为解决电网物资品种繁多、规格多样、数量巨大、用途广泛、受政策和投资影响大等特点所导致的预测模型构建困难的问题。首先，通过德尔菲法和灰色关联分析法（gray correlation analysis, GRA）筛选影响基建、业扩及抢修项目物资需求数量的因素。其次，利用引入自适应惯性因子和学习因子的改进粒子群算法调整极限学习机的最佳参数组合，训练各类配网项目物资需求预测模型。最后，以南方电网深圳市某供电局2020—2022年基建项目10 kV电力电缆需求情况为例，将GRA-IPSO-ELM（grey relational analysis, improved particle swarm optimization, and extreme learning machines）德尔菲法和灰色关联分析法模型与常见的4种预测模型的结果进行对比。结果表明，相较于ELM模型、支持向量机模型以及PSO-ELM模型，GRA-IPSO-ELM模型预测准确率得到10.38%、5.37%、3.83%的提升，可见，所提出的模型实现了对配网物资需求数量准确且高效的预测。

In order to solve the problem of difficulty in constructing forecasting models caused by the characteristics of power grid materials, such as many varieties, diverse specifications, huge quantities, wide range of uses, and great influence by policies and investments. Firstly, the factors affecting the quantity of material demand for infrastructure, business expansion, and emergency repair projects were screened by the Delphi method and gray correlation analysis (GRA). Secondly, an improved particle swarm algorithm that introduced adaptive inertia factor and learning factor was utilized to adjust the optimal parameter combinations of the extreme learning machine, and train the material demand prediction models for various distribution network projects. Finally, the results of the GRA-IPSO-ELM (grey relational analysis, improved particle swarm optimization, and extreme learning machines) model were compared with the results of four common forecasting models by taking the demand of 10 kV power cables of a power grid for 2020—2022 infrastructure projects as an example. The results show that the prediction accuracy of the GRA-IPSO-ELM model is improved by 10.38%, 5.37% and 3.83% compared with the ELM model, the support vector machine model and the PSO-ELM model, which shows that the model proposed in this paper realizes accurate and efficient prediction of the quantity of material demand in the distribution network.