System and method for generating battery alarms in infusion devices

A computer-implemented method for training a neural network to predict a remaining battery capacity state, the neural network including at least an input layer and an output layer, the method comprising:

• (i) receiving, from one or more rechargeable batteries over time, a plurality of measured battery parameters including at least voltages of the one or more rechargeable batteries, currents of the one or more rechargeable batteries, temperatures of the one or more rechargeable batteries, remaining voltages of the one or more rechargeable batteries, and full charge voltages of the one or more rechargeable batteries;
• (ii) preprocessing the measured battery parameters to generate input feature vectors, the preprocessing including determining rates of changes of the voltages over time using a time stamp associated with the measured battery parameters;
• (iii) generating a training data set comprising:
• the input feature vectors derived from the measured battery parameters, and corresponding target output vectors that indicate at least a low battery state, a very low battery state, and a depleted battery state;
• (iv) initializing weights and biases of a neural network such that output nodes of the output layer of the neural network correspond to the low battery state, the very low battery state, and the depleted battery state;
• (v) training the neural network by minimizing a loss function;
• and
• (vi) storing the trained neural network for use in generating battery alarm indications corresponding to the low battery state, the very low battery state, and the depleted battery state.