Regional suitability assessment for straw-based power generation: A machine learning approach

Abstract

Assessing regional suitability of straw-based power generation (SPG) is critical to site selection on straw-based power plants (SPPs). However, the traditional evaluation methods often exhibit subjectivity in determining index weights, lacking a standard method for weight assignments This can hinder the quntification of evaluation accuracy. Machine learning (ML) can derive the feature importance based on empirical data and model training, and quantify the accuracy of model predictions. This study proposes an effective regional suitability assessment framework for SPG from the perspective of resource endowment, population and economic development based on ML. We selected 12 provinces with abundant agricultural straw resources in China and collect data of 1037 counties. 20 indicators with respect to resource endowment, population and economic development were mapped to feature variables of ML classifiers. Various classifiers including the base classifier, the Bagging classifier and the Boosting classifier were employed to predict the regional suitability of SPG. The results indicate that Gradient Boost Decision Tree(GBDT) and eXtreme Gradient Boosting (XGBoost) are the optimal models for predicting regional suitability of SPG in different scenarios respectively. GBDT has the highest Precision (97.94%) and can avoid wrong selection of counties suitable for SPG, while XGBoost has the highest Recall (96.67%) and can avoid miss selection of counties suitable for SPG. The feature importance ranking are also identified based on the optimal models. The results not only provide an effective tool for regional suitability evaluation of SPG but also provide important reference for the preliminary site selection of SPPs.