Customer Intelligence Data-driven analytics for understanding utility customer behavior Arjen Zondervan (Alliander, Liander Klant & Markt) Maarten Wolf (Alliander, Liandon) Sasha Aravkin (IBM Research) 1 Alliander Intro Alliander is an energy network group, distributing electricity and gas to around three million households in the Netherlands. 2 SERI Collaboration Alliander joined the SERI collaboration in 2012 to develop an advanced analytics competence and create more value from analytics. • Smarter Energy Research Institute (SERI) • Collaboration of three utility companies Alliander, DTE (Detroit) and Hydro Quebec • IBM Research (Watson Lab) as knowledge partner on advanced analytics and facilitator of the collaboration • Multidisciplinary team within Alliander: 3 Business units ‘Customer & Market’ (‘Klant & Markt’), Asset Management and IT • Alliander SERI team works in two streams: Asset Management models and Customer Intelligence (CI, topic of this presentation) • Goal of the project for Alliander: develop an advanced analytics competence, while creating business value through the models that are developed. Why Customer Intelligence? It is both necessary and possible for grid operators to predict the behavior of their customers. Necessity: Customer side Possibility: Data analytics side Customers role is changing • From passive loads to ‘prosumers’ • Generating energy and supplying back to the grid • Organizing themselves in cooperatives • Adopting possibly disruptive technologies (PV/EV/heat pumps) • More vocal: e.g. social media More data • More and more data about our customers and our assets (digitalization) • More and more external data can be acquired at decreasing cost • Better IT systems to store, link and prepare data for analysis We want to influence customer behavior more • Energy savings programs • Support sustainable energy • Demand/response and peakshaving • Smart meter roll out Better analytics • Better analytics methods (data mining algorithms) • From looking back and describing to predicting and optimizing based on that prediction • Better tools and IT to handle large data sets It is both necessary and possible for grid operators to predict the behavior of their customers and adapt their strategy and their operations to these predictions. 4 Example Applications of CI Predicting customer behavior has numerous applications which can deliver serious business value for grid operators. Focus of Alliander SERI CI team the last 2 years Energy Saving Potential Predict energy savings potential of (groups of) customers Use for savings project location selection or providing individual benchmarks Fraud detection Predict based on usage data and customer data which customers/ areas are high risk for hosting illegal weed growing operations Adoption of PV/EV/heatpumps Predict which customers are most likely to adopt EV/PV/heatpumps Model spread of new technologies over service area to prepare the grid Future focus of Alliander SERI CI team Customer Satisfaction Predict which customers/group have potential/risk for increasing/ decreasing customer satisfaction Determine which variables are most predictive of customer satisfaction Demand Response Predict which customers will respond to demand/response programs Predict the shift in demand achieved through e.g. variable rates Customer Contact Predict which (group of) customers is likely to contact us, when and why Pre-empt or optimize the contact for costs or customer satisfaction PV-model Predicting PV-adoption allows Alliander to support the energy transition and prepare its assets for the additional load. • Popularity of solar panels (PV) has increased dramatically over the last couple of years. Situation Complication Solution Use Case • This growth is predicted to continue by a factor 4 to 16 in 2020. • In order to stimulate the energy transition as effectively as possible, we need to know where the highest potential for PV is. • The adoption of PV causes a very local significant extra load on the grid with possible disruptions and outages as a result. • Building a predictive model based on customer data to predict which customers are most likely to adopt PV and predict the spread of PV over the Liander grid over time. • Use the model to predict the PV distribution in the province of Flevoland up to 2030 to assess the impact on the grid and identify potential problems. 6 PV-model To predict the location of future PV installation we have developed a distribution model. Model development is aimed at growth prediction. PV probability input model output installed PV logistic regression PV probability per household PV distribution household demographics MC sampling expected PV distribution estimate of PV growth PV predicted growth – how much and where installed PV subsidy economic prospects survival analysis household demographics growth curves PV installation PV price development Datum Titel van de presentatie 7 Flevoland PV penetration Use case: Predict the PV distribution in Flevoland to assess the impact on the grid and identify potential problems for the grid. household with specific characteristics … PV probability … PV distribution PV scenarios household distribution 8 Flevoland PV penetration Use case: Predict the PV distribution in Flevoland to assess the impact on the grid and identify potential problems for the grid. risk map 9 Questions?