Key
3379
Conference Year
2014
Keywords
Central Plant Optimization, Model Predictive Control, Cost Optimization, Thermal Energy Storage
Abstract
An optimization framework is used in order to determine how to distribute both hot and cold water loads across a central energy plant including heat pump chillers, conventional chillers, water heaters, and hot and cold water (thermal energy) storage. The objective of the optimization framework is to minimize cost in response to both real-time energy prices and demand charges. The linear programming framework used allows for the optimal solution to be found in real-time. Real-time optimization lead to two separate applications: A planning tool and a real-time optimization tool. In the planning tool the optimization is performed repeatedly with a sliding horizon accepting a subset of the optimized distribution trajectory horizon as each subsequent optimization problem is solved. This is the same strategy as model predictive control except that in the design and planning tool the optimization is working on a given set of loads, weather (e.g. TMY data), and real-time pricing data and does not need to predict these values. By choosing the varying lengths of the horizon (2 to 10 days) and size of the accepted subset (1 to 24 hours), the design and planning tool can be used to find the design year’s optimal distribution trajectory in less than 5 minutes for interactive plant design, or the design and planning tool can perform a high fidelity run in a few hours. The fast solution times also allow for the optimization framework to be used in real-time to optimize the load distribution of an operational central plant using a desktop computer or microcontroller in an onsite Enterprise controller. In the real-time optimization tool Model Predictive Control is used; estimation, prediction, and optimization are performed to find the optimal distribution of loads for duration of the horizon in the presence of disturbances. The first distribution trajectory in the horizon is then applied to the central energy plant and the estimation, prediction, and optimization is repeated in 15 minutes using new plant telemetry and forecasts. Prediction is performed using a deterministic plus stochastic model where the deterministic portion of the model is a simplified system representing the load of all buildings connected to the central energy plant and the stochastic model is used to respond to disturbances in the load. The deterministic system uses forecasted weather, time of day, and day type in order to determine a predicted load. The estimator uses past data to determine the current state of the stochastic model; the current state is then projected forward and added to the deterministic system’s projection. In simulation, the system has demonstrated more than 10% savings over other schedule based control trajectories even when the subplants are assumed to be running optimally in both cases (i.e., optimal chiller staging, etc.). For large plants this can mean savings of more than US $1 million per year.
Model Predictive Control for Central Plant Optimization with Thermal Energy Storage