Stochasticity mitigation in deployed AI agents

A computing device comprising:

• one or more processors;
• a memory in communication with the one or more processors, the memory having computer-readable instructions stored thereupon which, when executed by the one or more processors, cause the computing device to:
• receive an agent that maps a current state of a mechanical system to a delta value, wherein the delta value represents a difference between a fixed state of the mechanical system and a fluctuating state of the mechanical system;
• generate an action in part by performing an undelta operation on the delta value;
• and
• apply the modified action to control an aspect of the mechanical system.

A non-transitory computer-readable storage medium comprising instructions that, when executed by a processor, cause the processor to:

• train a machine learning model based on states collected from a chiller system, wherein the machine learning model is trained to generate delta values that represent differences between chilled water setpoint values and outside air temperature values;
• use the machine learning model to infer a plurality of delta values of the chiller system from a plurality of sets of states;
• generate an agent based on the plurality of sets of states and the plurality of inferred delta values;
• and
• wherein the agent generates a chiller water setpoint action from one of the plurality of delta values and applies the chiller water setpoint action to configure the cooler system.