考虑相关因素的长短时记忆网络短期负荷预测方法

Short-term Load Forecasting Based on Long Short-term Memory Network Considering Related Factors

电力市场环境下短期负荷预测是电力系统发电调度计划的重要基础,负荷预测的准确性对电力系统安全经济运行具有重要意义。为考虑相关因素对短期负荷的影响,提高短期负荷预测的准确率,在历史负荷数据的基础上,引入天气、节假日等相关因素信息,构造长短时记忆(LSTM)网络模型对日前96点负荷进行预测。利用广东某市的2011-2015年的历史实际负荷数据作为训练数据,2016年的数据作为测试数据进行模拟预测,并与传统人工神经网络方法和单纯考虑历史数据的LSTM网络模型的结果作比较。预测结果表明,考虑相关因素的影响后,基于LSTM网络的全年短期负荷预测准确率达到97.6%,节假日的平均预测准确率达到95.8%,能够有效提高短期负荷预测的准确率。

Short-term load forecasting is the important fundamental of power system generation dispatching. The accuracy of short-term load forecasting is of great significance to the safe and economic operation of power system. In order to consider the impact of relevant factors of the short-term load and improve the accuracy of short-term load forecasting, the information of weather, holidays, typhoon and other related factors were introduced on the basis of historical load data to construct a long short-term memory(LSTM) network model. The LSTM network model was used to predict the 96-point load before the day. The historical load data of a city in Guangdong from 2011 to 2015 was used as training data, and the data of 2016 as testing data to predict the short-term load. Moreover, the forecasting results were compared with the results of the common neural network method and the LSTM network model considering historical data alone. The results show that the annual short-term load forecasting accuracy based on LSTM network reaches 97.6% and the average accuracy of holidays reaches 95.8% after considering the influence of relevant factors, which can effectively improve the accuracy of short-term load forecasting.