基于联合模型的短期电力负荷预测方法

Short term Load Forecasting Method Based on Combined Model

摘要:

为了准确预测电力负荷并提高电力系统调节和调度的灵活性、准确性，提出了基于差分自回归滑动平均和长短期记忆神经网络的短期负荷联合模型预测方法，以避免单一预测模型可能难以满足预测准确需求的情况。首先，使用差分自回归滑动平均和长短期记忆神经网络单一模型对短期电力负荷开展预测；然后，使用改进的粒子群优化算法对联合模型权重进行寻优；最后，利用最优权重将单一模型预测结果进行合并得到最终的预测结果。验证结果表明，所建立的联合模型能够对短期电力负荷进行准确的预测，且联合模型的预测精度要优于差分自回归滑动平均、长短期记忆神经网络和BP神经网络等单一模型，具有一定的工程应用价值。

Abstract:

In order to accurately forecast power load and improve the flexibility and accuracy of power system regulation and scheduling, a short term load forecasting method based on combined model of auto regressive integrated moving average (ARIMA) and long short term memory (LSTM) neural network is proposed to avoid that a single prediction model may be difficult to meet the prediction accuracy requirement. Firstly, the two single models of ARIMA and LSTM are used to forecast the short term load, and then the hybrid particle swarm optimization (PSO) algorithm is used to optimize the weight of combined model. Finally, the forecasting results of the single model are combined with the optimal weight to obtain the final forecasting result. The verification results show that the proposed combined model can accurately forecast the short term load, and its forecasting accuracy is better than that of single models of ARIMA, LSTM and back propagation neural network (BPNN), which has certain engineering application value.