Transforming the EV Charging Market: The Potential of AI and ML

It can be arguably said that the future belongs to EVs, be it cars, bikes, or eVTOL aircraft for urban air mobility. In fact, according to Statista, the electric vehicle market is anticipated to achieve $858 billion, with an expected compound annual growth rate of 17.02%. Moreover, with the artificial intelligence revolution at its peak and stretching over many industries and applications, the EV market is sure to get a boost. 

 

Electric vehicle charging. Image used courtesy of Pexels

 

The battery is the most important component of an electric vehicle, as it defines the vehicle's shelf life. In the EV revolution, the optimization of artificial intelligence (AI) and machine learning (ML) technology is required for efficiency and effectiveness; therefore, the spotlight is on the use of AI and ML.

This article will explore the ways in which AI and ML are set to make an impact on the EV industry.

 

Battery Performance and Charging Efficiency

Battery performance is the most important factor to consider when it comes to vehicles. The battery holds 25% of the total cost of an electric vehicle, so increasing the battery life is a priority. 

AI and ML evaluate the battery performance of electric vehicles based on several variables, such as transmission utilization trends, acceleration patterns, deceleration patterns, headlights, and HVAC. Artificial Intelligence has also been used to make EV charging faster, compared to the time it takes for a traditional vehicle to stop at a gas station. 

 

Increasing Battery-Life-Prediction Accuracy

Machine learning has opportunities in battery lifecycle management. By combining advanced electronics with the Internet of Things, digital twins, and data science, machine learning uses predictive intelligence to anticipate battery life, identify issues such as breakdowns, leakage, and heating, and determine causes before such issues occur. 

Moreover, the U.S. National Renewable Energy Laboratory (NREL) applied ML algorithms to prove how ML can deliver increased life-prediction accuracy and demonstrated how ML algorithms offered more accurate battery life predictions.

Battery lifetime models with a reduced order use algebraic and differential equations to simulate the physical degradation processes. These models are easy to interpret and have good extrapolation abilities from small datasets. They have the added advantage of being quick to compute. 

However, accurately identifying the model equations can be challenging. Often, researchers do not statistically justify selected models compared to alternatives. To tackle this issue, NREL has implemented machine learning (ML) algorithms to automatically generate thousands of equation components and narrow down millions of possible combinations to select a simplified model that strikes a balance between predictive accuracy and simplicity.

The ML algorithm identified pertinent physical equations similar to those employed by professionals but without any preconceived notions, resulting in a "rediscovery" of the equations.

 

Battery aging data. Image used courtesy of NREL

 

The above images show the prediction of expert and ML models in different temperature conditions. Two models are used to analyze aging batteries: 

(i) A model based on expert judgment from literature, represented by a black line with a gray 90% confidence interval. 

(ii) A machine learning model shown in red. Based on the ~8 months of training data, the ML model outperforms the literature model in terms of accuracy. When projected into the future, the ML model predicts a longer calendar life of 40% to 130%, depending on the aging condition.

 

Reducing Battery Costs

When compared to conventional internal combustion vehicles, EVs have failed to capture the interest of consumers primarily due to their higher upfront costs, limited driving range, and insufficient supporting infrastructure. To address the issue of high upfront costs, it is crucial to reduce the battery pack's cost.

Machine learning offers a cost-effective and time-efficient method for identifying low-cost and high-performing battery materials, as well as improving battery-manufacturing efficiency. The utilization of AI algorithms and controllers can enhance the precision of estimating driving range and improve energy management, resulting in extended driving range and reduced apprehension among consumers regarding "range anxiety."

 

Choosing the Optimal Location for EV Charging Stations

Electric vehicle charging stations (EVCS) are categorized into Level 1 (120 V) (240 V), Level 2, and DC fast charging based on charging voltage. 

Tesla has installed 1,971 DC fast charging stations globally for user convenience. Level 1 chargers are placed in parking lots, while DC fast chargers are placed near high-volume road access points. 

The ideal positioning of charging stations is crucial in reducing range anxiety and increasing the acceptance of EVs. The EVCS placement depends on factors such as charging demand, infrastructure, economy, and power-grid security. To identify the ideal placement, multiple objective optimization functions (MOOPs) can be employed, including minimizing costs, maximizing net present value, and prioritizing placement in unpopulated regions. 

To determine MOOPs, computational intelligence (CI) techniques like swarm intelligence and evolutionary algorithms can be utilized. In addition, machine learning algorithms can be utilized to prepare data or models. ML clustering algorithms can be used to form zones, and ML methods are used to model EV user demand and traffic occupancy. 

Determining the best location for electric vehicle charging stations can be achieved by addressing the MOOP using computational intelligence techniques such as particle swarm optimization and genetic algorithms. However, ML algorithms like K-means clustering, Bayesian networks, and neural networks can also be utilized for identifying the optimal charging site without needing to formulate a MOOP. 

Furthermore, agent-based models provide a more realistic approach to determining placement and sizing, as they account for different agents and their characteristics. agent-based models have been effectively used to simulate and analyze the potential impact of EVCS placement and charging capacity on EV adoption and charging usage over short and long periods, facilitating the development of effective EVCS deployment strategies.

Easier access to charging stations and smart grids can change the game for the EV industry. AI can help with the minimal congestion at charging ports, as AI can control and evaluate the resources and sites for the charging stations.   

Recently, Google announced a new feature for EV owners that uses AI to locate the right EV station and best route.

 

The role of AI in the mass adoption of EVs. Image used courtesy of Science Direct

 

Better Battery Management 

Battery management systems (BMS) are critical components in electric vehicles as they ensure the safe and energy-efficient operation of the battery pack. For battery-state estimation and diagnosis, the BMS analyzes the voltage, current, and temperature of individual battery modules to calculate the state of charge (SOC) and state of health (SOH).

However, these estimations are challenging due to inconsistencies in battery-cell performance and nonlinear battery characteristics. While traditional methods, such as equivalent circuit models (ECM) and electrochemical models (EM), can estimate SOC and SOH, these models can be computationally expensive and less accurate. Machine learning techniques like neural networks and support vector machines have shown promise in improving SOC and SOH estimations with low computational requirements. 

Recurrent neural network (RNN)-based models are useful in evaluating battery history and dynamic aging, and LSTM and RNN-GRU architectures can capture long-term battery characteristics. Hybrid models combining ML and ECM/EM can also provide accurate predictions. 

Despite advancements, there are still hurdles to overcome in battery degradation modeling. For instance, pure machine learning models struggle with representing battery aging mechanisms, and external operating conditions that impact battery degradation are varied and complex. Additionally, most research is cell-level rather than battery-pack-level. 

BMS is also responsible for minimizing the risks of battery failure during EV operations, and ML approaches are being developed for early detection of location, time, and cause of battery cell or module failures in an EV. Faults in battery cells can be caused by operational abuse or mechanical damage, and detecting mechanical faults requires changes in vibration.

ML-based state estimation is promising in improving battery management in EVs due to its lower computational demand and accuracy, but more research is needed to address existing challenges in real-time EV applications. Additionally, early detection of battery cell or module failures can be achieved using ML approaches, but detecting mechanical faults requires changes in vibration, and research is ongoing.

 

The Effect of AI and ML on EVs

The increasing use of AI and ML in everyday life cannot be denied. They have the potential to reformulate every aspect of our lives. AI and ML provide key solutions for reducing challenges related to EV adoption and expanding the environment-friendly automotive system, which not only enhances performance efficiency but also user safety. 

eInfochips has expertise in autonomous vehicles, EV charging, V2X, telematics, infotainment, and multimedia and helps its global clients with safety and efficiency.