USING TIME SERIES AND LSTM NEURAL NETWORKS FOR IDENTIFICATION OF HIVE STATES

Bees are extremely important for ecosystems and agriculture, yet their populations are declining due to various threats. This creates a critical need for effective systems to monitor honeybee colony health, allowing beekeepers to react promptly. Traditional methods have limitations, leading to growing interest in artificial intelligence applications. This work aims to develop models based on the LSTM architecture to identify different beehive states by analyzing time-series data from sensors (weight, temperature). Data from Kaggle was used, undergoing thorough preparation including aggregation, interpolation, derivative feature calculation, and segmentation. Two LSTM models were trained – one for point states (swarming, feeding, honey harvesting) and another for global states (normal, colony collapse, spring build-up, honey flow). They demonstrated high accuracy (96% and 92% respectively), confirming the approach's effectiveness.