereon so as to: receive a plurality of electric drive system data inputs; input the plurality of electric drive system data inputs into a trained multi-horizon time forecasting model comprising a deep neural network; obtain, from the multihorizon time forecasting model, at least one of a first predicted time series predicting a temperature of a rotor component and a second predicted time series predicting a temperature of a winding hotspot of the stator; determine, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor component; determine, if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit; output a signal to one or more controller of the vehicle if the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and / or the winding hotspot limit is predicted to be exceeded according to the second predicted time series. Advantageously, the control system provides accurate prediction of time series data relating to the temperature of the electric motor, particularly of the rotor, stator and / or other components thereof. This enables predictive maintenance or early fault detection to thereby increase reliability. Optionally, the control system can implement the computer-implemented method of any preceding paragraph. According to yet another aspect of the invention, there is provided a system comprising the control system of any preceding paragraph and an electric drive system. According to yet another aspect of the invention, there is provided a vehicle comprising the system of any preceding paragraph or the control system of any preceding paragraph. 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 any way 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 schematic representation of a system comprising a control system, an electric drive system and a controller of a vehicle suitable for implementing embodiments of the invention; Figure 2 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 3 shows operation of a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 4 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 5 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 6A shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 6B shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 7 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 8 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 9 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 10 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 11 shows operation of a method of predicting temperatures over time in an electric drive system according to embodiments of the invention; Figure 12 shows a method of training a multi-horizon forecasting model according to embodiments of the invention. Figures 13A, 13B, 13C show graphs illustrating predicted temperature values of a rotor component compared to measured values of the rotor component; Figure 14 shows a computer system suitable for implementing a method of predicting temperatures over time in an electric drive system in accordance with embodiments of the invention; and Figure 15 shows a vehicle in accordance with embodiments of the invention. DETAILED DESCRIPTION According to embodiments of the invention, temperature(s) of a rotor component and / or a winding hotspot of a stator of an electric drive system of a vehicle can be predicted using a control system that controls the electric drive system. Monitoring these temperatures of an electric drive system is beneficial for predictive maintenance and early fault detection, and therefore increases reliability of the electric drive system. Temperature variation of various parts of the electric drive system depends on the duty cycle of the electric motor and the engineering design of the motor. Therefore, temperature variation can be used to predict malfunctioning of the motor. Figure 1 shows a schematic representation of a system 130 comprising a control system 100, an electric drive system 110 and a controller 120 of a vehicle suitable for implementing embodiments of the invention. A control system 100 for controlling an electric drive system 110 of a vehicle in accordance with an embodiment of the present invention is described herein with reference to the accompanying Figure 1. With reference to Figure 1, the electric drive system 110 comprises a rotor 112, a stator 114 and a cooling system 116. The control system 100 comprises one or more controller 102, the control system 100 comprising one or more processors 104. The one or more processors 104 are collectively configured to: receive a plurality of electric drive system data inputs 118; input the plurality of electric drive system data inputs 118 into a trained multi-horizon time forecasting model comprising a deep neural network; obtain, from the multi-horizon time forecasting model, at least one of a first predicted time series predicting a temperature of a rotor 112 component and a second predicted time series predicting a temperature of a winding hotspot of the stator 114; determine, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor 112 component; determine, if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit; output a signal 106 to one or more controller 120 of the vehicle if the rotor 112 component temperature limit is predicted to be exceeded according to the first predicted time series and / or the winding hotspot limit is predicted to be exceeded according to the second predicted time series. Winding hotspots are point(s) in the electric drive system 110 which are at a high or the highest temperature due to flux leakage from the windings. It is widely known how to find and measure winding hotspots on electric drive systems, as well as transformers, etc.. The control system 100 enables monitoring of important temperatures within an electric drive system 110 during its operation and can enable early action on maintenance and / or changing the operation of the electric drive system 110 to prevent faults, reduce wear by predicting the important temperatures, the important temperatures being the winding hotspot of the stator 114 and / or temperature of the rotor 112 component. By using a trained multi-horizon time forecasting model, more accurate prediction can be carried out, and it provides predictions for multiple time points in the future. The trained multi-horizon time forecasting model can handle several variables, with or without knowledge on how they interact. In some embodiments, the one or more controller 102 of the control system 100 comprises memory storing the multi-horizon time forecasting model, and in other embodiments the one or more controller 102 is configured to access memory storing the multi-horizon time forecasting model, via an input / output interface, for example. In some embodiments, the electric drive system 110 is configured to receive a control signal 122 from the one or more controller 120 of the vehicle. The control signal 122 may cause the electric drive system 110 to increase a rotor coolant flow rate and / or a stator coolant flow rate associated with the cooling system 116 in response to receiving, at the one or more controller 120 of the vehicle, the signal 106 that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. In some examples, the rotor coolant flow rate is increased specifically for when the rotor 112 component temperature limit is predicted to be exceeded, and the stator coolant flow rate is increased specifically for when the winding hotspot limit is predicted to be exceeded. In other examples, both of the rotor coolant flow rate and the stator coolant flow rate are increased when either of the rotor 112 component temperature limit is predicted to be exceeded and the winding hotspot limit is predicted to be exceeded. In some embodiments, the control signal 122 may cause the electric drive system 110 to limit the maximum current provided to the coils of the stator 114 and / or limit the maximum torque output of the electric drive system 110 in response to receiving, at the one or more controller 120 of the vehicle, the signal 106 that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. In other embodiments, the electric drive system 110 is configured to receive a control signal from the control system 100. In some embodiments, the control system 100 does not output signal 106 to one or more controller 120 of the vehicle. The control system 100 may provide some or all the functionality of the one or more controller 120 of the vehicle. The one or more controller 120 can be referred to as an electronic control unit or ECU. The control system 100 can also be referred to as an electronic control unitor ECU. The electric drive system 110, control system 100 and one or more controller 120 are therefore arranged to protect components of the electric drive system 110 by limiting the operation of the electric drive system 110 in response to receiving the signal 106 at the one or more controller 120. This may therefore prevent the temperature limit(s) from being exceeded, which may have led to increased wear and / or failure of components of the electric drive system 110. As shown in Figure 1, in one or more embodiments according to the present invention, the control system 100 and the electric drive system 100 can be part of a system 130. In other embodiments the one or more controller 120 can also form part of the system 130. Figure 2 shows a flow chart illustrating a method 200 of predicting temperatures over time in an electric drive system 110 according to embodiments of the invention, that can be implemented in the control system 100 illustrated in Figure 1. Figure 3 further illustration operation of method 200 in accordance with an embodiment of the invention. The computer-implemented method 200 begins in block 210. In block 210, the method 200 comprises receiving a plurality of electric drive system data inputs 118. The method proceeds to block 220, which comprises inputting the plurality of electric drive system data inputs 118 into a trained multi-horizon time forecasting model 300 comprising a deep neural network. The trained multi-horizon time forecasting model 300 is arranged to provide at least one predicted time series 310 as an output. The method 200 continues in block 220 with obtaining, from the multi-horizon time forecasting model 300, the least one predicted time series 310, which can be a first predicted time series predicting a temperature of a rotor 112 component, and / or a second predicted time series predicting a temperature of a winding hotspot of the stator 114. The method 200 further continues at block 230 with determining, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor 112 component, and if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit of the stator 114. The one or more temperature values 320 which exceed a temperature limit, whether it is one or more temperature values of the rotor 112 component that exceed a temperature limit of the rotor component and / or it is one or more temperature values of the winding hotspot that exceed a winding hotspot limit are provided as a signal 106 for output. The method 200 continues in block 240 with outputting the signal 106 to one or more controller 120 of the vehicle if the rotor 112 component temperature limit is predicted to be exceeded according to the first predicted time series and / or if the winding hotspot limit is predicted to be exceeded according to the second predicted time series. As described herein, in some embodiments the signal 106 is not output to one or more controller 120, and the control system 100 functions as the one or more controller 120 of the vehicle. The plurality of electric drive system data inputs 118 can comprise any of: a rotation speed of the rotor 112 as a function of time; a torque output of the electric drive system 110 as a function of time; a rotor coolant flow rate of the cooling system 116 as a function of time; a stator coolant flow rate of the cooling system 116 as a function of time; a rotor coolant inlet temperature of the cooling system 116 as a function of time; a stator coolant inlet temperature of the cooling system 116 as a function of time. These data inputs 118 can be obtained from sensor data. The sensors used to provide the sensor data can be sensors 119 of the electric drive system 110 as shown in Figure 1, or can be other sensors within the vehicle. The rotation speed of the rotor 112 may be in revolutions per minute (RPM) or metres per second (m / s) as a function of time. The torque output of the electric drive system 110 can be in Newton metres (Nm) as a function of time. The rotor coolant flow rate and stator coolant flow rate can be volume flow rates in litres per minute as a function of time. The temperatures of the rotor coolant inlet and the stator coolant inlet can be provided in °C as a function of time. In other embodiments, the plurality of electric drive system 110 data inputs 118 can also or alternatively include: value of the current (RMS) used to excite the stator coils, assuming a 3-phase balanced excitation, in amperes as a function of time; temperature measurement from multiple sensors located on the electric drive, in °C as a function of time. The temperature of the rotor 112 component and / or the winding hotspot temperature can therefore be predicted without requiring sensor data of the temperature of the rotor 112 component and the winding hotspot temperature. The temperature predictions can be made with minimal information about the electric drive system 110 environment. The method 200 may comprise obtaining both of the first predicted time series and the second predicted time series from the trained multi-horizon time forecasting model, and may comprise outputting the signal to the one or more controller if one or both of the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and the winding hotspot limit is predicted to be exceeded according to the second predicted time series. For the rotor 112 component, athermal hotspot Ison the magnets of the rotor component and their maximum permissible temperature depends on the demagnetisation temperature at which the permanent magnets significantly lose their magnetic performance. Demagnetisation temperature for NdFeB magnets, which can be used in rotor 112 components disclosed herein, is 155 °C. This can be defined as the rotor component temperature limit. Alternatively a safety factor may be applied to the demagnetisation temperature. For example, using a safety factor of 5 °C, the temperature limit for the rotor component can be 150 °C during operation. For other types of magnet, the temperature limit can be set as : the demagnetisation temperature minus 5 °C. Different safety factor amounts are envisaged. For the winding hotspot limit on the stator, the stator windings have a maximum permissible temperature which depends on the winding insulation technique used and the stator core's saturation limits. An example temperature limit used for the stator windings during the devising of this invention was 170 °C. In some examples, the temperature limit of the winding hotspot can be set as: the largest of the stator core insulation's maximum permissible temperature and the stator core saturation temperature minus 5 °C as a safety factor. Different safety factor amounts are envisaged. By obtaining predicted time series for the temperature of both the rotor 112 component and the winding hotspot, this incorporates redundancy into the monitoring of the electric drive system, as only one of the rotor component temperature limit and winding hotspot limit needs to be exceeded for the signal to be sent to the one or more controller, thereby increasing reliability. Providing fail-safe or backup procedures for determining and predicting performance of the electric drive system 110, thereby providing redundancy, is beneficial for increasing reliability and preventing faults and damage to components of the electric drive system 110. Figure 4 shows a flow chart illustrating part of a method 200 of predicting temperatures over time in an electric drive system 11-according to embodiments of the invention. The dotted line represents that the method illustrated in Figure 4 as incorporating the steps of method 200 illustrated in Figure 2. The method 200 in the embodiment illustrated in Figure 4 proceeds to block 400 from block 250, which comprises increasing the rotor coolant flow rate and / or the stator coolant flow rate in response to receiving, at the one or more controller 120 of the vehicle, the signal 106 that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. In some examples, the rotor coolant flow rate is increased specifically for when the rotor component temperature limit is predicted to be exceeded, and the stator coolant flow rate is increased specifically for when the winding hotspot limit is predicted to be exceeded. In other examples, both of the rotor coolant flow rate and the stator coolant flow rate are increased when either of the rotor component temperature limit is predicted to be exceeded and the winding hotspot limit is predicted to be exceeded. The method 200 proceeds to block 410 from block 400, which comprises limiting the maximum current provided to the coils of the stator 114 and / or limiting the maximum torque output of the electric drive system 110 in response to receiving, at the one or more controller 120 of the vehicle, the signal 106 that the rotor 112 component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. In certain embodiments, block 410 and block 400 are both initiated in response to receiving, at the one or more controller 120 of the vehicle, the signal 106 that the rotor 112 component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. In other words, block 410 is not dependent upon block 400 being performed first before being enacted. In some embodiments block 400 and 410 can be performed in conjunction. In other embodiments, block 410 is enacted after block 400 has been performed. For example block 410 can be performed in dependence upon block 400 being performed. For example, The limiting of the maximum current provided to the coils of the stator 114 and / or limiting the maximum torque output can be done after the rotor coolant flow rate and / or the stator coolant flow rate have been increased. For example the electric drive system 110 can first attempt to prevent the temperature limit(s) from being exceeded by increasing the coolant flow rates, and if it is still predicted that one or both of the temperature limit(s) will be exceeded, the maximum current provided to the coils of the stator and / or the maximum torque output of the electric drive system can be limited. After block 400 is performed, the blocks 210-250 described herein may be repeated to determine if the rotor component temperature limit and / or the winding hotspot limit is predicted to be exceeded after the coolantflow rate(s) has been increased. The computer-implemented method 200 may pause for a predetermined time after block 400 is performed before blocks 210-250 are repeated, to allow any impact caused by the increase in the coolant flow rate(s) to take effect. The method 200 illustrated in Figure 4 therefore can protect components of the electric drive system 110 by altering the coolant flow rate(s) and / or limiting the operation of the electric drive system 110 in response to receiving the signal 106 at the one or more controller 120. This may therefore prevent the temperature limit(s) from being exceeded, which may have led to increased wear and / or failure of components of the electric drive system 110. This increases reliability of the electric drive system 110 and reduces the frequency of maintenance required. Figure 5 shows a flow chart illustrating part of a method 200 of predicting temperatures over time in an electric drive system 110 according to embodiments of the invention. The dotted line represents that the method 200 illustrated in Figure 5 as incorporating the steps of method 200 illustrated in Figure 2. The method 200 in the embodiment illustrated in Figure 5 proceeds to block 500 from block 250, which comprises: obtaining sensor measurements. The sensor measurements can be sensor measurements of the temperature of the rotor 112 component. Alternatively, or in addition, the sensor measurements can be of the temperature of the winding hotspot of the stator 114. The method 200 illustrated in Figure 5 further comprises block 510, which comprises determining, when the first predicted time series has been obtained, errors between the measurements of the rotor 112 component to corresponding data points of the first predicted time series to obtain a first error time series. Alternatively, or in addition to this, block 510 comprises determining, when the second predicted time series has been obtained, errors between the measurements of the winding hotspot of the stator 114 to corresponding data points of the second predicted time series to obtain a second error time series. The method illustrated in Figure 5 can also incorporate blocks 400, 410 in Figure 4, and the blocks 500, 510 can be performed either before, after or simultaneously as each of blocks 400, 410. Each of the first predicted time series and the second predicted time series have data points at fixed time intervals proceeding from an origin which represents the current time. As time progresses the first predicted time series and the second predicted time series updates to predict new data points for the time series. The measurements of the rotor 112 component and / or the winding hotspot may be collected as real-time or “live” data, although there is likely to be some inherent delay in the data collection. The real-time data points can then be compared to corresponding data points in the respective predicted time series to determine the difference between corresponding data points, which are referred to as errors between the measurements of the rotor 112 component or winding hotspot and the corresponding first predicted or second predicted time series data points. Over time as more and more errors are determined this produces the respective first error time series and the second error time series. Figure 6A shows a flow chart illustrating a method 200 of predicting temperatures over time in an electric drive system according to embodiments of the invention. Figure 6A shows an example of how the first error time series and the second error time series may be used to detect whether a fault has occurred in the electric drive system 110. The dotted line illustrates that this embodiment is a continuation of method 200 and incorporates the blocks shown in Figure 2. The blocks of Figure 5 are also incorporated, and optionally the blocks of Figure 4. The method 200 proceeds with block 600, determining whether one or more data points of the first error time series exceed a first error limit. The method then proceeds with block 610, determining whether one or more data points of the second error time series exceed a second error limit. The method further proceeds in block 620 with determining, if either of the first error limit and the second error limit have been exceeded, that a fault in the electric drive system 110 has occurred. The method proceeds in block 630 with outputting an indication to the one or more controller 120 of the vehicle that the fault has occurred. In some examples, the indication that the fault has occurred comprises one or more Diagnostic Trouble Codes (DTCs) and / or Data Identifiers (DIDs). The first error limit and the second error limit may be set as 20 °C and 30 °C respectively. This is based on a calibration done during the devising of this invention, using Ansys® motorcad simulation and testing. The example first error limit and second limits were obtained by: intentionally decreasing / increasing cooling system parameters and monitoring increases / decreases in rotor component / winding hotspot temperatures through test-calibrated motorcad simulation. In this way, an understanding was obtained of what amount of error in rotor component / winding hotspot temperatures can be caused by significant variation in the inputs. Figure 6B shows another flow chart illustrating a method 200 of predicting temperatures over time in an electric driver system 110 according to embodiments of the invention. In the embodiment of Figure 6B, only one of the first predicted time series and the second predicted time series were obtained. Determining whether a fault has occurred is therefore based only on one of the predicted time series. The method is a continuation of method 200 as illustrated by the dotted line and incorporates the blocks shown in Figure 2. The blocks of Figure 5 are also incorporated, and optionally the blocks of Figure 4. The method in Figure 6B proceeds with block 640, determining whether one or more data points of the first error time series exceed a first error limit. Alternatively block 640 comprises determining whether one or more data points of the second error time series exceed a second error limit. The method then proceeds with block 650, determining, if the firsterror limit has been exceeded, that a fault in the electric drive system 110 has occurred. Alternatively block 650 comprises determining, if the second error limit has been exceeded, that a fault in the electric drive system 110 has occurred. The method 200 then proceeds to block 660, outputting an indication to the one or more controller 120 of the vehicle that the fault has occurred. For other embodiments, there is no outputting of the indication to the one or more controller 120 as shown in Figures 6A, 6B. Instead the control system 100 has the functionality of the one or more controller 120 of the vehicle. The indication may be output to an infotainment system of the vehicle, and / or the control system 100 may take other action, e.g. limiting the operation of the electric drive system 110, in response to the fault occurring. The method 200 in the embodiments of 6A and 6B provide the advantage that not only is the electric drive system 110 monitored for predictions on the rotor 112 component and the winding hotspot temperatures to determine early fault detection, but real-time measurements of the rotor 112 component and winding hotspot temperatures are used to determine “live” faults. This again builds redundancy into the system 130. Predicting the temperatures and determining whether they exceed a temperature limit is done without using the real-time measurements of the rotor 112 component and winding hotspot temperatures. The system 130 therefore does not rely on real-time measurements only to predict / determine faults have occurred, which increases reliability. It has been found during the training of the model and during Design failure Mode and Effect Analysis (DFMEA) of the cooling system 116 that different types of faults can be diagnosed based on which error limits have been exceeded. For example, several issues such as leak of coolant from the cooling system 116, or low volume of coolant in the cooling system 116 lead to abnormal temperatures in the electric drive components. Figure 7 shows a flow chart illustrating a method of predicting temperatures over time in an electric drive system 110 according to embodiments of the invention in which the fault is diagnosed depending on which of the first error limit and the second error limit have been succeeded. Figure 8 shows another flow chart illustrating how faults are diagnosed. In Figure 7, the block 620 illustrates that this Figure is a continuation of method 200 shown in Figure 6A. The method then proceeds to one of blocks 632, 634,636 for outputting the indication according to block 630, depending on which fault is diagnosed. The dotted lines to each of the blocks 632, 634, 636 show that these blocks are alternative to one another. In Figure 7, the method additionally comprises: in block 632, outputting, if it is determined that the first error limit has been exceeded and the second error limit has not been exceeded, an indication that the fault is due to either a low flow of oil to a shaft of the rotor or that there is a high inlet oil temperature to the shaft of the rotor. In block 634, the method proceeds with outputting, if it is determined that the first error limit has not been exceeded and the second error limit has been exceeded, an indication that the fault is due either to a low flow of oil to the stator or there is a high inlet oil temperature to the stator. In block 636, the method proceeds with outputting, if it is determined that the first error limit has been exceeded and the second error limit has been exceeded, an indication that the fault is due to a low oil level in a sump of the electric drive system 110 or that a heat exchanger of the vehicle has low efficiency. As shown in Figure 8, the method illustrated in Figure 7 can alternatively be illustrated as a decision flow chart. In the Example of Figure 8 it is shown to determine whether the first error limit has been exceeded first, but the method could alternatively find whether the second error limit has been exceeded first. In Figure 8 at the first block 800, it is determined whether the first error limit has been exceeded or not. If the first error limit has been exceeded, the decision flow chart goes to block 820, and if the first error limit has not been exceeded, the method proceeds to block 810. In both blocks 810, 820, it is then determined if the second error limit has been exceeded. Starting from block 820, if the second error limit has been exceeded, the decision flowchart goes to block 636 as shown in Figure 7. The indication in block 636 is that the fault is due to a low oil level in a sump of the electric drive system or that a heat exchanger of the vehicle has low efficiency. Going back to block 820, if the second error limit has not been exceeded, the decision flowchart foes to block 632 as shown in Figure 7. The indication in block 632 that the fault is due to either a low flow of oil to a shaft of the rotor or that there is a high inlet oil temperature to the shaft of the rotor. Going to block 810, if the second error limit has been exceeded, the decision flow chart foes to block 634 as shown in Figure 7. The indication in block 634 is that the fault is due either to a low flow of oil to the stator or there is a high inlet oil temperature to the stator. If in block 810 the second error limit has not been exceeded, the decision flowchart goes to block 830, in which no error limit has been exceeded and so there is no output of an indication that a fault has occurred. The method illustrated in Figures 7 and 8 thereby provide a method which, as well as predicting key temperatures of an electric drive system 110 for monitoring and early fault or damage detection, also provides live fault detection. The particular types of error determined have been found to relate to specific problems of the electric drive system 110 operating outside of normal operating conditions. Figure 9 shows a flow chart illustrating a method 200 of predicting temperatures over time in an electric drive system 110 according to embodiments of the invention. The dotted line shows that this is a continuation of method 200 illustrated in Figure 6A or Figures 7 and 8. In Figure 9, the method additionally comprises in block 900, outputting a notification to an infotainment system of the vehicle, the notification indicating that the fault has occurred and / or indicating the warning that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded. Figure 10 shows a flow chart illustrating a method 200 of predicting temperatures over time in an electric drive system 110 according to embodiments of the invention, the dotted line shows that this is a continuation of method 200 illustrated in Figure 6A or Figures 7, 8 and optionally Figure 9. In block 1000, the method additionally comprises: limiting the maximum current provided to the coils of the stator 114 and / or limiting the maximum torque output of the electric drive system 110 in response to receiving, at the one or more controller 120 of the vehicle, the indication that the fault has occurred. Advantageously the vehicle therefore may change the operation of the electric drive system 110 in order to prevent further damage to the electric drive system 110, thereby reducing maintenance requirements. Figure 11 shows operation of a method 200 of predicting temperatures over time in an electric drive system 110 according to embodiments of the invention. As shown in Figure 11, the blocks of Figure 3 can be performed, resulting in the output of signal 106 to one or more controller 120 of the vehicle if the rotor 112 component temperature limit is predicted to be exceeded according to the first predicted time series and / or the winding hotspot limit is predicted to be exceeded according to the second predicted time series. Additionally, as shown in Figure 11, sensor measurements 119 are obtained and in block 1100 the first and / or second error time series are determined between the sensor measurements and the first predicted time series 310 and / or the second predicted time series 310. In block 1110 of Figure 11, it is determined whether any of the errors exceed a first error limit and / or second error limit. In block 1120, a fault is determined if either error limit is exceeded. The diagnosed fault may be determined as shown in Figures 7 and 8. The indication that the fault has occurred is then provided to one or more controller 120 of the vehicle. The one or more controller 120 may output to an infotainment system 1130 indication that a fault has occurred. The one or more controller 120 may also output an indication if the rotor 112 component temperature and / or winding hotspot temperature limit exceeded. In some embodiments, method 200 additionally comprises transmitting, to an electronic device via wireless communication: the indication that the fault has occurred and / or the warning that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded; the first predicted time series; and the second predicted time series. The electronic device is arranged to provide a Guided User Interface for displaying on a display of the electronic device: the indication that the fault has occurred and / or the warning; the first predicted time series; and the second predicted time series. Therefore a user of the electronic device can be informed of the indication(s) and / or fault(s). the user of the electronic device does therefore not have to be within and / or operating the vehicle to learn of the indication(s) and / or fault(s). As well as being beneficial for the normal operator (i.e. driver and / or owner) of the vehicle, it also benefits servicing people to determine what type of maintenance / repair could need performing on the vehicle without needing to enter the vehicle, and may even be provided before a vehicle arrives at the servicing location, which can increase efficiency of sourcing parts, booking appointment slots, etc.. In the embodiments presented herein, a rotor 112 component temperature is predicted and optionally is measured. In some examples, the rotor component can be one of: a rotor magnet; a rotor bar. For example it may be a rotor magnet for an electric drive system 110 with a permanent magnet synchronous machine (PMSM), or a rotor bar where the electric drive system 110 has an induction motor (IM). In some embodiments of the present invention, the multi-horizon forecasting model is a Temporal Fusion Transformer, which is trained similar to other training methods for deep neural networks. The Temporal Fusion Transformer provides other benefits over other multihorizon forecast models in that it has a series of features which increases performance of the model, as disclosed herein. Compared to general sequential neural network architectures, temporal fusion transformers work better when some exogenous outputs are not known in the future. With the use of the Temporal Fusion Transformer, dynamic temperature prediction made be made with greater accuracy and by the use of accessory data from the electric drive system 110. In the embodiments illustrated throughout, a trained multi-horizon time forecasting model is used which comprises a deep neural network that has been trained. Multi-horizon forecasts provide prediction of variables-of-interest at multiple future time steps. In contrast to one-step-ahead predictions, multi-horizon forecasts provide users and / or systems with access to estimates across the entire path, allowing them to optimize their actions at multiple steps in the future. In the embodiments illustrated throughout, The trained deep neural network comprises an input layer having input nodes for receiving an input vector ln = {h, I2,...., In}; an output layer having output nodes for outputting an output vector Om = {O1, O2,...., Om} at the output 15 nodes of the output layer; one or more hidden layers having connecting nodes connecting the input layer to the output layer, the connections of the connecting nodes having weights. Each value of the input vector ln takes a value of one of the plurality of electric drive system data inputs 118, so that all of the plurality of electric drive system data inputs 118 are fed in as the input vector ln. Each of the values of the output vector takes either a predicted temperature value of the rotor component or a predicted temperature value of the winding hotspot. The output vector may only have one value if only one predicted time series is being produced. The values of each In of an input vector ln and the values Om of the corresponding output vector Om relate to substantially the same moment in time. In practice, due to differences in sensors 119 used and variations in the electronics, there may be small deviations in the exact moment in time that each value of the plurality of electric drive system data inputs 118 was recorded. Over time successive values of each data input 118 are fed into the neural network, and a time series of the predicted temperature(s) are produced. Figure 12 shows a method of training a multi-horizon forecasting model according to embodiments of the invention. As mentioned above, the deep neural network has: an input layer having input nodes for receiving an input vector ln = {h, In}; an output layer having output nodes for outputting an output vector Om = {O1, O2,...., Om} at the output nodes of the output layer; one or more hidden layers having connecting nodes connecting the input layer to the output layer, the connections of the connecting nodes having weights. As illustrated in Figure 1200, the method begins at block 1210 with providing a store of training data comprising training example inputoutput pairs. Each input-output pair comprises an input vector ln = {h, I2,...., In} and a corresponding output vector Om = {O1, O2,...., Om}. Each In value of said input vector ln represents a recorded value of one of a plurality of electric drive system data inputs 118. Each Om value of said output vector Om represents a recorded temperature of a rotor 112 component of the electric drive system 110 and / or a recorded temperature of a winding hotspot temperature of the electric drive system 110 that corresponds to the recorded values of the plurality of electric drive system data inputs 118 in the input vector ln of the input-output pair. Each input-output pair corresponds to measurements collected from sensors 119 at a particular moment in time. In this example, the sensors 119 include temperature sensors for collecting temperatures of the rotor 112 component and / or the temperature of the winding hotspot. In some examples, each input-output pair represents data points for a plurality of recorded time series for each ln value and Om value. As mentioned above, the values of each ln of an input vector ln and the values Om of the corresponding output vector Om relate to substantially the same moment in time. In practice, due to differences in sensors 119 used and variations in the electronics, there may be small deviations in the exact moment in time that each value of the plurality of electric drive system data inputs 118 and the recorded temperatures was recorded. The training method comprises in block 1220 inputting the input vectors from one or more of the input-output pairs from the store of training data into the input nodes of the input layer of the deep neural network to produce a predicted output vector Pm = {Pi, P2,..., Pm} for each of the one or more input-output pairs, wherein each Pm value provides a predicted value of the corresponding Om value from the corresponding input-output pair. The training method further comprises in block 1230 determining an objective function characterising an error between the values of the predicted output vector Pm compared to the corresponding values of the output vector Om for each of the one or more inputoutput pairs. The method comprises in block 1240 using an appropriate optimisation algorithm operating on the objective function, updating the weights of the connections of the connecting nodes of the hidden layers of the deep neural network to seek to minimise the objective function. Table 1 shows example training data. Speed J r pm) O Torque -0.25 RotorGHF-ow JL / minJ 1.2 0.05 RotorOU Inlet (0} 6-3.7 StaiorQhhMet (C) RotorHotspot (C) WandlngHotspot (<) 67.25. 59 65.94 O -0..01 1.23 5.94 6-3.5 66.5 63.56 67,25 i O 1.23 5.94 60,5 56.5 63,16 67,25.: 0.0.3 -0.53 1.24 5.57 60..8 61.9 63,03 67.15: -2..53 -0.2 1.23 5.73 61.1 52,3 63.14 67.66 -12.28 -0,25 1.23 5.73 61.1 62.3 63.34 57.06 -2G.75 -013 X, — —r 5 32 61.5 62.5 63.40 56.95 -19.97 -0..13 5 33 61.5 52.0 63.4S 66-96 -19.55 -0.65 - iw 62,.5 52.9 63 56 66.85: -20.53 -G.15 •3.75 5 5 62...5 62.9 63.6 66.86: -207 -015 5.45 5 32 62.8 63 64.06 55.86: -3931 -0..1 015 5 22 52.3 63 64.20 -13.14 -0.15 5 45 5 63.3 63.1 64.49 56.88: -21.01 -0.1 0.45 -p 633 63.1 64.66 66.88: -21.22 -0.11 5.4 3 - 3 63,5 53.2 61,44 66.36 -18.32 -0.12 5.43 1 52 63.5 63,2 64.58 55.96 -23.1 -012 5.45 5 32 63.9 63.3 64.82 55.86 -20.15 -0,07 0 ^5 5 52 63.9 £3.3 64.86 56.86 -11.47 -0..45 0.44 -c 63.9 63.4 64.04 55,88: £.95 -0.3 0.44 5.70 63,5 63.4 64,74 55 85 20,44 -0.34 1.02 5.53 64,1 63.2 65,22 66.36: Table 1 As shown in the first six columns, the plurality of electric drive system data inputs 118 used for the values of the input vector were: a rotation speed of the rotor as a function of time; a torque output of the electric drive system as a function of time; a rotor coolant flow rate of the cooling system as a function of time (rotor oil flow); a stator coolant flow rate of the cooling system as a function of time (stator oil flow); a rotor coolant inlet temperature of the cooling system as a function of time (rotor oil inlet); a stator coolant inlet temperature of the cooling system as a function of time (stator oil inlet). As shown in Table 1, the values used for the output vector values were measured rotor component temperature (rotor hotspot) as a function of time and winding hotspot temperature as a function of time. Each row of Table 1 represents an input-output pair. During development of the embodiments of this invention, 30000 data points were used to train the specific multi-horizon forecasting model. In some embodiments, the objective function (also known as a loss function or cost function) can be a MSELoss function. In some embodiments, the objective function can be a Mean Squared Error function. In some embodiments, substantially the same type of electric drive system 110 or the same type of electric drive system 110 is used to train the multi-horizon forecasting model as it is implemented on. For example the electric drive system 110 used to train the model may have at least substantially the same type of rotor 112, stator 114 and cooling system 116. Figure 13A illustrates a graph showing predicted rotor 112 component temperatures provided by the trained multi-horizon forecasting model over time, with the predicted values shown in blue and the actual measured temperature shown in black. Figure 13A has temperature scaled to 1. Figure 13B also illustrates a graph showing predicted rotor 112 component temperatures provided by the trained multi-horizon forecasting model over time, with the predicted values shown in blue and the actual measured temperature shown in black. As shown in Figures 13A, 13B, the trained multi-horizon forecasting model produces very accurate predictions, with Figure 13C illustrating a graph that shows the error of the predicted rotor component temperatures compared to the measured temperature. The mean squared error in the graph shown in Figure 13C was 0.566°C. Embodiments of the invention have a Temporal Fusion Transformer as the multi-horizon forecasting model. The architecture of the Temporal Fusion Transformer is known, for example as detailed in “Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting”, Lim et. Al (University of Oxford, Google Cloud Al, September 29, 2020). Temporal Fusion Transformer (TFT) is an attention-based architecture that combines high-performance multi-horizon forecasting with interpretable insights into temporal dynamics. To learn temporal relationships at different scales, TFT uses recurrent layers for local processing and interpretable self-attention layers for long-term dependencies. TFT utilizes specialized components to select relevant features and a series of gating layers to suppress unnecessary components, enabling high performance in a wide range of scenarios. Owing to its unique architecture, TFT models supports types of features: I) temporal data with known inputs into the future ii) temporal data known only up to the present and ill) exogenous categorical / static variables, also known as time-invariant features. Being a multihorizon forecasting model, it can output multi-step predictions of one or more target variables — including prediction intervals. The use of specialized components also facilitates interpretability; in particular, TFT enables three valuable interpretability use cases: helping users identify (i) globally important variables for the prediction problem, (II) persistent temporal patterns, and (iii) significant events. In some embodiments, the Temporal Fusion Transformer is trained according to method 1200 described herein, and provides additional features over other multi-horizon forecasting models which provide further benefits and more accurate predictions. Firstly, the Temporal Fusion Transformer uses “Gating" mechanisms, within the hidden layers of the deep neural network to skip over unused components of the Temporal Fusion Transformer architecture. This provides adaptive depth and network complexity, and can determine unimportant, unused inputs. Nodes can be dropped to prevent overfitting. The Temporal Fusion Transformer also uses “Variable selection networks” within the hidden layers of the deep neural network. While multiple variables may be available as inputs to the deep neural network, their relevance and specific contribution to the output(s) are typically unknown. TFT is designed to provide instance-wise variable selection through the use of variable selection networks applied to both static covariates and time-dependent covariates. Using the variable selection networks means that all static, past and future inputs make use of separate variable selection networks with distinct weights. The Temporal Fusion Transformer also uses a self-attention mechanism within the hidden layers of the neural network, known as a “multi-head attention block” that identifies the long-range patterns the time series may hold and prioritizes the most relevant patterns. Each attention head can focus on a different temporal pattern. The multi-head attention block learns long-term relationships across different time steps, which can be used to modify from multi-head attention in other transformer-based architectures. The TFT has a temporal self-attention layer, which has the multi-head attention block followed by a gated layer. To overcome the major shortcoming of other Recurring Neural Networks (RNN) such as Long Short-Term Memory (LSTM) to capture long term dependencies in timeseries forecasting, a novel interpretable multi-head attention is used. The attention mechanism can be described as an attempt to mimic the action of selectively concentrating on a few relevant things instead of everything. The central idea behind attention is not to throw away any intermediate encoder states in LSTMs. Instead, these states are used to construct the context vectors required by the decoder to generate the output. The Temporal Fusion Transformer also uses Long Short-Term Memory (LSTM) encoders / decoders within the hidden layers of the neural network to summarize shorter patterns. The LSTM blocks of the Transformer are used to identify relationships of time steps with their surrounding values, whereas long-range relationships are left to the attention heads of the attention block. LSTM encoders and decoders are used between the Variable Selection blocks and the gated layer just before a static enrichment layer in a Temporal Fusion Decoder of the TFT. These Encoders and Decoders learn local temporal relationships present in the dataset from the past inputs and known future inputs. The temporal features generated by these Encoders and Decoders then serve as inputs into the Temporal Fusion Decoder itself where the temporal self-attention layer is present. The Static Enrichment layer referred to above, amongst other purposes, increases the significance of the short term temporal features so that these features are not lost while processing through later stages of the model architecture. Temporal Fusion Decoder referred to above is the region within the TFT architecture responsible for identification of long term temporal feature within the time series without compromising the short term temporal feature. This advantageously improves the accuracy of prediction. Both the number of LSTM layers and number of attention heads are model hyper-parameters and are tuned on test data while developing the prediction model. According to an example of training the TFT, the time-series for the input vector and output vector, as shown above in Table 1, are recorded for 50,000 times during the operation. From these, 60 percent of the data i.e., the initial 30,000 logged data points are used for training and remaining are used for model validation. Feature scaling can be used in this example so that all the variables are scaled to be between 0-1, since the dataset contains features that are varying in degrees of magnitude, range and units. The TFT model consists of the following hyperparameters which need to be selected by the user in order to obtain the required model performance for the dataset: 1) EPOCHS - denote the number of training cycles: one forward pass, followed by one backward pass of the entire training set. Usually, the model will have to go through several hundred epochs to calibrate itself and reach an acceptable accuracy. 2) INLEN - sets the input_chunk_length or size (node count) of the input layer (input vector). 3) HIDDEN - sets the size of the hidden layers. Too many hidden nodes can result in overfitting if the source data do not contain complex patterns. Too few nodes will oversimplify: the model will not catch all the significant patterns in a complex time series. 4) LSTMLAYERS - number of layers for the Long Short Term Memory encoder / decoder. 5) ATTHEADS - denotes the number of attention heads. 6) DROPOUT RATE - To prevent the model from overfitting, by digging itself deeper and deeper into a single configuration, the model should randomly select and switch off a fraction of its nodes during each of its training passes. Hence, a dropout rate is used. 7) BATCH SIZE - It is the number of training observations the model processes before it updates its weights. Its upper limit is defined by the available memory and the number of observations. 8) N_FC - denotes a chosen default number of forecast periods. The TFT model is set up with the hyperparameters listed above. For a deterministic forecast, a MSELoss function can be used. Then, the model is trained on the training data (first 30,000 points) using default values of the above hyperparameters. In each epoch or iteration, the TFT model tries to minimize the MSELoss associated with its two prediction time series and training data time series. The trained model can be further tuned using a Randomized Grid Search method to obtain the best deterministic parametrization (hyperparameters) of the TFT model, which is then validated on the validation data (remaining 20,000 points). Certain methods and systems as described herein may be implemented by a processor that processes program code that is retrieved from a non-transitory storage medium. Figure 14 shows a computer system suitable for implementing a method of predicting temperatures over time in an electric drive system in accordance with embodiments of the invention. Figure 14 illustrates an example of a computer system 1400 operable to implement the described methods according to embodiments of the invention. Computer system 1400 includes memory 1410, one or more processors 1420, network interface 1430, and non-transitory computer-readable medium 1440. The computer-readable medium 1440 can be any medium that can contain, store, or maintain programs and / or data for use by or in connection with an instruction execution system or other system for giving effect to instructions. Computer-readable medium 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 storage such as, for example, a USB drive or disk. In Figure 14, the computer-readable storage medium 1440 comprises program code to perform a method or implement a device or system as described herein. For example, the program code when executed may implement a method corresponding to the example shown in Figure 2. Figure 15 shows a vehicle 1500 in accordance with embodiments of the invention. Vehicle 1500 can comprise any system 130, control system 100, electric drive system 110, controller(s) 120 as disclosed in the embodiments of the invention. 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.

Claims

1. A computer-implemented method of predicting temperatures over time in an electric drive system of a vehicle, the electric drive system comprising a rotor, a stator and a cooling system, the method comprising:receiving a plurality of electric drive system data inputs;inputting the plurality of electric drive system data inputs into a trained multi-horizon time forecasting model comprising a deep neural network;obtaining, from the multi-horizon time forecasting model, at least one of a first predicted time series predicting a temperature of a rotor component, and a second predicted time series predicting a temperature of a winding hotspot of the stator;determining, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor component, and if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit;outputting a signal to one or more controller of the vehicle if the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and / or if the winding hotspot limit is predicted to be exceeded according to the second predicted time series.

2. A method as claimed in claim 1, wherein the plurality of electric drive system data inputs comprise any of: a rotation speed of the rotor as a function of time; a torque output of the electric drive system as a function of time; a rotor coolant flow rate of the cooling system as a function of time; a stator coolant flow rate of the cooling system as a function of time; a rotor coolant inlet temperature of the cooling system as a function of time; a stator coolant inlet temperature of the cooling system as a function of time.

3. A method as claimed in any preceding claim, wherein the method comprises obtaining both of the first predicted time series and the second predicted time series from the trained multi-horizon time forecasting model, wherein the method comprises outputting the signal to the one or more controller if one or both of the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and the winding hotspot limit is predicted to be exceeded according to the second predicted time series.

4. The method of any preceding claim, wherein the method additionally comprises:increasing the rotor coolant flow rate and / or the stator coolant flow rate in response to receiving, at the one or more controller of the vehicle, the signal that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded; and / orlimiting the maximum current provided to the coils of the stator and / or limiting the maximum torque output of the electric drive system in response to receiving, at the one or more controller of the vehicle, the signal that the rotor component temperature limit is predicted to be exceeded and / or the winding hotspot limit is predicted to be exceeded.

5. The method of any preceding claim, wherein the method additionally comprises:obtaining sensor measurements of the temperature of the rotor component;determining, when the first predicted time series has been obtained, errors between the measurements of the rotor component to corresponding data points of the first predicted time series to obtain a first error time series.

6. The method of any preceding claim, wherein the method additionally comprises:obtaining sensor measurements of the temperature of the winding hotspot of the stator;determining, when the second predicted time series has been obtained, errors between the measurements of the winding hotspot of the stator to corresponding data points of the second predicted time series to obtain a second error time series.

7. The method of claim 6, when dependent upon claim 5, wherein the method additionally comprises: determining whether one or more data points of the first error time series exceed a first error limit; determining whether one or more data points of the second error time series exceed a second error limit; determining, if either of the first error limit and the second error limit have been exceeded, that a fault in the electric drive system has occurred;outputting an indication to the one or more controller of the vehicle that the fault has occurred.

8. The method of claim 7, wherein the method additionally comprises:outputting, if it is determined that the first error limit has been exceeded and the second error limit has not been exceeded, an indication that the fault is due to either a low flow of oil to a shaft of the rotor or that there is a high inlet oil temperature to the shaft of the rotor;outputting, if it is determined that the first error limit has not been exceeded and the second error limit has been exceeded, an indication that the fault is due either to a low flow of oil to the stator or there is a high inlet oil temperature to the stator;outputting, if it is determined that the first error limit has been exceeded and the second error limit has been exceeded, an indication that the fault is due to a low oil level in a sump of the electric drive system or that a heat exchanger of the vehicle has low efficiency.

9. The method of any of claims 7 to 8, wherein the method additionally comprises: limiting the maximum current provided to the coils of the stator and / or limiting the maximum torque output of the electric drive system in response to receiving, at the one or more controller of the vehicle, the indication that the fault has occurred.

10. The method of any preceding claim, wherein the trained multi-horizon time forecasting model is a trained Temporal Fusion Transformer model.

11. Computer readable instructions which, when executed by a computer, are arranged to perform a method according to any of claims 1 to 10.

12. A control system for controlling an electric drive system of a vehicle, the electric drive system comprising a rotor, a stator and a cooling system the control system comprising one or more controller, the control system comprising one or more processors collectively configured to:receive a plurality of electric drive system data inputs;input the plurality of electric drive system data inputs into a trained multi-horizon time forecasting model comprising a deep neural network;obtain, from the multi-horizon time forecasting model, at least one of a first predicted time series predicting a temperature of a rotor component and a second predicted time series predicting a temperature of a winding hotspot of the stator;determine, if the first predicted time series is obtained, whether the first predicted time series contains one or more temperature values which are predicted to exceed a temperature limit of the rotor component;determine, if the second predicted time series is obtained, whether the second predicted time series contains one or more temperature values which are predicted to exceed a winding hotspot limit;output a signal to one or more controller of the vehicle if the rotor component temperature limit is predicted to be exceeded according to the first predicted time series and / or the winding hotspot limit is predicted to be exceeded according to the second predicted time series.

13. A system comprising the control system of claim 12 and an electric drive system.

14. A vehicle comprising the system of claim 13 or the control system of claim 12.

15. A method of training a multi-horizon forecasting model for predicting temperatures over time in an electric drive system ofa vehicle, the multi-horizon forecasting model comprising a deep neural network, the deep neural network having:an input layer having input nodes for receiving an input vector ln = {h, I2,...., In};an output layer having output nodes for outputting an output vector Om = {O1,02,...., Om} at the output nodes of the output layer;one or more hidden layers having connecting nodes connecting the input layer to the output layer, the connections of the connecting nodes having weights;wherein the method comprises:providing a store of training data comprising training example input-output pairs, wherein each input-output pair comprises an input vector ln = {h, I2,...., In} and a corresponding output vector Om = {O1, O2,...., Om}, wherein each ln value of said input vector ln represents a recorded value of one of a plurality of electric drive system data inputs, wherein each Om value of said output vector Om represents a recorded temperature of a rotor component of the electric drive system and / or a recorded temperature of a winding hotspot temperature of the electric drive system that corresponds to the recorded values of the plurality of electric drive system data inputs in the input vector ln of the input-output pair;inputting the input vectors ln from one or more of the input-output pairs from the store of training data into the input nodes of the input layer of the deep neural network to produce a predicted output vector Pm = {Pi, P2,..., Pm} for each of the one or more input-output pair, wherein each Pm value provides a predicted value of the corresponding Om value from the corresponding input-output pair;determining an objective function characterising an error between the values of the predicted output vector Pm compared to the corresponding values of the output vector Om for each of the one or more input-output pairs.using an appropriate optimisation algorithm operating on the objective function, updating the weights of the connections of the connecting nodes of the hidden layers of the deep neural network to seek to minimise the objective function.

Images