Artificial Intelligence (AI) has attracted increasing attention from the scientific and technological community, thanks to recent advancements in the “generative AI” space such as ChatGPT. The global market size for AI is expected to grow annually at a CAGR of 39.4% through 2028.1 Because AI is now more capable than ever of providing solutions to complex problems, several industries are already benefiting from this advanced technology.

What is AI and how does it work? 

As most readers are aware, AI is a technology that allows computers to learn from data, experiences, and situations, continually improving performance without being explicitly programmed. AI uses complex algorithms and mathematical models to process large amounts of data and identify patterns, trends, and relationships to make decisions or predictions.

In simple terms, AI is able to “learn” from a vast set of data and subsequently use that learning to solve problems and make decisions autonomously. The technology has revolutionized numerous industries, from the automotive to healthcare, financial analysis and logistics.

One of the sectors that is exploring AI with great interest is the electric power industry, both for mobility and energy storage. The power system is becoming increasingly complex and distributed, with many different sources of energy, and its management requires ever-greater efficiency and reliability. This is where AI comes into play to predict the demand and supply of electric energy and optimize the use of available resources.

In particular, AI can be used in a battery energy storage system (BESS) to optimize the use of stored energy, as well as in the design and manufacture of its components. This becomes even more crucial as companies and countries are setting sustainability goals in the coming years.

What are the challenges and opportunities for the future?

In this article, we will explore in detail some possible use cases in which AI can improve the efficiency and reliability of energy solutions, predict demand to ensure a constant supply, optimize the use of stored energy, and improve maintenance management and problem diagnosis. Here are some leading examples we’ve seen:

Electric mobility optimization: An interesting case study where AI can play a fundamental role is in electric mobility. According to Statista, U.S. Electric vehicle (EV) energy demand alone is expected to increase to about 107 terawatt hours by 2035, up from 4.7 terawatt hours in 2020. Thanks to AI systems, it’s possible to predict the energy demand for charging EV’s in a certain area, taking into account information on traffic, availability of charging stations, data from smart meters, and driver habits.

Identifying low energy demand periods for charging and programming the charging of vehicles to take advantage of these periods using demand management optimization models can reduce energy demand during peak consumption times. Detecting energy demand spikes in the power grid and providing signals to EV users to slow down charging during these periods can reduce the risk of grid overload and improve the stability of the system.

Peak shaving management: One of the main uses of a BESS is to store excess electrical energy when demand is low and release it to the grid when demand is high. This helps reduce the burden on energy generators and improve the stability of the grid.

To maximize the efficiency of energy storage systems, their operation needs to be coordinated with energy demand. This is where AI, particularly machine learning algorithms, becomes useful.

The use of AI in peak shaving management allows for improved accuracy in forecasting energy demand, thus increasing efficiency. This means that energy storage systems can be programmed more precisely to store and distribute energy, minimizing waste and maximizing benefits by using machine learning algorithms such as artificial neural networks, support vector machines, decision trees, and random forests.

To train AI for peak shaving management in energy storage systems, historical data on energy demand, charging and discharging activities, and grid behavior are utilized. Machine learning algorithms are then trained with this data to predict demand spikes and make recommendations to optimize systems.

Another exciting dimension occurs when AI models are fed with spot and future predicted pricing data from system operators such as the California Independent System Operator (CALISO) or the Electric Reliability Council of Texas (ERCOT). From the insights obtained from this data, AI can generate more precise predictions on energy demand and the need for charging or discharging the BESS. By identifying when prices are highest, BESS operators can maximize profits by targeting energy discharge during highly profitable windows during the day.

Battery health management: AI in BESS management can be used to predict faults and maximize performance and lifespan, thus increasing revenue.

To achieve these goals, machine learning algorithms need to be trained on historical operational data of the energy storage systems, such as temperature, battery charging and discharging levels, electric current, voltage, and pressure.

Gaussian process regression (GPR) is an algorithm used to predict battery health, based on a Bayesian approach so the identification of one observation depends on the previous one. As data points are gathered, they are inputted into the algorithm for future use, producing the “posterior” graph. Lastly, once all observations are made, the “prediction” graph can be made to predict hypothetical observations at different values of x as if they are made in the same series and therefore dependent on previous observations in the series.2

AI interprets this data to create predictive models on the degradation of the components of the BESS. These models are used to predict when a battery, inverter, or other component may fail, allowing system managers to schedule preventive interventions to avoid costly problems in the future.

AI can be further used to maximize the performance of the BESS by using information on system operation and degradation to adapt system programming to reduce degradation speed. For example, machine learning algorithms can suggest reducing battery discharge depth by predicting the state of charge (SOC), or reduce charging current to increase battery lifespan cycles. We’ve seen that degradation modeling based on neural networks compares favorably with electromechanical modeling, based on recent academic work at UC Berkeley.

Data can make or break your AI efforts

There are, of course, important challenges to overcome when using AI in the energy sector. Accurate and reliable data is essential for effectively implementing AI, and integrating into existing energy storage systems can be costly and complex.

To train AI to predict faults and maximize performance, large quantities of accurate and detailed operational data – both real-time and historical – must be collected. An at-scale battery farm, for example, can produce 50-100 million data points every second. Moreover, the collected data must also be accompanied by additional data on usage conditions, such as ambient temperature, frequency of charging and discharging, and utilization modes. Computational workloads with such large data sets can become demanding and require both big data architecture and active data curation strategies. Consideration must be given to performance and cost at every step.

This large quantity of accurate and detailed operational data is in turn used to train machine learning algorithms, which create predictive models on the degradation of energy storage system components. These models are used to optimize system programming and scheduling of preventive interventions, ultimately increasing system performance and lifespan.

Just because the data exists in an organization, however, doesn’t mean that it can be effectively used for AI applications. We’ve explored multiple sides of this topic in previous issues, including a look at the complexities involved in capturing battery assembly data, ensuring battery lot traceability, and the use of “manufacturing travelers” to automate data collection.

Threading and making use of all available data types can reveal statistical correlations in the data to seed machine learning algorithms in Peaxy’s Machine Learning Manager module, part of our battery analytics solution. See our previous article on anomaly detection using ML-enabled data pipelines for more information.

Organizations attempting an AI initiative are faced with questions around how much battery data to store, how to parameterize large amounts of data from across different systems and sources of input, and how to thread it correctly to enable traceability down to an individual serial number or unique identifier. These are just some of the areas where Peaxy has helped customers with their analytics efforts.

Some additional promising areas where AI can come into play include:

  • Battery design and development: By analyzing vast amounts of data and identifying patterns, AI assists researchers in discovering new materials and designs for batteries. A company we work with, for example, uses P2D modeling to optimize dispatch, leading to an increased cycle life in laboratory experiments by 50%. We’ve seen automation of this process lead to revenue increases of 20%.
  • Cybersecurity and grid protection: By detecting and predicting potential threats or vulnerabilities, AI helps safeguard energy storage systems and the wider power grid from cyber attacks.
  • Regulatory compliance and reporting: By automating data collection, analysis, and reporting processes, AI saves time and resources, ensuring operators meet regulatory requirements more efficiently.

If you’re contemplating an AI initiative in the area of battery analytics for manufacturing or BESS operations, it’s crucial to partner with a company that has a proven ability to ingest, aggregate, tag, and thread your data. Peaxy will work with you on the “first mile” data capture challenge by providing edge devices to ensure your analytics capabilities are comprehensive. These capabilities will ensure your “last mile” analytics in the areas discussed above are as robust and impactful as possible. Please reach out to us for details using the button below.

2 Sit, H. (2019, October 20). Quick Start to Gaussian Process Regression. Medium.