GBDT与感知机融合的充电桩故障诊断方案
Fault Diagnosis of Charging Piles by Fusing Gradient Boosting Decision Tree and Perceptron
张震 1郭恩伯 2郭恩仲 3许成乾4
作者信息
- 1. 天津平高易电科技有限公司,天津 300000
- 2. 合肥工业大学管理学院,合肥 230009
- 3. 西交利物浦大学先进技术学院,苏州 215028
- 4. 天津大学 电气自动化与信息工程学院,天津 300072
- 折叠
摘要
为解决充电桩故障诊断中普遍存在的特征提取难题以及复杂的人工智能模型容易产生过拟合问题,该文提出一种基于GBDT与MLP融合的新方法.根据集成学习理论,建立多个独立的GBDT,组建多个全连接的GBDT层,最后连接一个MLP进行特征学习与分类.该融合方案避免了手动提取特征的困难,在降低对单个模型性能和复杂度依赖的同时,其融合性能得到提升而且更加稳定.在公开数据集上的实验结果表明,该文提出的方案优于典型的独立机器学习方案.
Abstract
In order to solve the common problems of feature extraction and the over-fitting of complex artificial intel-ligence models for fault diagnosis of charging piles,a new method based on gradient boosting decision tree and mul-ti-layer perceptron fusion is proposed in this paper.According to the ensemble learning theory,multiple independent models based on gradient boosting decision tree are established,which are used to construct multiple fully connected layers of gradient boosting decision tree,and finally a multi-layer perceptron is arranged at the end of the network for feature learning and classification.This fusion scheme avoids the difficulty on manually extracting features.More-over,while reducing the dependence on the performance and complexity of an individual model,the fusion perfor-mance is still improved and more stable.Experimental results on the public dataset show that the proposed scheme is superior to typical individual machine learning schemes.
关键词
充电桩/故障诊断/集成学习/GBDT/MLPKey words
charging pile/fault diagnosis/ensemble learning/gradient boosting decision tree(GBDT)/multi-layer per-ceptron(MLP)引用本文复制引用
基金项目
天津平高充电桩管理平台后端业务实现和功能测试服务项目(2021GKF-1093)
出版年
2024
NETL
NSTL
万方数据