Forecasting
Forecasts and early warnings from your sensor and demand data, worked out on the device itself. It works offline, with no cloud in the loop.
Overview
The Agentyk Forecasting Engine runs on the device that collects the data, from a gateway down to a small sensor board. It looks at a signal's recent history and forecasts where it is heading, with a likely range around the forecast. There is no model to train or pick for each signal.
Agentyk Forecasting Engine
On the deviceSees trouble coming in your data, right where it is measured.
Temperature, vibration, pressure and energy use often drift for a while before something goes wrong, and by the time a fixed alarm level is crossed there is little time left to plan. Sending every reading to the cloud to keep an eye on it costs bandwidth, and does not work where there is no connection.
The Agentyk Forecasting Engine forecasts each signal ahead on the device, with a likely range that shows how certain the forecast is. From that range it can warn you early, for example when a reading is likely to reach its alarm level within the next hours, and say in plain words which measurements moved. It also follows series with daily, weekly or seasonal patterns, such as energy use or demand. One pretrained model covers your signals, so nobody has to train or tune a model for each one.
- Forecasts with a likely range, made on the device
- Early warnings before a reading reaches its alarm level
- A plain-language reason with every warning
- Seasonal and demand series as well as sensor readings
- Works offline: no cloud and no connection needed
- A small, self-contained engine with no dependencies
- Licensed per device and checked on the device itself, with an evaluation mode to try it first