MODEL FOR PROCESSING HISTORICAL TIMESERIES AND ESTABLISHING RULESET FOR ANOMALY DETECTION IN CURRENT SENSOR DATA AND GENERAL-PURPOSE FORECASTING FOR SMART FARMING IN LATVIA

In this article the model is created for establishing ruleset for further anomaly detection in data from smart farming sensors which could be utilized in data cleaning, data analysis and general-purpose forecasting. Model is based on assumption that processes related to smart farming are tied to nature in particular area, for example, Latvia. In case we have historical data for some metric, for example, air temperature, then we can build such a model as long-term knowledge of phenomena provides the basement for such a model to be successful, for example, we know that in Latvia there are four repeating seasons spring, summer, autumn, and winter, but there are even shorter cycles as twelve months with certain known characteristics for each month and day night cycles. Air temperature is more affected by day and night cycles during summer season, less in other seasons. Created model will show calculations based on air temperature, but similar approach could be utilized for other metrics which depend on location and seasonality of nature, such as soil moisture level, sunlight, etc. In case, approach is applied elsewhere it is assumed that rulesets are regenerated based on local historical timeseries data.