基于矩阵分解的电力物资需求预测

Prediction of Power Grid Material demand based on Based on matrix decomposition

摘要:

准确预测变电站及配网工程的物资需求,对于节约工程成本,提高资金利用率,具有重要意义.尽管研究者在电力物资需求预测方面已经开展了一系列的研究,提出了很多预测模型和算法,例如基于神经网络的算法,然而,这些算法普遍存在输入数据过于简单和理想、仅对少数几种物资的需求量进行了预测实验、预测的准确率偏低等不足.因此,目前生产系统普遍采用人工方式进行电力物资需求预测,由有经验的领域专家根据工程初步设计方案预测各类物资的需求量.为了解决现有电力物资需求预测方法存在的不足,本文提出基于矩阵分解的预测方法,以电网建设项目物资需求历史数据和项目计划的部分物资作为输入,通过矩阵分解算法对项目其他物资需求用量进行预测.矩阵分解算法不需要大量的历史用量数据,只用部分项目的物资数据就能进行预测,且算法不需要提前进行训练.

Abstract:

It is of great significance to accurately predict the material demand of substation and distribution network for saving project cost and improving capital utilization. Researchers have carried out a series of studies on power material demand forecasting, and proposed many prediction models and algorithms such as neural networkbased algorithms. However, these algorithms have several disadvantages. Specifically, these algorithms can only process simple and ideal input, predict the demand of limited materials, and suffer from the problem of low accuracy. As a result, in current production systems, the demand of electric power materials is predicted by experienced experts according to the preliminary design scheme of the project manually. In order to solve the existing shortcomings of the current demand forecasting methods, this paper proposes a forecasting method based on matrix decomposition. The method takes the historical data of the power grid construction project material requirements and part of the project plan as input, and use matrix decomposition algorithm to predict the demand for other materials in the project. The matrix decomposition algorithm can be implemented with the material data of some projects instead of a large amount of historical usage data. In addition, the developed model does not need to be trained in advance.