As a user I want to create a forecast for an attribute based on the attribute's past data using the weighted exponential average algorithm; the forecast data should then be retrievable from the existing forecast HTTP API.

1. Add the configuration item forecast to the attribute to generate a forecast; the value type should be a JSON schema that allows all relevant parameters to be configured for weighted exponential average calculation (see below)
2. Backend reacts to configuration item CRUD and starts/stops a scheduled task to periodically calculate forecast data and to store it using the AssetPredictedDatapointService

Forecast configuration item

• type - The algorithm to use for now only wea supported (this can be used as @JsonTypeName for deserialisation)
• pastPeriod - The period of time between each historical value to look back; should be a valid ISO8601 duration (called P in the formula)
• pastCount The number of periods to look back; positive integer (called R in the formula, >=1)
• forecastPeriod - The period of time between each forecast value to calculate; should be a valid ISO8601 duration
• forecastCount - The number of forecast periods to calculate; positive integer

For example, the following defines an hourly forecast for 24h based on the data from the same day a week earlier using the past 4 weeks:

{
  "type": "wea",
  "pastPeriod": "P7D",
  "pastCount": 4,
  "forecastPeriod": "PT1H",
  "forecastCount": 24
}

Weighted Exponential Average Calculations

a = 2/(R+1)
Attribute_t = Attribute_t-p * a + (1-a) * (Attribute_t-pN (1-a) * (Attribute_t-3P))
for R = 1: Attribute_t = Attribute_t-P * a
for R = 2: Attribute_t = Attribute_t-P * a + (1-a) * Attribute_t-2P
for R = 3: Attribute_t = Attribute_t-P * a + (1-a) * (Attribute_t-2P * a + (1-a) * Attribute_t-3P)
for R = 4: Attribute_t = Attribute_t-P * a + (1-a) * (Attribute_t-2P * a + (1-a) * (Attribute_t-3P * a + (1-a) * Attribute_t-4P))
etc...

Notes

• Forecast data points should have consistent timestamps so each iteration replaces the previous value(s)
• Forecast service needs to find all 'Forecast' configuration items at startup and also monitor AssetEvents (see AgentService monitoring of AgentLink configuration items as an example)
• Use a single scheduled task for generating all forecast data in series (to minimise load on the system)

Tasks

• Create abstract ForecastConfiguration class with type field and create concrete sub type ForecastConfigurationWeightedExponentialAverage
• Add FORECAST meta item descriptor to MetaItemType with type of ForecastConfiguration
• Create ForecastService (see AgentService for guidance on initialising, getting FORECAST attributes and reacting to asset changes and see ForecastWindService for guidance on storing predicted data)