Having begun the analytics journey over a decade now as a research associate at IISc, Deepika Sandeep is currently the senior program manager of AI and ML at BLP Clean Energy, a company operating in renewable energy management and Industry 4.0 space. She has worked in areas such as training machine learning models to recognise handwritten characters, build OCR system, neuroscience application of machine learning, natural language processing for pharmacovigilance and more.
In her current role, she is responsible for devising innovation strategy and analytics roadmap for the organisation, analytics solution implementation, building and mentoring analytics teams. She also takes care of the development and deployment of analytics, AI and BI components of our product.
Analytics India Magazine got in touch with Sandeep to get an insight into how energy companies are using technologies such as analytics and AI, state of energy analytics in India, challenges faced by the company and more.
AIM: How are the energy companies resorting to technologies such as cloud and analytics to come up with a better and greener country? Please highlight 2-3 important instances?
DS: Primary source of data at renewable energy power plants is the SCADA system, in which all the data from sensors mounted on assets, get congregated. Our data lake resides on Google Cloud Platform. Predictive analytics on data is carried out using analytics models, on the cloud and the outcomes of analyses are stored in the output data store on the cloud. This interfaces with BI tools for real-time visualisation and reporting to facilitate remote, live monitoring of assets.
Analytics tools help the Operations and Maintenance (O&M) teams to conduct remote health assessment on a real-time basis. Live reporting of asset performance through various Key Performance Indicators, derived using statistical computing, allows them to keep a tab on each component’s condition in each individual asset. Also, the fact that energy generation is affected by myriad factors such as weather, asset performance in history, the lifespan of assets etc., necessitates multivariate analysis which is possible only through machine learning and high-speed computing.
AIM: How is the state of energy analytics companies in India?
DS: In alignment with the Indian government’s mandate of growing renewable energy to 175 GW by the year 2022, the process of commissioning of renewable energy projects has become quicker and easier. Adoption of analytics and AI in the renewable energy industry is still at its infancy. For AI to enter the mainstream in the renewable energy sector, a change of mindset and willingness to embrace data-driven solutions, among owners and operators of farms is imperative. However, with the decline of margins, asset owners are increasingly feeling the need for remote monitoring, early warning systems and automation. This helps them cut OpEx significantly. Hence, a steep rise in the number and penetration of analytics firms in the renewable energy sector is foreseen in the next few years.
AIM: How is BLP Clean Energy using data analytics and AI to provide clean energy to the world? Please highlight some use cases?
DS: We, at BLP, started out as asset owners and our digital journey began as a consequence of a fire incident of one of our own assets. Owing to this incident, we realised that plain vanilla monitoring does not suffice and predictive analytics is the need of the hour. Together with our deep domain know-how, we leverage AI and ML to predict critical component failures, breakdowns and other abnormalities with the help of self-learning models trained on historical data. We provide timely, actionable insights for more informed decision-making by site engineers. We also have a robust energy forecasting solution in place, to provide day-ahead generation forecasts and intraday revisions using ensemble methods for forecasting, based on trends in historical generation data and weather forecasts. In order to reduce uncertainty in demand and generation and to ensure safe and reliable operation of the grid, system operators rely on load and generation forecasts to balance electricity supply and demand.
AIM: What does the analytics tool-kit at BLP Clean Energy look like?
DS: Input raw data collated from IoT devices are stored in a data ingestion layer on the cloud, comprised of GCP ecosystem components like BigQuery, Bigtable, FileStore and MySQL databases. Analytics and machine learning models are implemented using Python. The output storage layer is made up of relational schema on MySQL and a time series database, InfluxDB. For building dashboards that represent analytics insights to end users, we employ open-source visualisation tools like Grafana and Google Data Studio.
AIM: How important is it to make machines smarter in today’s digital revolution age? How is BLP Clean Energy ensuring it?
DS: In today’s era of connected devices, it is essential for every device or asset in a power plant to be smart and able to communicate with other networked devices. In the context of the renewable energy industry, this means the ability to connect all elements of power production and consumption, remotely gain visibility into operations at the site and provide control at every step of the energy pipeline from the source to the supply. With the advent of IoT infrastructure, this connected ecosystem has become the norm. The vision for our organisation is to transform every renewable energy power plant into a digital power plant. We make this possible by being able to remotely help improve the productivity of assets and control their operations, through mobile applications on-site engineers’ smartphones or other handheld devices.
AIM: What are the intelligent solutions you are providing to ensure sustainable and efficient energy usage and generation?
DS: We create solutions for energy consumption pattern recognition in manufacturing industries and provide a recommendation on the optimisation of energy consumption. Another example is that of an automation solution to furnish optimal cleaning recommendation of solar modules based on their condition, rather than the conventional practice of a periodic cleaning schedule on a round-robin basis. This solution aims at alleviating losses in generation due to dust accumulation on modules, in utility-scale solar power generation plants and automation of module cleaning schedule based on how soiled various modules are, thence minimising operational expenditure incurred in manually cleaning modules.
AIM: With loads of data that you might be dealing on a daily basis, how do you ensure efficient use of this data? What are the various data points you collect?
DS: Process automation is key in this context. The end-to-end process of data flow is automated, right from data collection at the site, data storage locally at the site, edge computing, data storage on the cloud, processing and analysis on the cloud, visualisation and reporting. Real-time alerts are automatically issued through a closed loop workflow on our mobile application. We ensure a tangible value proposition from analytics on our product, through our constantly evolving analytics landscape, in terms of new use case additions, based on data collected.
Varied sources of data points collected include key parameters necessary to monitor the performance of renewable energy assets which are collected as part of SCADA data, configuration details such as design, layout, capacity, geo-locations of assets and farms, weather data and maintenance records on component repairs and replacements.
AIM: Please tell us about Orion, the AI-based IoT platform by the company. How is artificial intelligence and machine learning crucial to it?
DS: Orion is a predictive intelligence platform, which helps convert power plants from being reactive to being proactive, using the strength of big data analytics, machine learning and industry expertise. We capitalise on the fact that we were asset owners and have collected data from self-owned farms, for a considerable duration of time of over six years now and have in-depth domain expertise in this space. Our machine learning algorithms are able to exploit an ample amount of data to learn from and predict failures in future, based on the behavioural patterns learnt from past data.
AIM: What is the roadmap for analytics at BLP Clean Energy for this year?
DS: We are looking to tap on to external sources of data, especially unstructured data in the form of images captured by drones, satellites, etc., and using this in congruence with the data we already have in hand, for more efficient generation forecasting and defect detection solutions. The objective is to leverage deep learning for visual analytics, in order to deliver a more engaging end-user experience. We are also evaluating additional value-adds from new age technologies like digital twins for Industry 4.0 solutions. Boosting automation through robotic process automation for asset management is on our agenda as well.