基于神经网络的房地产项目成本智能化归集研究

彭宇涵

摘  要

近年来，房地产企业利润日趋微薄，利润增速呈下降趋势，房地产项目成本管理对提高房地产企业利润的影响越来越大。在房地产全过程成本管理中，动态成本控制阶段尤为重要，成本归集是动态成本控制中的关键环节，即将签约合同清单项金额与房地产企业目标成本科目进行匹配，以便于在项目实施阶段通过统一的成本科目口径对比分析本项目目标成本是否超支。现阶段成本归集工作主要依靠人工操作，成本归集的准确性直接取决于成本管理人员的项目管理经验，这样的做法造成过往已完项目的经验数据无法共享。

本文聚焦成本归集工作，研究如何通过人工智能技术实现成本项目自动匹配，形成成本归集的标准化体系流程，减轻成本管理人员的工作负担。通过对成本归集工作的目的和流程进行深入研究，并在大量分析文本分类相关研究的基础上，本文提出通过CNN神经网络拟合现有成本归集分类依据和经验的模型，实现合同清单项与目标成本科目自动匹配的功能。

本文首先研究了文本预处理的相关方法，重点研究了one-hot词向量嵌入、Word2Vec词向量模型等文本预处理方法，分析了不同文本处理方法的优缺点，最终确定以Word2Vec模型作为本文词向量生成模型。然后通过试验对比分析了CNN、RNN、LSTM三种模型在本文数据分类情况下的性能，最终选取基于CNN神经网络的分类模型。在确定了文本预处理和分类模型的前提下，本文构建了基于CNN的成本项目自动匹配模型，并针对模型参数进行了试验优化，同时针对本文文本数据的自身缺陷进行了数据优化，最终实现了成本项目自动匹配模型性能提升，成功解决了本文开篇提出的问题。最后根据本文研究的不足之处提出改进意见和进一步研究的方向。

关键词： 房地产成本管理；成本归集；成本项目自动匹配；文本分类；CNN

Abstract

In recent years, the profits of real estate enterprises have become increasingly meager, and the growth rate of profits has shown a downward trend. The cost management of real estate enterprises has an increasing impact on improving the profits of real estate enterprises. In the whole-process cost management of real estate, the dynamic cost control stage is particularly important. Cost collection is a key link in dynamic cost control, that is, the amount of the contract list item is matched with the target cost account of the real estate enterprise, so as to facilitate the unification of the project implementation stage Compare and analyze whether the target cost of this project is overrun. At this stage, the cost collection work mainly relies on manual operations, and the accuracy of cost collection directly depends on the project management experience of cost managers. This approach results in that the experience data of completed projects in the past cannot be shared.

This paper focuses on cost collection work, and studies how to achieve automatic matching of cost items through artificial intelligence technology, form a standardized system process for cost collection, and reduce the workload of cost management personnel. Through in-depth research on the purpose and process of cost collection work, and on the basis of a large number of studies related to text classification, this paper proposes a model that fits the existing cost collection classification basis and experience through CNN neural network to realize contract list items. The function of automatic matching with target cost accounts.

This paper firstly studies the related methods of text preprocessing, focusing on text preprocessing methods such as one-hot word vector embedding, Word2Vec word vector model, etc., analyzes the advantages and disadvantages of different text processing methods, and finally decides to use the Word2Vec model as the word vector in this paper. Generate the model. Then, the classification model suitable for the data format of this paper is selected, and the performance of the three models of CNN, RNN, and LSTM in the case of data classification in this paper is compared and analyzed through experiments, and the classification model based on CNN neural network is finally selected.On the premise of determining the text preprocessing and classification models, this paper builds a cost item automatic matching model based on CNN, and conducts experimental optimization for model parameters. The performance of the cost item automatic matching model is improved, and the problem raised at the beginning of this paper is successfully solved. Finally, according to the shortcomings of this study, suggestions for improvement and further research directions are proposed.

Key words: Real estate cost management, Cost collection, Cost item automatic matching, Text-calssification, CNN