Data-enabled predictive control (DeePC) leverages system measurements in characterizing system dynamics for optimal control. The performance of DeePC relies on optimizing its hyperparameters, especially in noisy systems where the optimal hyperparameters adapt over time. Existing hyperparameter tuning approaches for DeePC are more than often computationally inefficient or overly conservative. This paper proposes an adaptive DeePC where we guide its hyperparameters adaption through reinforcement learning. We start with establishing the relationship between the system I/O behavior and DeePC hyperparameters. Then we formulate the hyperparameter tuning as a sequential decision-making problem, and we address the decision-making through reinforcement learning. We implement offline training to gain a reinforcement learning model, and we integrate the trained model with DeePC to adjust its hyperparameters adaptively in real time. We conduct numerical simulations with diverse noisy conditions, and the results demonstrate the identification of near-optimal hyperparameters and the robustness of the proposed approach against noises in the control.

The integration of solar power has been increasing as the green energy transition rolls out. The penetration of solar power challenges the grid stability and energy scheduling, due to its intermittent energy generation. Accurate and near real-time solar power prediction is of critical importance to tolerant and support the permeation of distributed and volatile solar power production in the energy system. In this paper, we propose a deep-learning based ultra-short-term solar power prediction with data reconstruction. We decompose the data for the prediction to facilitate extensive exploration of the spatial and temporal dependencies within the data. Particularly, we reconstruct the data into low- and high-frequency components, using ensemble empirical model decomposition with adaptive noise (CEEMDAN). We integrate meteorological data with those two components, and employ deep-learning models to capture long- and short-term dependencies towards the target prediction period. In this way, we excessively exploit the features in historical data in predicting a ultra-short-term solar power production. Furthermore, as ultra-short-term prediction is vulnerable to local optima, we modify the optimization in our deep-learning training by penalizing long prediction intervals. Numerical experiments with diverse settings demonstrate that, compared to baseline models, the proposed method achieves improved generalization in data reconstruction and higher prediction accuracy for ultra-short-term solar power production.

Probabilistic power flow quantifies voltage and phase angle uncertainty under variable photovoltaic generation, but repeated AC Monte Carlo simulation is costly. A local topology change also modifies the electrical operator and state dimension when only a small target data set is available. We propose a Basis Constrained Graph Convolutional Network (BCGCN) for few-shot adaptation after local bus additions in small-scale grids. BCGCN predicts nonlinear residuals around a first-order solution using graph Laplacian and Proper Orthogonal Decomposition modes. It transfers source coordinates; adapts only the new bus rows, rotation, readouts, and correction gate; and freezes the backbone. The experimental results indicate that BCGCN leads all four reported errors on the IEEE 14 and IEEE 57 expansions. IEEE 118 and Polish 2746 establish the scale boundary. BCGCN wins only 9 of 64 IEEE 118 error cells and none on Polish 2746, while retaining compact updates. The paired IEEE 118 PV study shows that target pilots reduce zero-shot error and residual correction removes most high variability linearization error. BCGCN is therefore effective for local few-shot adaptation in small grids but not an accuracy-preserving adapter for large networks.

This work concerns the control of unknown nonlinear systems corrupted by disturbances. For such systems, we propose an anti-disturbance dual control approach with active learning of the disturbances. Our approach holds the dual property of handling the two tasks simultaneously and iteratively: (i) learn the disturbances affecting the system and (ii) drive the system output towards a reference trajectory. Particularly, we model nonlinear system dynamics using a specialized neural network (SNN). This SNN formulates the disturbances via the designed additive and multiplicative disturbance components. We consider both additive and multiplicative disturbances for precise description and recognition of disturbance profile. We achieve the disturbance recognition in the SNN via the design of a Bayesian-based active learning approach, which allows the disturbance learning to be decoupled from the control law derivation. Such a decoupling contributes to the control robustness in the existence of varying and abrupt disturbances. We derive the dual control law based on the active learning of the SNN, and validate our approach via one-time and Monte Carlo simulations. The results demonstrate a fast disturbance recognition by our method in real-time and the robustness of control of unknown systems with abrupt disturbances. We evaluate our approach on the speed control of high-speed train, and the results manifest efficient control of the train speed with disturbance resilience, without prior knowledge about the train dynamics and the disturbances imposed on the train. We openly released the code for this work for reproduction purpose.