Power generation is rapidly being integrated with IoT. AI is driving this and, perhaps, the most exciting technique is digital replica, or digital twin (DT).
Sanand Sule,
CTO Climate Connect Technologies
Wherever data, technology, and the need for smart decision-making overlap, Artificial Intelligence (AI) starts to make an impact. Whether determining the fastest route via Google Maps, early diagnosis of cancer cases, or customer behaviour analysis for targeted e-commerce, AI has exciting potential across an array of sectors; but its applications in power systems and the wider sector are particularly important. Because this will underpin the transformation of many other sectors and by extension the country, it is still early days for the adoption of AI. However, though, it has the capacity to drive the ‘Make in India’ campaign. By accelerating deployment of renewable power, helping to ramp-down fossil-fuel plants, and reducing the national fuel import bill, this will increase national energy independence, ultimately enabling the cheapest power possible.
AI is a broad term encompassing a wide array of concepts. But for the power sector it has two core pillars. The most recognisable of these is physical robotics such as flying drones for tracking and cleaning. More excitingly, ‘bots’ are already being developed for diving and sailing to enable construction and maintenance of offshore wind and solar plants which are much further out in the sea.
The other pillar, which is less visible but more vital, is machine learning. This is the ability of machines to learn without being explicitly programmed to do so. It is how the ‘intelligence’ of AI is gained. A machine is provided with huge amounts of data and certain parameters. It then sets about detecting patterns, associations, and insights far beyond human capacity. The ML-based tools, algorithms and modelling techniques of AI are starting to transform the power sector; particularly, renewable power generation and grid load management. Prominent examples being the impact of techniques such as Support Vector Machine (SVM) and deep learning based Artificial Neural Networks (ANNs). Power plant generation should be considered together with demand forecasting for grid load management. AI can crunch the huge volumes of data that are changing on a second-by-second basis to better match supply and demand, and so greatly reduce overall grid-volatility.
ANN based models can analyse the DC and AC data streams concurrently through each inverter. If there is a reduction in AC values with no corresponding reduction in DC values, this indicates a potential inverter fault. Lost energy may be turning into heat and raising the inverter temperature, or there may be a ventilation issue. An alarm is automatically, then, raised by the system to schedule a maintenance check. This form of prediction and complex decision-making can help reduce the downtime of equipment. In extreme cases, it can even prevent fires that could occur because of this increasing heat.
To maintain a high functioning power plant, scheduling of tasks such as physical inspections, and module-cleaning sessions should be automated. For example, whenever a dip is detected in the string efficiency, a physical check is scheduled on that basis or when there is a dip in performance ratio (PR) at a solar plant, a reduction may be detected by the system over consecutive days with similar irradiance, but high wind conditions. This indicates soiling of the panel, so the system then raises an alarm to schedule a cleaning session. That is on the surface of panel, but detection can also be for factors within the panel. For example, for physical stress or condition of internal chemicals,the angle and rotation of the panels can then be adjusted for optimal generation.
Different groups of bots will use various forms of ML to enable different types of optimisation. Another example being image analysis through Convolutional Neural Network (CNN). This can identify if a rat has chewed through a cable or a wire is about to touch the ground and cause earthing. So, it still takes some significant effort to employ AI appropriately as careful tailoring is required for each use-case.
However, the knowledge gained in individual plants can give a cumulative benefit through another ML technique called transfer learning (TL). TL inductive transfer is a ML design approach which uses knowledge gained whilst solving one problem and applies it to a different but comparable problem. The most common application of TL is in image recognition; it is exactly what Google uses. When commissioning of a new plant is considered, there is typically no historical data to work with. TL can use a generalised model based upon cumulative learnings from previous deployments to markedly reduce uncertainty for new sites.
Renewable generation plants already have a wide-span of sensor technology installed, providing enormous amounts of data. Most plants do provide good quality data, yet much of this remains largely under-analysed. Most plant operators still use spreadsheets to manage reporting and analysis. Amongst other problems, this can cause reliability issues such as invalid data entries. Solar and wind plants are for the most part closed systems. So, if enough sensors (including cameras for image analysis) are installed, everything aside from weather can be captured and accurately monitored. AI can then mitigate a wide spectrum of risks thorough monitoring, supervision, forecasting, and analysis. Pattern recognition and data anomaly detection can be used to identify subtle but significant trends that might indicate an upcoming component or system failure.
Like many sectors, power generation is rapidly being integrated with internet-of-things (IoT) technology. AI is driving this and, perhaps, the most exciting technique is digital replica, or digital twin (DT). This can provide a virtual copy that enables exact comparison with any real system and have enough sensors installed which will keep a DT fully up-to-date. Mapping drones can support this to give the exact specifications for multiple plants, right down to component-level. For a power plant, DT can be used to detect and isolate faults, perform diagnostics and troubleshooting, recommend corrective actions, and determine the ideal maintenance schedule. DT also provides a user-friendly interface, especially compared to what is currently available. Custom screens can be created down to panel and component level. This will enable even better optimisation and resource allocation.
Whilst AI is perceived as a future technology, when associated with renewables, it is simply an operational tool that enables optimisation. It can be applied to all generation types, including legacy thermal plants. It would be remiss not to mention this because the applications are equally significant. On a basic level, there is the obvious working of equipment such as pumps and boilers. Health and safety is also a more imperative concern than for renewable plants, as there are far higher risks from maloperation. However, it is the highly complex core combustion process itself where the greatest and widest benefits can be felt. AI can fine-tune the optimal temperature and conditions for power production with minimised carbon emissions. In this way, AI can help to manage the ramp-down of our fossil fuel use and national environmental impact.
AI has huge potential to improve plant performance, limit component downtimes and reduce operational costs; in-turn enabling more profitability for plant operators and a more stable supply on the grid-side. It can help manage and optimise each node of an interconnected ecosystem, whilst retaining human-oversight in place of heavy manual-intervention. This is the ultimate ambition for all AI techniques.
