Numerical prediction model feature selection method based on model performance and calculation cost

• Main Authors: ZHAN HOUJIAN, TANG JIACHUN, HUANG RONGXIANG, GUI MEIPING, ZHUO YIXIN, XU XIAOHONG, QIN FANGLU, TANG JIAN, HUANG KUI, YANG YITING, QIN YIMING, QI HUANXING, HU JIAQIU, TAN XIN
• Format: Patent
• Language: Chinese; English
• Published: 01.03.2024
• Subjects: CALCULATING; CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTINGELECTRIC POWER; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; ELECTRICITY; GENERATION; PHYSICS; SYSTEMS FOR STORING ELECTRIC ENERGY

The invention discloses a numerical prediction model feature selection method based on model performance and calculation cost, and belongs to the technical field of feature screening. According to the method, the importance degree of each feature of the stacking model is effectively explained by adopting a model interpretability algorithm, further, a reliable basis is provided for the optimization problem of the stacking model by establishing a feature selection model, feature optimization selection in the numerical prediction model establishment process is realized in the two dimensions of model performance and calculation cost, and the calculation cost is reduced. The problems of trial and error, blindness, weak interpretability and the like in feature selection in the establishment process of a numerical prediction model are solved. 本发明公开了基于模型性能与计算代价的数值预测模型特征甄选方法，属于特征筛选技术领域，该方法包括：提出了基于模型性能与计算代价的数值预测模型特征甄选方法，通过采用模型可解释性算法对堆叠模型的各特征重要程度作了有效的解释，进一步，通过建立特征甄选模型，为堆叠模型的优选问题提供了可靠的依据，在模型性能和计算代价两个维度上，实现了数值预测模型建立过程中的特征