This paper describes a data-driven approach for real-time control of a physical system. Specifically, this paper focuses on the cooperative wind farm control where the objective is to maximize the ...total wind farm power production by using control actions as an input and measured power as an output. For real time, data-driven wind farm control, it is imperative that the optimization algorithm is able to improve a target wind farm power production by executing as small number of trial actions as possible using the wind farm power monitoring data. To achieve this goal, we develop a Bayesian ascent (BA) algorithm by incorporating into the Bayesian optimization framework a strategy that regulates the search domain, as used in the trust region method. The BA algorithm is composed of two iterative phases, namely, learning and optimization phases. In the learning phase, the BA algorithm approximates the target function using Gaussian process regression to fit the measured input and output of the target system. In the optimization phase, the BA algorithm determines the next sampling point to learn more about the target function (exploration) as well as to improve the target value (exploitation). Specifically, the sampling strategy is designed to ensure that the input is selected within a trust region to improve the target value monotonically by gradually changing the input for a target system. The results from simulation studies using an analytical wind farm power function and experimental studies using scaled wind turbines show that the BA algorithm can achieve an almost monotonic increase in the target value.
We propose a physics-inspired data-driven model that can estimate the power outputs of all wind turbines in any layout under any wind conditions. The proposed model comprises two parts: (1) ...representing a wind farm configuration with the current wind conditions as a graph, and (2) processing the graph input and estimating power outputs of all the wind turbines using a physics-induced graph neural network (PGNN). By utilizing the form of an engineering wake interaction model as a basis function, PGNN effectively imposes physics-induced bias for modelling the interaction among wind turbines into the network structure. simulation study shows that the combination of a graph representation of a wind farm and PGNN produce not only accurate and generalizable estimations but also physically explainable estimations. That is, the computing and reasoning procedures of PGNN can be understood by analyzing the intermediate features of the model. We also conduct a layout optimization experiment to show the effectiveness of PGNN as a differentiable surrogate model for wind farm power estimations.
•Propose a graph representation of a wind farm with considering wind conditions.•Propose a physics-inspired data-driven model for the wind farm power estimation task.•Combination of the wind farm graph and the model produce accurate power estimations.•The model serves as a differentiable surrogate model for wind farm power estimation.
•The continuous wake model describes well the wake profile behind a wind turbine.•The expected wind farm power is expressed as a differentiable function.•SCP can be employed to efficiently optimize ...the layout of a large-scale wind farm.•The optimized wind farm layout increases the wind farm power efficiency by 7.3%.
This paper describes an efficient method for optimizing the placement of wind turbines to maximize the expected wind farm power. In a wind farm, the energy production of the downstream wind turbines decreases due to reduced wind speed and increased level of turbulence caused by the wakes formed by the upstream wind turbines. As a result, the wake interference among wind turbines lower the overall power efficiency of the wind farm. To improve the overall efficiency of a wind farm, researchers have studied the wind farm layout optimization problem to find the placement locations of wind turbines that maximize the expected wind farm power. Most studies on wind farm layout optimization employ heuristic search-based optimization algorithms. In spite of their simplicity, optimization algorithms based on heuristic search are computationally expensive and have limitation in optimizing the locations of a large number of wind turbines since the computational time for the search tends to increase exponentially with increasing number of wind turbines. This study employs a mathematical optimization scheme to efficiently and effectively optimize the locations of a large number of wind turbines with respect to maximizing the wind farm power production. To formulate the mathematical optimization problem, we derive a continuous wake model and express the expected wind farm power as a continuous and smooth function in terms of the locations of the wind turbines. The constructed wind farm power function is then maximized using sequential convex programming (SCP) for the nonlinear mathematical problem. We show how SCP can be used to evaluate the efficiency of an existing wind farm and to optimize a wind farm layout consisting of 80 wind turbines.
We systematically develop a learning-based treatment of stochastic optimal control (SOC), relying on direct optimization of parametric control policies. We propose a derivation of adjoint sensitivity ...results for stochastic differential equations through direct application of variational calculus. Then, given an objective function for a predetermined task specifying the desiderata for the controller, we optimize their parameters via iterative gradient descent methods. In doing so, we extend the range of applicability of classical SOC techniques, often requiring strict assumptions on the functional form of system and control. We verify the performance of the proposed approach on a continuous-time, finite horizon portfolio optimization with proportional transaction costs.
Energy storage systems (ESSs) have been considered to be an effective solution to reduce the spatial and temporal imbalance between the stochastic energy generation and the demand. To effectively ...utilize an ESS, an approach of jointly sharing and operating an ESS has been proposed in a conceptual way. However, there is a lack of analytic approaches designed to optimize the operation of such a system considering interactions among users in a game theoretic perspective. In this study, we propose the energy capacity trading and operation (ECTO) game where each agent determines two actions, capacity trading, and the 24-hour ahead charging-discharging scheduling with the capacity that will be assigned, to minimize the energy operation cost. We then propose a distributed optimization strategy to find a generalized Nash equilibrium for the proposed ECTO game. Simulation studies show that when optimally operated, a shared ESS can decrease both the total energy operating cost and the peak-to-average ratio of the energy for the entire grid compared to the conventional ESS control strategy without ESS sharing.
Due to the recent advances in manufacturing systems, the semiconductor FABs have become larger, and thus, more overhead hoist transporters (OHTs) need to be operated. In this article, we propose a ...cooperative zone-based rebalancing algorithm to allocate idle overhead hoist vehicles in a semiconductor FAB. The proposed model is composed of two parts: (i) a state representation learning part that extracts the localized embedding of each agent using a graph neural network; and (ii) a policy learning part that makes a rebalancing action using the constructed embedding. By conducting both representation learning and policy learning in a single framework, the proposed method can train the decentralized policy for agents to rebalance OHTs cooperatively. The experiments show that the proposed method can significantly reduce the average retrieval time while reducing the OHT utilization ratio. In addition, we investigated the transferable capability of the suggested algorithm by testing the policy on unseen dynamic scenarios without further training.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Neutral boundary layer (NBL) flow fields, commonly used in turbine load studies and design, are generated using spectral procedures in stochastic simulation. For large utility-scale turbines, stable ...boundary layer (SBL) flow fields are of great interest because they are often accompanied by enhanced wind shear, wind veer, and even low-level jets (LLJs). The generation of SBL flow fields, in contrast to simpler stochastic simulation for NBL, requires computational fluid dynamics (CFD) procedures to capture the physics and noted characteristics-such as shear and veer-that are distinct from those seen in NBL flows. At present, large-eddy simulation (LES) is the most efficient CFD procedure for SBL flow field generation and related wind turbine loads studies. Design standards, such as from the International Electrotechnical Commission (IEC), provide guidance albeit with simplifying assumptions (one such deals with assuming constant variance of turbulence over the rotor) and recommend standard target turbulence power spectra and coherence functions to allow NBL flow field simulation. In contrast, a systematic SBL flow field simulation procedure has not been offered for design or for site assessment. It is instructive to compare LES-generated SBL flow fields with stochastic NBL flow fields and associated loads which we evaluate for a 5-MW turbine; in doing so, we seek to isolate distinguishing characteristics of wind shear, wind veer, and turbulence variation over the rotor plane in the alternative flow fields and in the turbine loads. Because of known differences in NBL-stochastic and SBL-LES wind fields but an industry preference for simpler stochastic simulation in design practice, this study investigates if one can reproduce stable atmospheric conditions using stochastic approaches with appropriate corrections for shear, veer, turbulence, etc. We find that such simple tuning cannot consistently match turbine target SBL load statistics, even though this is possible in some cases. As such, when there is a need to consider different stability regimes encountered by a wind turbine, easy solutions do not exist and large-eddy simulation at least for the stable boundary layer is needed.
In this study, we propose a contextual Bayesian optimization with Trust-Region (CBOTR), an extended version of Bayesian optimization (BO) that can find an optimum input of a target system (or unknown ...function) through the iterative learning and sampling procedure. CBOTR adds two features to BO: (1) CBOTR can take into account context information which modifies the input and output relationship of a target system, and (2) CBOTR restricts the searching space for the next input to be selected so that it can rapidly find an optimum. The results from simulation studies using a set of benchmark functions and a wind farm power simulator showed that the CBOTR algorithm can achieve an almost optimum target value by taking a small number of trial actions (samplings). The proposed algorithm particularly suits well to determine the joint optimal operational conditions of wind turbines in a wind farm for maximizing the total energy production, in that the complex interaction among wind turbines in a wind farm is difficult to model using an analytical model and one needs to find the optimum operational conditions for varying wind conditions.
The hidden Markov model (HMM), used with Gaussian Process (GP) as an emission model, has been widely used to model sequential data in complex form. This study introduces the hybrid Bayesian HMM with ...GP emission using SM kernel (HMM-GPSM) to estimate the hidden state of each time-series observation, that is, sequentially observed from a single channel. We then propose a scalable inference method to train the HMM-GPSM using large-scale sequences of time-series dataset that has (1) a large number of sequences for state transitions and (2) a large number of data points in a time-series observation for each hidden state. For state transitions with a large number of sequences, we employ stochastic variational inference (SVI) to update the parameters of HMM-GPSM efficiently. Also, for each time-series observation that has a large number of data points, we propose the approximate GP emission using the Random Fourier Feature (RFF), which is constructed by using the spectral points that are sampled from the spectral density of SM kernel. We propose the efficient inference of the kernel hyperparameters of the approximate GP emission and corresponding HMM-GPSM. Specifically, we derive the training loss, that is, the evidence lower bound of the HMM-GPSM that can be scalably computed for a large number of time-series observations by employing the regularized lower bound of GP emission likelihood with KL divergence. The proposed methods can be used together to train HMM-GPSM with the sequential time-series dataset that contains both (1) and (2). We validate the proposed method on the synthetic and real datasets using the clustering accuracy, marginal likelihood, and training time as the performance metrics.