考虑温度模糊化的多层长短时记忆神经网络短期负荷预测

智能电表的普及为短期负荷预测提供了海量数据，使得负荷精细化预测成为可能，而温度是影响夏季负荷的重要因素。提出一种考虑温度模糊化的多层长短时记忆神经网络(ML-LSTM)短期负荷预测方法。利用隶属度函数将预测时刻的温度和当日的平均温度进行模糊化处理，减小夏季温度波动性对负荷预测的影响；建立含3层隐藏层的长短时记忆神经网络(LSTM)预测网络，并利用适应性矩估计(Adam)优化算法提高LSTM梯度参数的自适应性学习能力。利用西南某地区2018年6月至8月的实测温度和负荷数据进行验证，负荷预测结果表明，ML-LSTM模型比BP神经网络和支持向量机的负荷预测精度更高，且温度的模糊化处理提高了模型的泛化性。

Short-term load forecasting of multi-layer long short-term memory neural network considering temperature fuzziness

The popularity of smart meters provides a large amount of data for short-term load forecasting, which makes detailed load forecasting possible, and temperature is an important factor that affecting summer loads. A short-term load forecasting method of ML-LSTM(Multi-Layer Long Short-Term Memory neural network) considering temperature fuzziness is proposed. The membership function is used to blur the temperature at the forecasting time and the average temperature of the day, which reduces the influence of summer temperature fluctuation on load forecasting. A LSTM(Long Short-Term Memory) forecasting network with three hidden layers is built, and Adam(Adaptive moment estimation) optimization algorithm is used to improve the adaptive learning ability of LSTM gradient parameters. The actual-measured temperature and load data from June to August of 2018 in a certain area of southwest region are taken for verification, and the load forecasting results show that the load forecasting accuracy of ML-LSTM model is higher than that of BP neural network and support vector machine, and the fuzziness treatment of temperature improves the generalization of the model.