Distributed TCL Control Using Load Aggregation and Power Shifting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Controlling a heterogeneous population of Thermostatically Controlled Loads (TCLs) in power grids is challenging due to their unique and complex dynamics, large state action space, and the need to optimize multiple competing system-level objectives simultaneously, which existing centralized control methods struggle to address effectively.
Innovation Solution
A distributed control system comprising a central Load Aggregator Controller and building-level controllers, utilizing reinforcement learning and model predictive control, with an evolutionary strategy for policy updates, and modeling each TCL as a virtual battery to simplify the control problem and optimize grid-level objectives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If centralized control methods are used to manage TCLs, then system-level objectives can be optimized, but the complexity of controlling heterogeneous TCLs with unique dynamics increases significantly
Solution Approach 1:
The control system is segmented into a central Load Aggregator Controller that manages system-level objectives and building-level controllers that handle individual building TCLs. This segmentation allows the central controller to optimize aggregate demand without directly managing the complexity of each heterogeneous TCL, while building-level controllers handle local control tasks independently.
Solution Approach 2:
The Load Aggregator Controller acts as an intermediary between the utility company and individual building controllers. It aggregates predicted power demand from multiple buildings, applies convolutional filters to generate target load profiles, and distributes apportioned power shifts to building-level controllers, thereby mediating the control complexity.
2Adaptability or versatility
If individual building controllers manage their own TCLs independently, then local control flexibility is maintained, but aggregate power demand optimization across multiple buildings becomes difficult
Solution Approach 1:
The system merges individual building power demand predictions into an aggregate power demand profile at the Load Aggregator Controller. By combining data from multiple buildings and applying convolutional filters, the system optimizes aggregate demand while allowing each building to maintain its own control policies and flexibility through building-level controllers.
Solution Approach 2:
The system implements feedback loops where building-level controllers report predicted power demand to the Load Aggregator Controller, which then distributes apportioned power shifts back to individual buildings. This feedback mechanism enables coordinated optimization while preserving local control autonomy.
3Measurement precision
If the control system processes detailed individual TCL dynamics, then accurate control is achieved, but the computational burden and data processing requirements increase
Solution Approach 1:
The system uses predicted power demand data as a simplified representation (copy) of complex TCL dynamics rather than processing full detailed models. Building-level controllers generate predicted power demand profiles that capture essential behavior without requiring complete dynamic models, reducing computational burden while maintaining sufficient control accuracy.
Solution Approach 2:
Building-level controllers perform preliminary processing by generating predicted power demand profiles before sending data to the Load Aggregator Controller. This preliminary action filters and summarizes detailed TCL dynamics into manageable predictions, reducing the computational burden on the central system while preserving essential control information.
Data Source
AI summary
A computer implemented method for controlling a load aggregator for a grid includes receiving a predicted power demand over a horizon of time steps associated with one of at least two buildings, aggregating the predicted power demand at each time step to obtain an aggregate power demand, applying a learnable convolutional filter on the aggregate power demand to obtain a target load, computing a difference between the predicted power demand of the one building with the target load to obtain a power shift associated with the one building over the horizon of time steps, apportioning the power shift according to a learnable weighted vector to obtain an apportioned power shift, optimizing the learnable weighted vector and the learnable convolutional filter via an evolutionary strategy based update to obtain an optimized apportioned power shift, and transmitting the optimized apportioned power shift to a building level controller associated with the one building.