基于机器学习的短期电力负荷预测方法比较及改进研究

Comparison and improvement of short-term power load forecasting methods based on machine learningMonthly electricity load forecast based on time series analysis and SaDE-LSTM algorithm

摘要:

短期电力负荷预测的精确性是电力调度的基础和电力系统供需平衡的重要支撑，本文提出了一种基于机器学习的混合算法（SaDE-LSTM）对电力负荷进行短期预测。这一混合算法首先利用差分进化算法的自适应变异和交叉因子来优化长短期记忆算法的初始参数，在此基础上，运用寻优得到的参数训练长短期记忆算法从而得到优化后的预测结果。基于我国2004-2018年间月度社会用电负荷数据，本文对改进后的混合算法进行性能测试，对比采用支持向量机、反向传播神经网络、自回归积分滑动平均等模型的预测结果，通过各方法预测结果和真实结果对比分析证明，本文提出的SaDE-LSTM算法，在降低了对时间序列数据量的要求的基础上，相比传统算法有更高的预测精度，为虚拟电厂、负荷聚合商等对小样本和高精度预测有需求的主体提供参考。

Abstract:

The accuracy of short-term power load forecasting is the basis of power dispatching and guarantee supply and demand balance of power system. This paper proposes a hybrid algorithm based on machine learning (SaDE-LSTM) to predict the power load in a short term. In this hybrid algorithm, the parameters of the LSTM are first optimized by the adaptive mutation and crossover factor of the differential evolution algorithm, and then the LSTM is trained with the parameters obtained by the pre-optimization to obtain the optimal result. Based on the monthly social electricity load data from 2004 to 2018 in China, the performance test of the improved hybrid algorithm is carried out, and it is compared with Back Propagation (BP) neural network, support vector machine, and Autoregressive Integrated Moving Average (ARIMA) prediction model. The results show that the SaDE-LSTM algorithm proposed in this paper reduces the requirements for the amount of time series data, and has higher forecasting accuracy than traditional algorithms, which provides a reference for the subjects demanding for small samples and high-precision prediction such as virtual power plants and load aggregators.