FORECASTING THE WATER CONSUMPTION FOR A PUBLIC UTILITY COMPANY
Together with a municipal utility, we have developed a forecasting solution that enables the precise forecast of water requirements for an entire city on an hourly and daily basis.
As a municipal utility, the customer is responsible, among other things, for the entire water supply of a city in Germany. Exact predictions of the water demand are necessary for maintenance measures and water redistribution in e.g. burst pipes indispensable. For example, water redistribution measures in the event of a burst pipe require a precise assessment of the water demand at any time of the day.
Although the requirements follow a strong daily and weekly seasonality (given by working days and requirements in the industry), the influence of weather and other factors should not be neglected. At the start of the project, the customer creates the forecasts manually and only checks the influence of external factors on a random basis. In addition to the challenges in forecasting, there were also difficulties with the quality of the data.
Historically, measured values were collected manually, which led, for example, to missing values in the case of illnesses of individual employees. The aim of the project was therefore to predict the hourly water demand for the next seven days and the daily water demand for the next 30 days as precisely as possible.
In addition to historical data on water consumption over the past five years, various external influencing factors such as: B. Weather data (temperature, precipitation), school and semester breaks, other calendar information such as season and public holidays as well as special events (e.g. football games). In particular, the weather data was used to derive further variables: For example, the days since the last rainfall, the amount of rainfall in the last week or the days until the next rainfall can be used.
Based on the water demand history and other external factors, machine learning forecast models were trained to calculate future water demand. Of the influencing factors evaluated, the weather data and calendar information ultimately had the biggest influence on the forecasts.
With the help of the machine learning models, the hourly and daily water requirements could be predicted with a very high degree of accuracy. With an accuracy of 95.0% for the hourly forecasts and an accuracy of 94.6% for the daily forecasts. The predictions were therefore 32% and 43% more accurate than an individually constructed benchmark that is not based on machine learning algorithms. Due to the more precise demand forecasts, the municipal utility benefits from increased planning security for e.g. maintenance measures or water redistributions.