Probabilistic and Reinforcement Learning Track
2023 International Planning Competition
2023 International Planning Competition
Power Unit Commitment
| Example name | PowerGeneration |
| Action space | Dict |
| State space | Dict |
A number of power producers cooperate to meet daily demand that fluctuates according to the maximum temperature on a given day. A cost is incurred for every unit of power produced and income is received for every unit consumed by the demand. There is a large penalty for failing to meet demand on a given day and there are per-power plant penalties for deviating from the previous day’s production at each plant – some plants must pay higher operating costs for changes in production. Power generation is in integer units, consumption is real, and time steps are assumed to span 24 hours.
| Constant | Type | Desc |
|---|---|---|
| PROD_UNITS_MIN(plant) | int | Minimum unit to produce |
| PROD_UNITS_MAX(plant) | int | Maximum unit to produce |
| PROD_CHANGE_PENALTY(plant) | float32 | Penaly for changing the production amount |
| COST_PER_UNIT(plant) | float32 | Cost per power unit |
| INCOME_PER_UNIT | float32 | Income per power unit |
| TEMP_VARIANCE | float32 | Temperature change variance |
| DEMAND_EXP_COEF | float32 | Exp coefficient for the demand U shape |
| MIN_DEMAND_TEMP | float32 | Center of the demand U shape |
| MIN_CONSUMPTION | float32 | DC level of the demand U |
| UNFULFILLED_DEMAND_PENALTY | float32 | Penalty for producing too much power |
All of these can be read from the RDDLEnv interface and from the RDDL files.
The actions are the current amount of power each plant is required to produce.
| Action | Type | Desc |
|---|---|---|
| curProd(plant) | Discrete(PROD_UNITS_MIN, PROD_UNITS_MAX) | current production command to each plant |
The state space represents the temerature and the previous production of each of the plants in the problem.
| State | Type | Desc |
|---|---|---|
| temperature | Box(1, -inf, inf, float32) | Current temperature |
| prevProd(plant) | Discrete(-inf, inf) | previous power produced per plant |
The reward function is defined as
cost of supply per plant, demand income, demand exceeds supply penalty, steady-state penalties