Performing Department
(N/A)
Non Technical Summary
Revenue from wind energy is an important source of stable income for small and mid-size farms as the agricultural sector becomes more volatile. One of the biggest challenges facing wind energy development on small and mid-sized farms are the adverse effects of wind farms on bats. To reduce collision-related mortality of bats, curtailment -- stopping wind turbines from spinning during times of high bat activity -- is employed. While beneficial for bats, curtailment results in lost renewable energy and lost revenue, which reduces the annual lease payment to small and mid-size farms. This project aims to develop next-generation smart curtailment algorithms that save just as many bats as current technologies for less lost renewable energy. Smart curtailment involves fitting an algorithm to environmental data to predict times that are risky for bats (times to curtail) or not (times to not curtail). No current technologies acknowledge the dynamic cost of curtailment across wind speeds. The project team will design and implement a dynamic cost function for decision tree algorithms. The dynamic cost function will allow the algorithm to optimize a smart curtailment regime by explicitly accounting for the cost of curtailment at different wind speeds (the cost of curtailment is proportional to a cubic function of wind speed). The new software will be applied to prototypical data to assess its efficacy. The proposed software solution will create a win-win-win for bat conservation, small and mid-size farms that receive a share of wind energy revenues, and renewable energy generation in the United States.
Animal Health Component
50%
Research Effort Categories
Basic
(N/A)
Applied
50%
Developmental
50%
Goals / Objectives
The major goal of this project is to design, develop, and deploysoftware capable of fitting optimal smart curtailment algorithms for wind facilities on small and mid-size farms, a software solution to be called Optimized Smart CurtailmentSM. Smart curtailmentaims to identify environmental conditions under which bats are most at risk for collision with wind turbines and curtail (stop) operations during those times. Deploying smart curtailment requires an algorithm that classifies environmental conditions as risky to bats (curtail) or not (do not curtail). To fit optimal algorithms, the classification routine must account for the dynamic cost of curtailment: stopping wind turbines from spinning costs more in terms of lost renewable energy as wind speed increases. Thus, the underlying scientific objective is to developed a statistical approach and associated software for fitting classification routines with a dynamic cost function.To achieve the project's overarching goal of next-generation smart curtailment algorithms, four objectives must be met:Design and develop a dynamic cost function that can be used to fit classification and regression trees.Develop production-ready, scalable software that implements the dynamic cost function for classification and regression trees.Test Optimized Smart CurtailmentSM using simulated data.Develop marketing materials and a go-to-market strategy for Optimized Smart CurtailmentSM.
Project Methods
The dynamic cost function will be designed by extending an existing statistical method, classification and regression trees. Specifically, the cost matrix of the classification algorithm will be made a function of an auxilliary covariate (wind speed) and a new classification score will be developed to account for different costs per observation within a classification node. Computer code in the R statistical programming language will be writtinto implement the new cost function and to interact with existing software tools (e.g., R packages) for fitting classification algorithms. The entire suite of software will be bundled into an R package to generate scalable and reproducible smart curtailment algorithms.The new dynamic cost function will be evaluated bydetermining1) how well the newapproach approximates current classification modeloutput when given a "dynamic" cost function that is actually fixed, and 2) the relative stability of the classification output given different training data sets that are generated from the same global classification rules. The approach will be evaluated using prototypicaldata toquantifythe bat conservation benefit and renewable energy production using traditional curtailment approaches and the new Optimized Smart CurtailmentSMapproach, which uses the dynamic cost function.