VayuAI Corp. (“Vayu”) has developed a technology platform (see Fig. 1) that optimizes wind farm performance, increasing energy production and reducing O&M costs. Our platform is distributed across systems in the cloud (“Vayu Cloud”) and the edge (Vayu Edge”), recommending turbine yaw misalignments and control parameters to:
Minimize the impact of wake on power output
Mitigate the impact of loads on turbines
Recently, Vayu received patent protection on this technology, “Cloud Based Turbine Control Feedback Loop” (USPTO Application No. 16/212,546). This invention enables wind farms to use high performance computing and machine learning systems to process unlimited data, perform simulations at the wind farm level that prescribe the most efficient yaw settings for any given any set of wind farm conditions.
Financially, these improvements have a significant impact. Preliminary computational studies of a wide range of US wind farms (based on publicly available data) suggest AEP gains averaging 5% and total increased rated capacity for US wind farms of 6GW. Additionally, costs for operations and maintenance can be reduced substantially with Vayu optimization (dependent on wind farm conditions can be 2% or more).
Data collected at wind turbines is passed to Vayu Edge (resident on a Qualcomm RB5 at the wind farm turbine control center) via 1) wind farm ethernet connection to a SCADA server and 2) separate 5G private network connection to sensors that collect vibration and loads data from turbines. Vayu Edge processes the data to establish current wind farm conditions in a multifaceted process that involves correcting for sensor errors, miscalibrations and noise using machine learning. The inferred weather conditions allow Vayu Edge to recommend the optimal yaw settings (in accordance with manufacturer warranties) for implementation by wind farm control systems managed by third parties (OEMs and engineering services teams).
All wind farm turbine data aggregated by Vayu Edge is passed to Vayu Cloud (which harnesses the power of cloud computing to access significant amounts of processing power, memory and data storage) for model training and validation. All of the models used at the edge are trained in the cloud. This includes determining errors and turbine miscalibrations, as well as fine tuning Vayu models for wake, turbulence and turbine loads to further enhance optimal turbine positioning and control settings. All models are updated in real time and pushed back to Vayu Edge for continued implementation. In addition to the above processes, Vayu Cloud uses advanced modeling techniques to simulate baseline wind farm performance (without a Vayu implementation) for comparison.
Cloud Solution - Wind Farm Simulation and Optimization
As noted above, the Vayu Cloud solution is multifaceted. It leverages data and computational power to train models for three different purposes critical to a well functioning wind farm controls implementation:
Determination of current weather conditions,
Selection of optimal turbine positions and control parameters
Creation of a baseline standard for later comparison.
Together, these models comprise a simulation of the wind farm that accurately models power output, turbulence and loads at each turbine under current and future conditions. This allows the system to consider different turbine positions and control parameters, predict performance and select optimal instructions accordingly.
Establishing current weather conditions is not always straightforward, thanks in part to common sensor errors and general miscalibration of turbine nacelle position. This process is often further impacted by turbulence, topographical features and mixing of different wind directions. To tackle these issues we train machine learning models designed to predict current weather at each turbine by minimizing the error in power output prediction. These models are based on past and current power output both for the turbine of interest and nearby ‘impacted’ turbines.
Selecting optimal turbine position and control parameters is rooted in simulations of turbine wakes, turbulence and loads. This accurately predicts the future state of the wind farm given a set of control instructions. If this can be done effectively, it is straightforward to choose the path that will maximize performance. These models are rooted in physics, relying on proprietary analytical wake models and aeroelastic simulations that eclipse the accuracy of the current state-of-the-art in academic literature. They are tuned to best fit the specific turbines and atmospheric dynamics at each site.
This process is then augmented by machine learning to improve performance and better fit the specific wind farm. Machine learning accounts for topography, multiple wind directions, turbine deterioration and other site specific factors that traditional physics-based models are generally unable to incorporate. As more data is collected over time the performance of machine learning improves, surpassing the physics-based models and eventually fully supplanting them.
These simulations are also effective in creating a baseline performance metric for comparison. If we can accurately model wind farm performance it is relatively simple to determine how the turbines would behave under normal operation and use that behavior as input to the model. The resultant output is an accurate picture of overall performance without a Vayu implementation.
Fig. 1: VayuAI Corp. Wind Systems Optimization Platform