Consumption measurements
Selected smart-meter consumption feeds were collected through Kafka at 15-minute granularity. Other collection paths supplied historical or less timely readings.
Energy forecasting · Utility case study
Energy forecasts.
€3M annual savings.
CloudHero put energy forecasting workflows into use for a utility company. The project owner reports 94%+ forecast accuracy and €3 million in annual energy cost savings, supported by consumption integration, forecast delivery and evaluation views.
Project outcomes
Forecast accuracy and annual savings.
- Reported forecast accuracy
- 94%+
- Reported annual energy cost savings
- €3M
Project outcomes supplied by CloudHero’s project owner. The accuracy calculation, evaluation period and savings baseline are not specified here. These figures describe this engagement and are not a forecast of results for other customers.
Evidence & scope
The scope behind the energy workflow.
- Consumption points referenced in the handover
- 400+
- Selected consumption-feed intervals
- 15 minutes
- Consumption forecast horizon
- 1 day ahead
The handover references approximately 403 consumption points; the figure is rounded down and is a historical scope observation, not a claim that every point had complete smart-meter data or an individual forecast. Fifteen minutes is selected source granularity, not end-to-end delivery latency. Day ahead describes the forecast use case, not an accuracy result.
The challenge
Bring the available data into the forecasting workflow.
Energy planning needs a view of expected consumption and generation. Different sources provide information at different times and levels of detail, so the workflow has to account for what is actually available.
The engineering interviews describe historical collection paths alongside selected smart-meter feeds. Data availability was a practical constraint on the forecasts and the way teams could evaluate them.
The workflow
From measurements to forecasts teams can inspect.
- 01Energy inputs
Consumption measurements and generation information.
- 02Prepared datasets
Source-aligned data in the BigQuery analytical environment.
- 03Forecast execution
Python library and notebook workflows put into production.
- 04Planning & evaluation
Forecast views, energy reporting and comparisons with recorded consumption.
Source integration
Work with the readings available.
Generation and energy position
Reporting brought consumption and generation information together. Selected industrial-data integrations used AWS services before transferring information into the Google Cloud analytical environment.
Putting forecasts into use
Deliver outputs and make comparison possible.
Forecasting development included Python libraries and notebooks. CloudHero’s engineering work put those workflows into production and made their outputs available through the data platform and dashboards.
Day-ahead consumption forecasts supported review of expected energy requirements. Separate evaluation views compared forecasts with recorded consumption, allowing the team to examine deviations over time.
The distinction between a forecast, an actual measurement and a delayed observation remained explicit. The project owner reports 94%+ forecast accuracy and €3 million in annual energy cost savings for this engagement. These outcomes are distinct from the historical consumption-point count, source intervals and forecast horizon shown above.
Delivered capability
Business results and reviewable forecasts.
Usable forecast outputs
Model workflows produce outputs that teams can review through dedicated views, rather than relying only on development notebooks.
Forecast-versus-actual evaluation
Separate dashboard views compare predicted consumption with recorded consumption over time.
Energy-position reporting
Consumption and generation information can be reviewed together to support energy planning discussions.
Related utility work
Explore the enterprise data platform and reporting story.
A separate case study covers enterprise data strategy, the BigQuery-centered platform and excavation reporting. Its sample work counts and reporting features are kept separate from the energy evidence.
Put your energy data and forecasts to work.
Discuss consumption sources, forecast delivery and how your team will review the results.