Design and Implementation of Hybrid Energy Storage System Integrating Lithium-ion Battery and Wind Turbine

A hybrid energy storage system (HESS) by integrating Lithium-Ion Battery and Wind Turbine System for Electric Vehicle is designed and implemented. An advanced model of lithium ion/wind turbine HESS model is developed to improve the battery’s charging capacity. The behavior of this model is tested by using Machine Learning Algorithms to identify remaining battery capacity of the energy storage components. The Machine learning Algorithms such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF) Decision Tree (DT) and Linear Regression (LR) are implemented which helps in predicting the State of Charge (SoC) in the electric vehicle to estimate the battery’s lifetime. The SoC of the EV without HESS is calculated and compared with the SoC of HESS implemented EV model. Sensors are used to measure the current and voltage parameters of the Lithium-Ion Battery setup (Internal Battery) and Storage Battery (External Battery). The measured data is sent to the LPC2148 ARM microcontroller Board. The data that is obtained from LPC2148 is given as input to the machine learning algorithms which are executed in the Computer. Performance metrics such as Mean Absolute Error (MAE), Root Mean Square (RMSE), R-Square Error (R2E) and Mean Square Error (MSE) are calculated and the SoC values are predicted. The Random Forest algorithm produces better results for the HESS model as the Error values produced are less than 2.5%. The battery capacity of the electric vehicle has been improved by using external battery that is charged by wind turbine. Thus, the driving range of the electric vehicle implemented with HESS is increased twice when compared to conventional electric vehicle [1].


INTRODUCTION
In electric vehicles, the energy storage system used are high cost, limited in range, slow in recharge. The electric vehicle development on these days is providing solutions for improving the lifetime of these storage systems, driving range, sizing specifications. To improve EV's lifetime and efficiency that is embedded with ESS, a lithium-ion battery is integrated with wind turbine. The HESS model contains two batteries where one battery is the Lithium-Ion battery setup already present in the electric vehicle; the other battery is the storage battery which is charged from the wind turbine. EV's range is based on the SoC of battery [7].
SoC is the percentage of remaining charge available in the battery. SoC is calculated using the battery parameters current, voltage and temperature taken from both the batteries in HESS model. The LPC2148 controller is used to acquire the battery parameters from sensors. The LPC2148 controller is programmed to obtain and store the real time dataset from the sensors. The dataset is then uploaded to the Jupyter Notebook through an USB module connected to the controller. The machine learning algorithms such as Support Vector Machine, Decision Tree, K-Nearest Neighbor, Random Forest and Linear Regression are used to calculate the remaining charge. The machine learning metrices are used in calculating performance of the algorithms. These algorithms provide the aging by predicting the time until which the battery charge lasts.

II. BACKGROUND WORKS
An effective method to predict aging of the battery is proposed by Prakash et al (2021). Through real time estimation battery health can be predicted accurately. and State of Health (SoH) and Remaining Useful Life (RUL) are predicted by calculating SoC of Lithium-Ion battery. The Machine learning Algorithms are used to calculate the SoH of the battery. The results obtained from the neural network algorithm shows the error rate of ±5%. The LSTM algorithm works better in predicting RUL with ±10 cycles of accuracy [2].
Meshabi et al (2021) et al proposed an advanced electrothermal modelling of a HESS integrating lithiumion batteries and supercapacitors. They have used particle swarm-Nelder-Mead (PSO-NM) optimization algorithm to evaluate the performance of the model. Electrothermal behaviors are analyzed in the hybrid system to estimate the performance. The experimental data is taken from the urban EV. The optimization algorithm produces high efficiency and the error obtained is less than 3% [3]. Lee et al (2021) proposed an energy management strategy for hybrid EV's namely Equivalent Consumption Management Strategy (ECMS) to reduce energy consumption. A reinforcement learning algorithm namely Deep Q-Networks (DQNs) algorithm is implemented to evaluate the control parameter for minimal energy consumption. The costate is estimated accurately using Deep Q-Networks (DQNs). The results IOP Publishing doi:10.1088/1757-899X/1291/1/012042 2 perform better than the already existing adaptive technique for the control concept. Thus, the proposed study is feasible in optimizing the energy consumption [4].
Huilong et al (2021) proposed a HESS which combines lithium-ion batteries and supercapacitors. This method helps to overcome the drawbacks of ESS including short cycle life, low power density and high cost. Optimized HESS and better power management system can achieve better driving range and life cycle. In this paper Pareto optimization technique is used which provides both optimal design and efficient control. Battery Life is significantly increased with the current solution [5].
Min et al (2021) proposed an adaptive battery SoC estimation method for EV by using BMS technique. BMS manages charging and discharging rate of battery which provides the state of battery. SoC provides the battery state which need to be predicted accurately. In this paper to predict SoC a Kalman-filter (EKF) based method is used. Conventional Coulomb counting method and EKF is combined to predict SoC which shows less than 2% error in predicting SoC [6].
An energy management along with the optimization approach is proposed by Zhou et al (2020) for plug-in hybrid electric vehicle (PHEV) with a HESS containing a Ni-Co-Mn battery pack and Li-Ti-O battery pack. The performance of this system is tested using a test bench. Combination of both Energy management system and hybrid energy management system in PHEV has increased the efficiency up to 2.5% and the lifecycle is improved up to 203% [9].

A. FLOW DIAGRAM
A flow of the proposed system is shown in Figure 1. The dataset is divided into testing and training dataset. The training data is the web data and the test data is the real-time data. Two different types of test data are fed into the algorithm. The first set of data is obtained from battery without integration of HESS model and the other data is evaluated from the battery in HESS model. The training data is given as input to ML algorithms. After training the dataset the testing is done which are then given to the model for predicting the battery's capacity. block diagram of the proposed system. The ARM 7 is the microcontroller used to get real-time data from sensors. Through kinetic energy the wind turbine is rotated. According to the speed of the wind turbine charge is generated and stored in the external storage battery. When the electrical motor in the EV runs in low speed the charge is taken from the external storage battery and in high-speed internal battery is used. The temperature of the battery is measured by LM395 sensor. The ACS712 sensor is used to measure the current discharge from the battery and the voltage of the battery is measured by voltage sensor. The data from LPC2148 Microcontroller is collected by CPC2102 USB to TTL module and stored in spreadsheet. Python IDLE is used for implementing the Machine learning algorithms to predict the aging of battery. Both classification and regression models use support vector machine. Decision boundaries are used in SVM where decision planes are predicted that are referred as hyperplane. The hyper plane separates two different set of points which has different class in the plane. The distance between the hyperplane and nearest points from either class is called is called as margin. The classification is done by figuring out the hyperplane that has the largest margin between the two classes. [7].
D. K-NEAREST NEIGHBOURS The K-nearest neighbours works by classifying the data greatest number of common points according to the distance. The data point that needs to be predicted are considered from K numbered nearest points which belong to a specific class. The greatest number of points that belong to the points are the considered as the predicted values.
E. DECISION TREE (DT) This algorithm works by creating a tree like structure. This tree structure is used to perform both classification and regression. The dataset is then divided into smaller nodes that are combined with the decision tree algorithm. The decision nodes and leaf nodes combine and form the tree structure. The nodes are the test data and the branch of the tree forms the result of the test data [8].
F. RANDOM FOREST (RF) Random Forest algorithm helps in solving both regression and classification problems. Ensemble learning concept is used by RF algorithm. The RF algorithm takes the average value of numerous decision subsets to improve the precision of the dataset and provides the accurate prediction.
G. LINEAR REGRESSION (LR) Linear Regression is based on supervised learning algorithm which performs regression for the given dataset. Independent variables are taken to identify the prediction values. Relationship between variables and forecasting are found out mostly using LR algorithm. For the given number of datasets LR algorithm plots the best fit line in the best possible way to fit the model.

A. EXPERIMENTAL SETUP
The experimental setup of the proposed system is illustrated in Figure 3. Sensors connected to the GPIO pins of LPC2148 MC collects the data from the current, voltage and temperature sensors. The data are collected from both internal and external battery. Data from LPC2148 is given to CPC2102 USB to TTL module which sends the data to an excel sheet displayed in PC.

E. ANALYSIS OF REGRESSION METRICES
The mathematical metrics which are used to identify the performance of the software model proposed is given below. Mean Absolute Error: MAE provides absolute difference between actual and predicted values.
(1) Where is the i'th expected value in the dataset, is the i'th predicted value. Mean Square Error: MSE is the squared difference between actual and predicted value.

MSE =^2 (2)
Root Mean Square Error: Where sqrt() is the square root function. R-Square Score: R-Square Value is a metric that tells the performance of your model. R 2 = 1 − sum squared regression / total sum of squares (4) SST = (5) Where y is each observed value minus the average of observed values Analysis Report for Regression Metrices Table 6.2 show that the performance of the different algorithms for both data sets in terms of the above defined metrics. The analysis table shows that Random Forest algorithm produces less than 2.5% of Mean Absolute Error when tested using both the datasets. To conclude the performance of the algorithms, MAE is considered and shows less than 2.5% error for Random forest algorithm.  Table 2 shows the comparison of regression metrices V.
CONCLUSION AND FUTURE WORK A. CONCLUSION In this paper the energy storage system is designed in order to predict the capacity of the Li-Ion battery. Which was developed and the ML algorithms used predicts the aging of the Battery. The Regression metrics such as Mean Absolute Error (MAE), Root Mean Square (RMSE), Mean Square Error (MSE), R-Square Error(R2E) are calculated and the SoC values are predicted. The Random Forest algorithm produces better results for the HESS model as the Mean Absolute Error values produced are less than 2.5%. This method helps the in estimating the driving range of the EV by predicting the remaining battery capacity. This technique of predicting the remaining charge is highly applicable for increasing the driving range during real time implementation.

B. FUTURE WORK
In further this work can be extended by implementing this technique using Deep Learning algorithms. The electric vehicle model can be created using simulation tool and the deep learning algorithms such as Convolutional Neural Networks (CNNs) , Artificial Neural Networks (ANNs) and Recurrent Neural Networks (RNNs), may produce more accurate results in estimating the remaining charge.