WO2020234743A1

WO2020234743A1 - Battery-powered propulsion unit for a vehicle and method thereof to estimate the state of charge

Info

Publication number: WO2020234743A1
Application number: PCT/IB2020/054683
Authority: WO
Inventors: Andrea Tonoli; Nicola Amati; Angelo BONFITTO; Francesco MONTI; Stefano FERACO
Original assignee: Politecnico Di Torino; Podium Engineering Srl
Priority date: 2019-05-17
Filing date: 2020-05-18
Publication date: 2020-11-26
Also published as: IT201900006987A1

Abstract

A vehicle propulsion unit comprises a battery having one or more electrochemical cells, the battery having an output voltage and an output current when delivering power; an electric motor driven by power delivered from the battery; a vehicle propeller driven by the electric motor; a battery output current, voltage and temperature sensing circuit; and a battery management system comprising processing circuitry coupled to the battery output current sensing circuit, the processing circuitry configured to model behavior of the battery with at least one recurrent neural network, the at least one recurrent neural network comprising a set of input layer nodes, a set of hidden layer nodes, and wherein at least some of the input layer nodes receives either an input representing a given state of charge of the battery or at least one input from a previous time state of an output layer node representing the estimation of the state of charge calculated by the processing circuitry.

Description

Battery propulsion unit for a vehicle and method thereof to estimate the state of charge

The present invention refers to a propulsion unit for a land, sea or aerial vehicle comprising a battery and an electric motor powered by the battery for the vehicle’s cruise motion, e.g. traction for a land vehicle.

A full electric or hybrid, including plug-in hybrid, vehicle comprises a battery to power a drivetrain and the battery performance and health are severely affected by the number and type of the charge/discharge cycles, and by environmental factors such as temperature and age. Moreover, a battery requires a constant and accurate monitoring to check its condition and, specifically, the level of the remaining available power, indicated by the State of Charge (SOC). An accurate and reliable knowledge of the SOC allows limiting psychological factors such as the range anxiety in electric vehicle. Additionally, it contributes to prevent accelerated ageing due to deep discharge, overheating and overcharging. However, the SOC cannot be directly measured, and its value can only be estimated from the measurement of other battery parameters, such as current, voltage, internal resistance and temperature. Common laboratory techniques for SOC estimation are based on the measurement of the open circuit voltage, internal impedance of the battery or on the current integration over time. Although accurate, these approaches, also known as direct methods, require energy dissipation, are not compatible with fast discharge rates and therefore are not suitable for real-time applications and vehicle applications, which cannot require the availability of a laboratory to analyze the battery. To overcome these issues, model-based, rule-based and artificial intelligence-based techniques have been proposed. The first family includes Kalman Filter (KF), Extended Kalman Filter (EKF) [12][13][14], Unscented Kalman Filter (UKF) [15][16], Adaptive Particle Filter (APF) [17], and Smooth Variable Structure Filter (SVSF) [18][19]. These solutions exploit the abovementioned direct methods for the tuning of the reference model and are heavily depending on its accuracy. Rule-based techniques, such as Fuzzy Logic (FL) [20][21][22] do not need any model but are strictly depending on the experience of the algorithm designer.

In this context, Artificial Intelligence-based solutions and in particular Artificial Neural Networks (ANNs), gained an increasing attention since they can be reliable and robust, independent from the cell chemical compounds, suitable for the representation of nonlinear dynamics and compatible with real-time applications. Nevertheless, the effectiveness and the accuracy of an ANN is strongly depending on the adopted architecture and on the selection of proper training datasets. An additional issue is that the resulting network could be too heavy in terms of memory occupation and processing time, demanding high performance computing units to be correctly deployed.

The analysis of the computational cost of the ANNs deployed on real control units and the validation on real driving test-cycles needs a deeper investigation. Indeed, several architectures have been proposed although with weak validations obtained by means of simple laboratory charge/discharge profiles and without addressing the computational cost analysis. Feedforward ANNs have been proposed in both a standard configuration [23][24] or combined with UKF [25] and EKF [11]. Such works presented a validation based on periodic laboratory discharge profiles. Additionally, their computational cost during training and deployment is not investigated. Similarly, an Elman ANN has been studied by [26] without a validation on real driving cycle profiles and no discussion on the weight of the algorithm.

In addition, US-A1-2018086222 discloses a Multilayer Recurrent ANN to model a battery and estimate the voltage and the temperature of a battery. However, in view of the lack of investigation, there appears to be room for focusing on an even more optimized neural network structure for the determination of the SOC.

The scope of the present invention is to provide an efficient and computationally optimized and simple architecture method to estimate the SOC of a drivetrain battery providing motion power to a vehicle.

The scope of the present invention is achieved by a vehicle propulsion unit comprising a battery having one or more electrochemical cells, the battery having an output voltage and an output current when delivering power;

an electric motor driven by power delivered from the battery; a vehicle propeller driven by the electric motor; a battery output current, voltage and temperature sensing circuit; and

a battery management system comprising processing circuitry coupled to the battery output current, voltage and temperature sensing circuit, the processing circuitry configured to model behavior of the battery with at least one recurrent neural network, the at least one recurrent neural network comprising a set of input layer nodes, a set of hidden layer nodes, and wherein at least some of the input layer nodes receives either at least one input representing a given, i.e. pre-defined, state of charge of the battery or at least one input from a previous time state of an output layer node, representing the estimation of the state of charge calculated by the processing circuitry.

Evaluation of SOC via an Autoregressive Neural Network according to the above provides a very efficient approach for the estimation of the SOC. Compared to other architectures, it is possible to use a limited number of nodes and obtain a precise estimation as shown by the benchmark example discussed below. In particular, the proposed ANN architecture of the invention shows a lower training time and better estimation performance when compared with other alternative, which will be discussed below in greater detail. Indeed, the feedback of the estimated output for the next calculation provides for a quite simple architecture, which is always preferable, without any compromise with the precision of the results.

According to a preferred embodiment of the present invention, the processing circuitry is configured to receive as given input of the input layer nodes after a new switch on of the unit, a previously stored value representing the state of charge when the drivetrain was last switched off before the new switch on.

Providing an initial value as input to the network avoids the risk of divergence. In particular, the initial value is the last estimation, i.e. the output of the proposed ANN, calculated immediately before the previous switching off of the vehicle. For example, as switching off is not always predictable, each estimated output value is stored in order to have a value ready after an abrupt switch off. In alternative or combination, when the propelling unit receives a switching off command e.g. from a command key operated by a user of the vehicle, such command also triggers the storing of the last estimated output for a later use as an initial input when the unit will be switched on.

According to a preferred embodiment of the present invention, the processing circuitry is configured to switch the input to the input layer nodes from the given state of charge value to the previous time state of the output layer node after a pre-defined and non-zero time interval.

This provides an adjusting parameter, i.e. the duration of the time interval, to optimize the proposed ANN.

According to a preferred embodiment, the hidden layer is a single hidden layer and hidden layer nodes are less than 10, preferably 8.

Such simple architecture is particularly optimized to estimate the SOC, preferably within a limited temperature range, e.g. from 10 to 40 degrees Celsius of the environment surrounding the battery, which is normally the case in vehicles including a heat exchanger to control the battery temperature.

There present invention is herein described according to non-limiting embodiments depicted in the attached drawings as mere explanatory examples. In particular:

Figure 1: shows a block diagram of a layout for training, test and validation of an ANN according to the present invention;

Figures 2 and 3: show a number of ANNs considered as a benchmark for the closed loop ANN according to the present invention wherein a) is a Feedforward Non-Linear Input-Output ANN, b) is a Multi-Layer Cascade Feedforward ANN, c) is a Recurrent Elman ANN, d) is a Multi-Layer recurrent ANN;

Figure 4: shows an ANN according to the present invention wherein a) is the open loop configuration for training and b) is the closed loop configuration for estimation;

Figure 5: shows Training profiles. a) Current charge/discharge. b) Voltage. c) Temperature. d) State of Charge;

Figure 6: shows test dataset obtained as a portion of the “United States Advanced Battery Consortium” (USABC) PHEV dynamic charge depleting duty cycle profile a) Current charge/discharge. b) Voltage. c) Temperature. d) State of Charge;

Figure 7: shows performance of the ANNs of figures 2 and 3, namely a) Feedforward Linear Input-Output. b) Multi-Layer Cascade Feedforward. c) Recurrent Elman ANN. d) Multi-Layer recurrent ANN. The SOC estimation and the Mean Square Error trend during the training process of each network are reported in the left-hand and right-hand column, respectively;

Figure 8: shows performance of the closed loop proposed ANN with 5 (a), 8 (b), 15 (c) and 20 (d) neurons or nodes in the hidden layer. The SOC estimation and the Mean Square Error trend during the training process of each network are reported in the left-hand and right-hand column, respectively;

Figure 9: shows a comparison of all the proposed architectures. a) Training time normalized with respect to the 100 Neurons proposed ANN. b) Maximum Relative Error (MRE) of estimation. c) Computational cost – program memory occupation. d) Computational cost – data memory occupation. The black circle indicates the ANN with the best performance, namely a closed loop ANN single layer of figure 4, 8 neurons;

Figure 10: shows validation on a real profile. a) Current charge/discharge. b) Voltage. c) Temperature. d) State of Charge: expected SOC (solid line), estimated SOC in simulation (dashed line) and estimated SOC on real electronic control unit (dash-dotted line);

Figure 11: shows graphs of a robustness analysis to the initial SOC estimation inaccuracies; in particular, a) shows the estimation output with error on the initial SOC ranging from -5% to +5%, b) the error between the expected and the estimated SOC, the grey area indicates the error tolerance region of ±5%, the solid lines represent the expected SOC and the two tolerance thresholds, the dashed lines are the SOC estimations;

Figure 12: shows graphs about robustness to the noise on the current measurement, a) current with noise of type 1, b) current with noise of type 2, c) SOC: expected without noise (solid line), estimated without noise (dashed line), estimated when the current is disturbed with the noise of type 1 (dash-dotted line) and estimated when the current is disturbed with the noise of type 2 (dotted line); and

Figure 13: shows a sketch of a propulsion unit controlled by the ANN of the present invention.

Figure 1 discloses a typical design flowchart of ANNs applied to the estimation of the SOC according to the present invention and comprising three phases: training, test and validation. The training phase is conducted exploiting datasets reproducing the widest possible range of the system dynamics to allow an accurate learning. In the case of the proposed method, the training datasets include the current, voltage and temperature measurements as network inputs and the SOC as the output target. Since the SOC is a not measurable quantity, its value is obtained from a look-up table battery model tuned by means of Ampère-hour counting method performed in a laboratory environment.

This model (block 1 in Figure 1) is then used as an emulator of the real battery to generate the expected SOC reference. The ANNs architectures (block 2) according to the invention are preferably designed and simulated in the Matlab/Simulink environment and then deployed on a real electronic control unit with a Hardware in the Loop (HIL) configuration (block 3) based, according to an embodiment, on an Atmel AVR 8-bit processor that has calculation performance comparable to that of common Battery Management System (BMS) and is in serial communication with a computer.

The design procedure can be summarized as follows: a) the training input dataset (current, voltage and temperature) is provided to the block 1 to obtain the training output target (SOC_1); b) the block 2 is fed with the obtained input/output dataset to perform the ANN training procedure; c) the test and validation steps are conducted by providing new datasets to the three blocks and comparing the outputs from the reference model of block 1 with the trained ANN model of block 2 and block 3. Feedforward ANNs do not require the feedback of the output estimation, while the recurrent ANNs have the feedback of the SOC estimation as the fourth input (dashed line in Figure 1).

As an example, the adopted battery pack is composed of 168 cells KOKAM SLPB 11543140H5 in the configuration 12p14s (p: parallel, s: series). The pack has a nominal voltage of 48 V, a nominal capacity of 60 Ah and is designed for a mild-hybrid electric vehicle with a peak electric power of around 20 kW, obtained considering a discharge rate of around 7C in nominal conditions. The main characteristics of the cell are reported in Table 1.

.

Typical Capacity (@0.5C, 4.2 V÷2.7 V, 25 °C)		5 Ah
Nominal Voltage		3.7 V
Cut-off voltage		2.7 V
Continuous current		150 A
Peak current		250 A
Cycle life (Charge/Discharge @ 1C)		> 800 cycles
Charge condition	Max. Current	10 A
	Voltage	4.2V ± 0.03 V
Operating Temperature	Charge		0 °C – 40 °C
	Discharge	-20 °C – 60 °C
Mass		128.0 ± 4 g
Dimension	Thickness	11.5 ± 0.2 mm
	Width	42.5 ± 0.5 mm
	Length	142.0 ± 0.5 mm

Figures 2 and 3 show several benchmark architectures taken as a comparison for the architecture of the invention, depicted in figure 4.

In particular, figure 2a and 2b show the respective feedforward configurations where the information flows in only one direction from the input to the output nodes without forming any recurrent cycle.

The Non-Linear Input-Output network (Figure 2a) is the simplest configuration included in this study, and here it is designed with a single hidden layer consisting of 10 neurons or nodes, d_x=2, a sigmoid HAF and a linear OAF. Essentially, the output y(n) of this network is the result of applying a non-linear function φ to the inputs x(n) according to the following equation:

y(n)= φ[x(n-1),x(n-2),… ,x(n- d_x )] (1)

where φ is the non-linear function modeled with the ANN.

The Multi-Layer Cascade Feedforward network (Figure 2b) exploits the formulation defined by Eq.(1). Each layer takes as input both the input and output of all previous ones. Further theoretical details can be found in [30]. The proposed configuration includes 3 hidden layers (12, 10 and 8 neurons or nodes, respectively), no delays on the inputs, hyperbolic tangent and linear as activation functions for the hidden and output layers, respectively.

In the recurrent Elman ANN (Figure 2c), the output of the hidden layer is fed back as input of the same layer. The designed network features 8 neurons or nodes in the hidden layer, d_x=1 and d_y=1, sigmoid and linear activation functions for the input and output layers, respectively. Basically, this network produces an output y(n) according to the following equation:

y(n)= φ[x(n- d_x),; ϕ(x(n- d_y))] (2)

where φ is the non-linear function modelled by the ANN, and ϕ is the function that the hidden layer applies to the input x(n). A detailed theoretical background can be found in [31] and in [32].

The Multi-Layer Recurrent network (Figure 2d), is designed with three hidden layers (5, 4 and 2 neurons or nodes, respectively), d_x=2 and d_y=2, a sigmoid HAF and a linear OAF. The output y(n) is computed by applying a non-linear function φ, modelled by the ANN network, to the inputs x(n), as in the following equation:

y(n)= φ[x(n-1),x(n- d_x ); ϕ₁ (x(n-1),x(n-d_y ));…; ϕ_i (x(n-1),x(n-d_y))] (3)

where the output of each i-hidden layer is fed back as an additional input of the same layer, and this layer applies its function ϕ_i to its inputs. A thorough background on recurrent ANNs can be found in [33].

The last architecture (Figure 4) is the ANN according to the present invention. This network is represented with a discrete nonlinear model and is commonly used for time series prediction tasks. It is mathematically defined as:

y(n)=φ[y(n-1),y(n-2),…,y(n-d_y);x(n-1),x(n-2),…,x(n-d_x)] (4)

where y(n)∈

and x(n)∈

denote the outputs and inputs of the proposed ANN model at the discrete timestep n, respectively. d_x and d_y are the input and output memory used in the model, respectively, and φ is the function, generally non-linear, represented by the ANN. During the proposed ANN regression procedure, the next value of the dependent output signal y(n) is regressed on the previous d_y values of the output signal and previous d_x values of the independent input signal.

According to the present invention, the proposed ANN is adopted in open loop (Figure 4a) during the training process and in closed loop (4b) during the estimation phase, i.e. when the network is deployed on the real application, in particular a control circuit receiving input signals from a plurality of sensors including respective current, voltage and temperature sensors applied to the battery, the latter being designed to provide power for driving, e.g. propelling, a land or sea or aerial vehicle via a drivetrain.

The open loop configuration is called Series-Parallel (SP) mode. In this configuration, the output regressor is:

y(n)=φ[

(n-1),

(n-2),…,

(n-d_y );x(n-1),x(n-2),…,x(n-d_x)] (5)

In this case, since the true output is available during the training of the network, it is used directly, instead of feeding back the estimated output. This has some advantages: the input to the artificial neural network is more accurate and the resulting network has a purely feedforward architecture, so a static backpropagation algorithm (Levenberg-Marquardt) can be used for the training process.

The closed-loop configuration is called Parallel (P) mode, where the output estimation is:

y(n)=φ[y(n-1),y(n-2),…,y(n-d_y);x(n-1),x(n-2),…,x(n-d_x)] (6)

When using the network in the Parallel mode, at the start of the computation, the estimation has an undetermined value and cannot be fed back to the ANN input because it may generate the divergence of the estimation over time. To overcome this issue, during an initial pre-defined time interval of computation e.g. of 1 second, the feedback signal is replaced by a constant value SOC_INIT (Figure 4b) that is the last SOC value recorded on a non-volatile memory at the previous shut down of the drivetrain and/or the vehicle. After such initial time interval, i.e. when the output estimation is stable, the input related to the SOC switches to the feedback signal.

Referring to Figure 4b and denoting by n=n₀ the time instant when the feedback signal switches from constantSOC_INIT to the estimated output, the characteristic equations of the model are written as:

y(n)=φ[SOC_INIT;x(n-1,x(n-2),…,x(n-d_x))],n<n₀ (7)

and

y(n)=φ[y(n-1),y(n-2),…,y(n-d_y );x(n-1),x(n-2),…,x(n-d_x )],n≥n₀ (8)

The adopted training dataset is obtained from a current charge/discharge profile of a real EV reported in [34]. The dataset is shown in Figure 5, where the values of the voltage (b), temperature (c) and SOC (d) are obtained from the reference model (the block 1 in Figure 1) when the input of the model is the current a). The data are sampled with a frequency of 10 Hz. Since this profile reduces the SOC of about 12,5%, thus, the ANN learning is performed repeating it 8 times to train the network on a full discharge with the SOC going from 1 to 0.

The test dataset is a portion of the “United States Advanced Battery Consortium” (USABC) PHEV dynamic charge depleting duty cycle profile [28] reported in Figure 6. It is adopted to perform the training and estimation accuracy evaluation along with the computational cost analysis of the five proposed ANN architectures.

The above mentioned ANNs are tested using the dataset reported in Figure 6. The aim is to evaluate the performance in terms of duration and precision of the training process, estimation accuracy and computational cost, namely memory and processor occupation, when the designed algorithms are deployed on an electronic control unit. Finally, the estimation accuracy and robustness of the ANN with the best performance is evaluated with an additional profile obtained from a real electric vehicle.

The first test is conducted to analyze the estimation accuracy and the training precision combined with its duration in number of training epochs for each ANN. The estimation accuracy is analysed by means of the Maximum Relative Error (MRE), computed in % as:

(9)

where n is the number of the measurements, SOC_exp and SOC_est are the expected and estimated SOC, respectively.

On the other hand, the training accuracy is evaluated by means of the Mean Square Error (MSE) obtained at the end of the learning process. The training is stopped when the MSE is equal or lower than 1e-13 or, alternatively, when it remains constant for a sufficiently large number of training epochs.

Figure 7 reports the results obtained by the Feedforward Linear Input-Output, Multi-Layer Cascade Feedforward, Recurrent Elman ANN and Multi-Layer recurrent ANN. For each of these networks, the training process is stopped because the training MSE is not reducing after one hundred epochs. Each ANN is tested in the configuration with the best compromise of number of neurons, layers, delays and activation/training functions. Different configurations, i.e. with larger number of neurons and layers, have been tested obtaining worst results in terms of training duration and the same accuracy (or worst if overfitting occurs). The plots on the left-hand of Figure 7 show that none of these networks provides a sufficiently accurate SOC estimation. Although the recurrent ANNs (Elman SRN and Multi-layer RNN) are characterized by slightly better results with respect to the feedforward ANNs (Linear Input-Output and Multi-Layer Cascade), the zoomed portions in c) and d) put in evidence that the estimated SOC (dashed line) results to be inaccurate (MRE around 5%) if compared to the expected one (solid line).

The performance of the proposed ANNs with one layer and 5, 8, 15 and 20 neurons or nodes in the hidden layer are presented in Figure 8. The training of each network is stopped because the MSE reaches an enough small value. The network with the best performance is the proposed ANN of figure 4 with 8 neurons (c). It has the minimum training MSE (9.8e-14), reached with the lowest number of training epochs, and the minimum estimation error MRE (0,35%).

The results obtained with all the proposed architectures are summarized and compared in Figure 9, where the proposed ANN with 8 neurons is marked with a black circle. The plots in the first row show the performance of the networks in terms of duration of the training process (a) normalized with respect to a proposed ANN with one layer and 100 neurons and of SOC estimation in terms of MRE (b). In this second plot the results of the proposed ANN with 50 and 100 neurons are not reported because of the huge simulation time required with performance that deteriorates at increasing the number of neurons. In general, proposed ANN architectures show a lower training time and better estimation performance when compared with other solutions. The results reported in c) and d) allow comparing the tested networks in terms of computational cost when they are deployed on an electronic control unit with an Atmel AVR-8 bit microcontroller, similar to the ones adopted in common BMS.

Finally, the accuracy of the proposed ANN with 8 neurons is tested on a profile obtained from a real electric vehicle [33]. The results are reported in Figure 10 where the dynamics of the current profile (a) are similar to those of the training dataset (Figure 5.a) but reveals a more aggressive driving style with peaks reaching 200 A. The SOC estimation (d) is accurate with the MRE lower than 0.35% as expected from the results illustrated in Figure 8. The solid line in d) represents the expected SOC, while the dashed and dash-dotted lines are the estimated SOC in simulation and on the control unit, respectively. The good correspondence between these last two, highlighted in the zoomed area, is a validation of the correctness of the deployment setup. It is worthy to notice that the estimation is never higher than the expected value, which is a conservative situation since it allows avoiding the over-estimation of the residual energy on the battery that can be a critical situation in EVs.

The final test conducted on the proposed ANN with 8 neurons aims to evaluate its robustness when the initial value of the SOC (SOC_INIT in Figure 4b) is not accurate and when the current measurement is affected by noise, that is a typical condition in real applications on board of a vehicle.

The first analysis is carried out by introducing a relative error in the maximum tolerance range of ±5% on the SOC_INIT and evaluating the capabilities of the network to recover this error or at least to keep it limited in between the lower and upper tolerance thresholds. The test is conducted adopting a portion of the validation dataset reported in the zoomed area of Figure 10d, testing the robustness on a range of ±5% centred on SOC=95%. The results are reported in Figure 11 where the estimation behaviour and the trend of the error are reported in a) and b), respectively and the error tolerance region is coloured in grey. The plots show that for errors lower than 4%, the estimation tends to converge to the expected value or to remain constant, while for higher errors, the estimation can diverge and exceed the tolerance range.

The second analysis is conducted by disturbing the measurement of the current provided as input to the network. Two different types of noise are summed up to the current profile of Figure 10: a 1 kHz pseudo-random Gaussian noise having zero mean value and standard deviation equal to 1.5 A (type 1, Figure 12a) and a 100 Hz pseudo-random Gaussian noise having zero mean value and standard deviation equal to 5 A (type 2, Figure 12b). The results are reported in (c), where the solid line is the expected value, the dashed line is the estimation without noise and the dash-dotted and dotted lines are the estimation affected by the type 1 and type 2 noise, respectively. The results show that the estimation performance is not affected by problems of noise on the current measurement that is typically the most disturbed signal in the real applications. Obviously, the ANN has not the capabilities to effectively compensate possible inaccuracies in the offset and gain calibration of the current sensors.

Figure 13 shows a sketch of a propulsion unit 110 comprising the battery, which may be a multiple cell battery or the like, a battery management system 160 comprising the circuitry described in the preceding paragraphs, an inverter 120 powered by the battery and connected to a current controller 130, an electric motor 140 driven by the inverter and a vehicle propeller 150 attached to electric motor 140. The propeller may be either directly driven by electric motor 140 such as in the case of a land vehicle wheel in a motor wheel, or a helical propeller in a sea or aerial vehicle, or a driveline may transfer mechanical power from electric motor 140 to propeller 150, such as a driveline in a hybrid traction vehicle, e.g. a hybrid car.

Controller 130 can be used to generate gate signals for the inverter 120. Accordingly, control of vehicle speed is performed by regulating the voltage or the flow of current from the inverter 120 through the stator of the motor 140. There are many control schemes that can be used in an electric vehicle drive system including current control, voltage control, and direct torque control. Selection of the characteristics of inverter 120 and selection of the control technique of the controller 130 can determine efficacy of the drive system.

It is also worth noting that the invention is able to estimate both a discharge of the battery due to propulsion and a recharge of the battery either during recharging from an electric power station or via an on board electric machine, such as a generator driven during a regenerative braking operation on a land vehicle. After a number of estimations, errors may propagate so that it is important to have a condition, i.e. the full charge condition, where there is high confidence that the estimation of the proposed ANN is particularly precise. Such condition is indeed the state of full charge. Therefore, whilst estimation of a SOC level after a partial recharge may be affected by a certain error range, according to the proposed ANN, it is verified that the estimation of the full charge state is very precise. Each time the full charge estimation feeds the input layer nodes, such value is very precise and this helps to avoid drifts and decrease of precision over time.

CITATIONS: Non Patent Literature

[1] Bishop J.D.K., Martin N.P.D., Boies A.M., Cost-effectiveness of alternative powertrains for reduced energy use and CO2 emissions in passenger vehicles, Applied Energy, Vol. 124, 44-61, 2014.
[2] Walther G., Wansart J., Kieckhafer K., Schneider E., Spengler T.S., Impact assessment in the automotive industry: mandatory market introduction of alternative powertrain technologies. System dynamics review, Vol 26, No. 3, 239-261, 2010.
[3] Ahman M., Assessing the future competitiveness of alternative powertrains, International Journal of vehicle design, Vol 33-4, 2003.
[4] Chan C.C and Chau K.T., Modern Electric Vehicle Technology, Oxford University Press, New York, 2002.
[5] Anderman M., Status and trends in the HEV/PHEC/EV battery industry, Rocky Mountain Institute, 2008.
[6] Chen X., Shen W., Tu Vo T., Cao Z., Kapor A., An overview of lithium-ion batteries for electric vehicles, IEEE IPEC Conference on Power and Energy, 2012.
[7] Leksono E., Haq I.N., Iqbal M., Soelami F.N., Merthayasa I., State of Charge (SoC) Estimation on LiFePO4 Battery Module Using Coulomb Counting Methods with Modified Peukert, IEEE 2013 Joint International Conference on Rural Information & Communication Technology and Electric-Vehicle Technology.
[8] Chang W.Y., The State of Charge Estimating Methods for Battery: A Review. Applied Mathematics
Vol 2013, 2013.
[9] Rivera-Barrera, J.P.; Muñoz-Galeano, N.; Sarmiento-Maldonado, H.O. SoC Estimation for Lithium-ion Batteries: Review and Future Challenges. Electronics 2017, 6, 102.
[10] Wei Z., Zhao J., Ji D., Tseng K. T., A multi-timescale estimator for battery state of charge and capacity dual estimation based on an online identified model, Applied Energy, Vol. 204, 1264-1274, 2017.
[11] Charkhgard M., Farrokhi M., State-of-Charge Estimation for Lithium-Ion Batteries Using Neural Networks and EKF, IEEE Trans. on Industrial Electronics, Vol. 57, No. 12, 2010.
[12] Jiang C., Taylor A., Duan C, Bai K., Extended Kalman Filter based battery state of charge (SOC) estimation for electric vehicles, in IEEE Transportation Electrification Conference and EXPO (ITEC), 2013.
[13] Pérez G., Garmendia M., Reynaud J. F., Crego J., Viscarret U., Enhanced closed loop State of Charge estimator for lithium-ion batteries based on Extended Kalman Filter, Applied Energy, Vol. 155, 834-845, 2015.
[14] Wang S., Fernandez C., Shang L., Li Z., Li J., Online state of charge estimation for the aerial lithium-ion battery packs based on the improved extended Kalman filter method, Journal of Energy Storage, Vol. 9, 69-83, 2017.
[15] He Z., Chen D., Pan C., Chen L., Wang. S., State of charge estimation of power Li-ion batteries using a hybrid estimation algorithm based on UKF, Electrochimica Acta, Vol. 211, 101-109, 2016.
[16] Yu Q., Xiong R., Lin C., Online estimation of state-of-charge based on H infinity and unscented Kalman filters for lithium ion batteries, Energy Procedia, Vol. 105, 2791-2796, 2017.
[17] Ye M., Guo H., Xiong R., Yang R., Model-based state-of-charge estimation approach of the Lithium-ion battery using an improved adaptive particle filter, Energy Procedia, Vol. 103, 394-399, 2016.
[18] Kim T., Wang Y., Sahinoglu Z., Wada T., Hara S., Qiao W., State of Charge Estimation Based on a Realtime Battery Model and Iterative Smooth Variable Structure Filter, IEEE Innovative Smart Grid Technologies – Asia, 2014.
[19] Zou Z., Xu J., Mi C., Cao B., Chen Z., Evaluation of Model Based State of Charge Estimation Methods for Lithium-Ion Batteries. Energies, 7, 5065-5082, 2014.
[20] Jiani D, Zhitao L, Youyi W, Changyun W, A Fuzzy Logic-based Model for Li-ion Battery with SOC and Temperature Effect, 11th IEEE Conference on Control & Automation (ICCA), Taiwan, 2014.
[21] Li H., Wang W., Su S., Lee Y., A Merged Fuzzy Neural Network and Its Applications in Battery State-of-Charge Estimation, IEEE Trans. Energy Convers., vol. 22, no. 3, pp. 697–708, 2007.
[22] Lin F.J, Huang M.S., Yeh P.Y., Tsai H.C., Kuan C.H., DSP Based Probabilistic Fuzzy Neural Network Control for Li-Ion Battery Charger, IEEE Trans. Power Electron., vol. 27, no. 8, pp. 3782–3794, 2012.
[23] Cai C.H., Du D., Liu Z.Y., Zhang Z., Artificial neural network in estimation of battery state of-charge (SOC) with nonconventional input variables selected by correlation analysis, Proceedings of International Conference on Machine Learning and Cybernetics, 2002.
[24] Soylu, E.; Soylu, T.; Bayir, R. Design and Implementation of SOC Prediction for a Li-Ion Battery Pack in an Electric Car with an Embedded System. Entropy 2017, 19, 146.
[25] He W., Williard N., Chen C., Pecht M., State of charge estimation for Li-ion batteries using neural network modeling and unscented Kalman filter-based error cancellation, Electrical Power and Energy Systems Vol. 62, 783–791, 2014.
[26] Shi Q., Zhang C., Cui N., Zhang X., Battery State-Of-Charge estimation in Electric Vehicle using Elman neural network method, IEEE Control Conference (CCC), 2010.
[27] Chaoui H, Ibe-Ekeocha C. C., State of Charge and State of Health Estimation for Lithium Batteries Using Recurrent Neural Networks, IEEE Transactions on Vehicular Technology, Vol. 66, No. 10, 8773-8783, 2017.
[28] Fan G., Pan K., Canova M. A Comparison of Model Order Reduction Techniques for Electrochemical Characterization of Lithium-ion Batteries, IEEE 54th Annual Conference on Decision and Control, 2015.
[29] Moré J.J., The Levenberg-Marquardt algorithm: Implementation and theory. In: Watson G.A. (eds) Numerical Analysis. Lecture Notes in Mathematics, vol 630. Springer, Berlin, Heidelberg, 1978.
[30] Warsito B., Santoso R., Suparti, Yasin H., Cascade Forward Neural Network for Time Series Prediction, Journal of Physics: Conf. Series 1025 012097, 2018.
[31] Seker S., Ayaz E., Turkcan E., Elman’s recurrent neural network applications to condition monitoring in nuclear power plant and rotating machinery, Engineering Applications of Artificial Intelligence, Vol. 16, pp. 647–656, 2013.
[32] Elman J. L., Finding Structure in Time, Cognitive Science, Vol. 14, pp. 179-211, 1990.
[33] Caterini A., A Novel Mathematical Framework for the Analysis of Neural Networks, UWSpace, 2017.
[34] Rezvanizanian S.M., Huang Y., Chuan J., Lee J., A Mobility Performance Assessment on Plug-in EV Battery, International Journal of Prognostic and Health Management, Vol.11, 2012.

Claims

A vehicle propulsion unit comprising:
a battery having one or more electrochemical cells, the battery having an output voltage and an output current when delivering power;
an electric motor driven by power delivered from the battery;
a vehicle propeller driven by the electric motor;
a battery output current, voltage and temperature sensing circuit; and
a battery management system comprising processing circuitry coupled to the battery output current sensing circuit, the processing circuitry configured to model behavior of the battery with at least one recurrent neural network, the at least one recurrent neural network comprising a set of input layer nodes, a set of hidden layer nodes, and wherein at least some of the input layer nodes receives either an input representing a given state of charge of the battery or at least one input from a previous time state of an output layer node representing the estimation of the state of charge calculated by the processing circuitry.
The propulsion unit according to claim 1, wherein the processing circuitry is configured to receive as given input of the input layer nodes after a new switch on of the unit, a previously stored value representing the state of charge when the drivetrain was last switched off before the new switch on.
The propulsion unit according to claim 1 or 2, wherein the processing circuitry is configured to switch the input to the input layer nodes from the given state of charge value to the previous time state of the output layer node after a pre-defined and non-zero time interval.
The propulsion unit according to any of the preceding claims, wherein the hidden layer is a single hidden layer and hidden layer nodes are less than 10, preferably 8.
Control method for a propulsion unit comprising:
a battery having one or more electrochemical cells, the battery having an output voltage and an output current when delivering power;
an electric motor driven by power delivered from the battery;
a vehicle propeller driven by the electric motor;
a battery output current, voltage and temperature sensing circuit; and
a battery management system comprising processing circuitry coupled to the battery output current sensing circuit, the method comprising the steps of:
modeling a behavior of the battery with at least one recurrent neural network, the at least one recurrent neural network comprising a set of input layer nodes, a set of hidden layer nodes, and
estimating via the processing circuitry a state of charge of the battery so that at least some of the input layer nodes receives either an input representing a given state of charge of the battery or at least one input from a previous time state of an output layer node representing the estimation of the state of charge calculated by the processing circuitry.
The method of claim 5, wherein the step of estimating further comprises the step of receiving as given input of the input layer nodes after a new switch on of the unit, a previously stored value representing the state of charge when the drivetrain was last switched off before the new switch on.
The method of claims 5 or 6, further comprising the step of switching the input to the input layer nodes from the given state of charge value to the previous time state of the output layer node after a pre-defined and non-zero time interval.
The method according to any of claims 5 to 7, wherein the hidden layer is a single hidden layer and hidden layer nodes are less than 10, preferably 8.