在一个我们除了增加日常生活用电而无所事事的社会里，能源消耗和相应的管理是一个重大问题。电力需求预测是电力系统运营商在电网管理中的重要组成部分。由于设计和确定合适的可再生能源系统和储能规模，预测特定家庭每日能源消耗的重要性也与最终用户有关。

本论文的目的是开发并训练一个能够以尽可能高的精度预测家庭用电量的计算系统。本文提出了一种基于人工神经网络（ANN）的短期负荷预测方法（STLF），在住宅能耗不可预测的情况下，仍能得到准确的预测结果。对记录的数据进行了分析，包括2015年至2018年特定家庭的每日用电量轨迹。随后，对神经网络的结构和训练算法进行了研究，以确定一个鲁棒的模型。此外，我们用不同的模型进行了实验，这些模型包含不同的输入，目的是比较不同参数对网络训练的相关性。最后，根据整个研究中收集到的见解，对优化模型进行预测，并在几个特定时间段进行比较。结果表明，在适当的输入和选择超参数的情况下，浅层神经网络可以为电力需求预测提供一定的精度。同时，开发和训练了基于人工神经网络的方法。

In a society where we do nothing but increase the use of electricity in our daily life, energy consumption and the corresponding management is a major issue. The prediction of electric energy demand is a key component, for the power system operators, in the management of the electrical grid. The importance of forecasting a particular household daily energy consumption does concern the end-user too, by reason of the design and sizing of a suitable renewable energy system and energy storage. The aim of this thesis is to develop and train a computing system capable of predicting, with best accuracy as possible, electricity consumption at household-level. This paper presents a Short Term Load Forecasting (STLF) with Artificial Neural Networks (ANN), which lead to accurate results in spite of the dwelling consumption unpredictability. The recorded data, containing the daily track of electricity consumption over a particular household from 2015 to 2018, was analysed. Subsequently, a study over the ANN architecture and training algorithms was carried out in order to define a robust model. Furthermore, several experiments were conducted with different models, containing distinct inputs, aiming to compare the relevance of a diversity of parameters for the network’s training. Finally, the forecasting of the optimal models, created with the insights collected over the whole research, was performed and compared in several specially selected time periods. The results showed how with the appropriate inputs and selection of hyperparameters, a shallow ANN can provide certain accuracy on the forecasting of electric energy demand. As well as a methodology to develop and train an artificial neural network.

1.  研究动机与项目背景

2.  引言

3.  人工神经网络ANN

4.  探索数据分析EDA

5.  构建过程

6.  研发测试

7.  训练过程

8.  预测

9.  成本分析

10.          环境影响

更多精彩文章请关注公众号：