基于改进LSTM的电力负荷预测与成本感知优化策略研究

Research on power load forecasting and cost perception optimization strategy based on improved LSTM

针对企业用电负荷预测计算复杂且准确度较低的问题,提出了一种基于改进粒子群算法(IPSO)与随差遗忘长短期记忆时间网络(EFFG-LSTM)的电力负荷预测与成本感知优化方法。该模型针对传统LSTM模型超参数随机选取的缺陷,利用IPSO算法实现了超参数的自适应寻优。对于LSTM单元缺乏误差跟随能力、易出现梯度消失的问题,采用一种随差遗忘门结构使其能够跟踪上一时刻的预测误差,同时动态调整遗忘门参数,并使用ReLU作为LSTM单元的激活函数。通过仿真算例表明,所提IPSO-EFFG-LSTM相比于EFFG-LSTM及LSTM模型在电力负荷预测上具有更高的准确度,且平均误差仅为5.2%。

Aiming at the problems of complex calculation and low accuracy of enterprise power load forecasting,a power load forecasting and cost perception optimization model based on Improved Partical Swarm Optimization(IPSO) and Error Following Forget Gate Long Short-Term Memory(EFFG-LSTM) is proposed.Aiming at the disadvantage that the super parameters of the traditional LSTM model are randomly selected,the IPSO algorithm is used to realize the adaptive optimization of the super parameters.For the shortcomings that LSTM unit lacks error following ability and is prone to gradient disappearance,a random difference forgetting gate structure is adopted to track the prediction error of the previous time,dynamically adjust the forgetting gate parameters,and use ReLU function as the activation function of LSTM unit.The simulation example shows that the IPSO-EFFG-LSTM proposed in this paper has higher accuracy in power load forecasting than EFFG-LSTM and LSTM models,and the error is only 5.2%.