Forecasting

Forecasting aims to help understand what future consumption may look like. It is available for different resolutions and units and is generated by analyzing each locations historic consumption, adding data from external factors and providing insights into what energy consumption may be by the end of a month, year or similar.

How does it work?

Our forecasting feature allows users to see how much energy they are projected to consume in the future. Eliq offers energy forecasting for several fuels Consumption data, on different timespans and units. There are many different use cases for energy forecasting, for example, helping the end users understand what the total amount of the energy bill will be to reduce bill-shock and customer dissatisfaction as well as helping the user action these insights before it is too late.

Forecast Insights are available for any Location in the Eliq platform.

API

The Insights API Endpoint Get Locations/{locationId}/forecast is used to request Daily or Monthly Consumption Forecast data.

The insights API Endpoint Get Locations/{locationId}/forecast/hourly is used to request Hourly Consumption Forecast data.

Daily Forecasting

This features provides a daily forecast for the coming days (maximum 31 days). This helps the end users to understand how much energy they will use in the current month.

Algorithm

The algorithm is based on version of supervised learning. For each user, we run an algorithm that adopts how much each forecasting parameter (such as previous patterns, weekend, temperature etc) should influence the forecast. Millions of different combinations are tested and evaluated against the last 8 weeks of data. The combination that has proven most successful is used to run the forecast.

Required data

Minimum 7 days of energy data, works best on 365 days of data.

Monthly Forecasting

This features provides a monthly forecast for the coming year (next 12 months). This helps the end users to understand how the energy usage is predicted to change during the year.

Algorithm

The algorithm analyses seasonal changes in the previous energy usage and extrapolates this forward. Unusual high or low monthly consumption is taken care of and filtered out. The last few months have a higher impact on the end result to faster adjust for bigger changes like buying an EV. In general, the forecast for the coming year will be similar to previous years consumption.

Required data

Minimum 1 month of energy data. (Recommended is minimum 6 months to get a reliable yearly forecast). Preferred is 12+ months.

Hourly forecast

This features provides an estimated hourly forecast for the next 24 hours for a Location.

The response is slightly different compared to daily or monthly forecast as it returns 3 values per hour; estimate , lower_bound and upper_bound . These 3 values aim to provide 3 pieces of information to the end user; Eliqs estimate for energy consumption for the hour as well as the upper and lower boundary of the estimate. This can be used to indicate to the end user an estimate span of lowest to highest forecast for each hour.

The purpose of providing 3 values is to support a user experience to help the end user trust the forecasted numbers. Forecasting consumption on an hourly basis may result in a widely differing outcome compared to the forecast due to sudden usage of energy intense appliances. By providing an hourly forecast with an upper and lower bound, a UI can indicate the best guess along with highest and lowest estimates for the same hour.

Algorithm

The hourly forecast algorithm analyses weekly and daily seasonality and extrapolates them into the future on an hourly level. Up to one month of historical data is used in the forecast, because recent data is the best indicator of short-term consumption patterns. The algorithm also calculates the 95% confidence interval for every estimated hour, which can be a good indicator of how confident the algorithm is about that exact estimate.

Required data

The location must have a minimum of 1 week of data on an hourly resolution or higher. The recommended amount of data, when the hourly forecast works best is ~1 month (28-31 days of hourly data).

Use Cases

Forecast summary

A forecast summary can be presented on home screen or card displaying the total forecasted cost and energy in the end of the current month. It can show a combination of Cost and Energy (left card) but also present in Energy only (right card) for example if cost is not available for the current Location.

Forecast in graphs

Forecast can be displayed in graphs, displayed with a lighter green color to emphasize it is estimated values.

Combining forecast and historic consumption data

Forecast data can be combined with historic consumption data in graph for periods that are not yet completed. Such as Month of a Year.

Forecast in monthly comparison card

In this Use Case, a Monthly Comparison is being shown for the ongoing month of April, combining historic consumption that is available for some days in April with a forecast for the remaining days of April.

‍

‍