Modifying operation of physical emission sources using a modified gradient descent model

A system comprising:

• one or more memory devices;
• and
• one or more processors configured to cause the system to:
• determine a plurality of constraints comprising one or more minimum unit numbers or one or more maximum unit numbers associated with one or more of a plurality of physical emissions sources;
• determine a plurality of emissions values for the plurality of physical emissions sources by:
• determining, utilizing a neural network, a feature representation for an entity corresponding to the plurality of physical emissions sources according to attributes associated with the entity:
• determining, utilizing the neural network, one or more additional feature representations for one or more additional entities according to additional attributes associated with the one or more additional entities;
• and
• determining the plurality of emissions values for the entity based on the one or more additional entities in response to distances between the feature representation and the one or more additional feature representations being within a threshold distance within a feature space;
• generate an action recommendation for modifying the plurality of physical emissions sources by utilizing a modified gradient descent model to iteratively adjust the plurality of emissions values for the plurality of physical emissions sources by determining a first set of adjusted emissions values for the plurality of physical emissions sources with according to the plurality of constraints and one or more target emissions values for the plurality of physical emissions sources and a second set of adjusted emissions values for the plurality of physical emissions sources without the plurality of constraints;
• and
• cause, according to the action recommendation, a source modification device to modify an operating temperature, an operating speed, an operating power, or an operating time of at least one physical emissions source of the plurality of physical emissions sources.

A computer-implemented method comprising:

• instructing a computing system to generate an action recommendation for modifying a plurality of physical emissions sources, wherein the computing system generates the action recommendation by:
• determining a plurality of constraints comprising one or more minimum unit numbers or one or more maximum unit numbers associated with one or more of the plurality of physical emissions sources;
• determining a plurality of emissions values for the plurality of physical emissions sources by:
• determining, utilizing a neural network, a feature representation for an entity corresponding to the plurality of physical emissions sources according to attributes associated with the entity;
• determining, utilizing the neural network, one or more additional feature representations for one or more additional entities according to additional attributes associated with the one or more additional entities;
• and
• determining the plurality of emissions values for the entity based on the one or more additional entities in response to distances between the feature representation and the one or more additional feature representations being within a threshold distance within a feature space;
• generating the action recommendation by utilizing a modified gradient descent model to iteratively adjust the plurality of emissions values for the plurality of physical emissions sources by determining a first set of adjusted emissions values for the plurality of physical emissions sources according to the plurality of constraints and one or more target emissions values for the plurality of physical emissions sources and a second set of adjusted emissions values for the plurality of physical emissions sources without the plurality of constraints;
• and
• modifying, according to the action recommendation, an operating temperature, an operating speed, an operating power, or an operating time of at least one physical emissions source of the plurality of physical emissions sources.