Anomaly detection in oil retail business

By Naveen Sikka, Principal Industry Consultant (Oil & Gas), SAS

The downstream segment consists of companies engaged in refining, processing and marketing of petroleum products. Retail outlets have installed retail automation systems for ensuring quality and quantity, customer confidence building, increasing the speed of transactions and operations and enabling direct interaction with the customers. This also enables oil marketing companies to monitor/analyse product stock/sales of retail outlets and control retail fuel prices at the outlets effectively and efficiently.

The automation system provided at retail outlets is generally connected to a central server. All the business data/transactions at the retail outlet are collected in the central server and used for business analysis. The automation vendors have integrated forecourt controller (FCCs) to transfer all transactional data from their respective FCC to the central server. FCCs are also capable of accepting and updating data pushed from the central servers. Typically, all dispenser sales data, pricing data, and inventory data at different time stamps get collected for different retail outlets at central servers. Further, the automation systems do have the capability to monitor flow rates from each dispenser. There are various other equipment installed at the retail outlet which help to monitor the fuel dispensed from each of the dispensers installed within the retail outlets.

In addition to the secondary sales retail data, the oil marketing companies also have the complete primary sales marketing data from the terminal for all the dispatches that take place to replenish the finished products such as motor spirit (MS), and diesel at various retail outlets. The primary sales data along with secondary sales data can help the oil marketing in checking the adulteration of MS and high-speed diesel with lower cost organic products (solvents such as mineral spirits, kerosene, rubber solvents, naphtha, and thinner) which is a widespread malpractice worldwide.

It is estimated that approximately eight per cent of transport fuel sold in India and worldwide is adulterated with hydrocarbons like pentane, kerosene, xylene and toluene. Through its work, SAS has seen that the following issues observed within the oil retail business can be resolved and curbed:

  • Forecourt controller tank gauge issue that reports erroneous data to central server
  • Primary sales transaction data issue
  • Revenue losses due to the procurement of additional material from multiple sources
  • Reporting of higher sales through various dispensers
  • Remain-in material during the unloading process
  • TT turn-around efficiencies

Since obtaining retail dispensation, forecourt tank and terminal transaction data is difficult due to privacy concerns, a data generator was used for creating large realistic smart retail datasets from a small seed of real data. The generator includes several user-controlled parameters. For instance, new datasets were created corresponding to consumers who are less or more “peaky” than those in the original sample. The real dataset obtained consists of over six months of data from around 50 retail outlets and analysed the following sets of data:

  • Reconciliation of wet stock deliveries to site from depots with sales and inventory on a periodic basis
  • Analysis of nozzle sales
  • Reconciliation between system accumulated and dispenser accumulated totals
  • Consolidation of grade totals
  • Tank gauge configuration
  • Tank delivery measurement & inventory reporting

The proposed benchmark was implemented using a state-of-the-art SAS visual investigator platform to detect anomalies in either fuel dispensation or adulteration of fuel. The models provided the following:

  • Advanced analytics techniques and machine learning algorithms to monitor the fuel dispensation from each dispenser based on current and historical data.
  • Generating a consolidated risk score for each dispenser at different time-periods based on data analytics principles by analysing parameter trends of each dispenser for different retail fuel stations.
  • Use analytical and decision-making tools combined with real time insights for a holistic understanding of fuel retail dispensation system parameter anomalies for early detection of frauds and anomalies and provide guidelines (prescriptive analysis) to the marketing team for corrective action.
  • Generate quick insights from the real-time streaming data, update key performance indicators and display results through dashboards, or other mechanisms for early detection of frauds and anomalies and provide discrepancy analysis.

The above analytical procedures have helped the oil marketing company to increase its consumer confidence and protect the goodwill of the companies responsible for selling fuel from the illegal activities of rogue dealers. Instead of failing a test which indicates that a particular fuel dispenser is normal, an alarm was generated when the given fuel dispenser passes a test that indicates fraud. Further, a failure to report data to a central server may also be indicative of fraud. In such an instance, an alarm should be generated, and the station operator must interrogate why the data was not provided as required. Alternatively, an independent, manual test could be performed at the station to confirm that fraudulent activity is taking place before any questions are asked.