摘要
由于现有预测方法收益不稳定,预测收入值与实际值存在差异,文章研究了基于深度强化学习的年度售电收入预测方法。通过收入因素分析模型来判断售电量对售电收入的影响,并对关键因素进行了量化分析;应用深度强化学习的方法,通过深度神经网络对分析模型进行优化;采用策略梯度公式对策略向量进行处理,使得模型能够学会预测售电收入的最佳策略;将年度的售电均价数据代入到预测模型中,计算得到总售电收入预测值。实验结果表明,经800次迭代,收益基本稳定在48左右,实验组的预测收入情况稳定增长0.5~2.0,与实际情况一致,能够更加稳定捕捉预算收入变化,提升了整体售电收入预测的准确性。
Due to the unstable income of the existing forecasting methods,the predicted income value is different from the actual value,this paper studies the annual electricity sales revenue forecasting method based on deep reinforcement learning.The influence of electricity sales on electricity sales revenue is judged by income factor analysis model,and the key factors are quantitatively analyzed.The method of deep reinforcement learning is applied to optimize the analysis model through deep neural network.The strategy vector is processed by the strategy gradient formula,so that the model can learn the best strategy to predict the income from electricity sales.By substituting the annual average price of electricity sold into the forecast model,the forecast value of total electricity sold revenue is calculated.The experimental results show that the profit is basically stable at about 48 after 800 times.The predicted income of the experimental group increased steadily from 0.5 to 2.0,which was consistent with the actual situation.It can more stably capture changes in budget revenue,and improve the accuracy of the overall electricity sales revenue forecast.
作者
江再玉
王晓敏
张婷
JIANG Zaiyu;WANG Xiaomin;ZHANG Ting(Beijing Zhongdian Puhua Information Technology Co.,Ltd.,Beijing 100192,China)
关键词
深度强化学习
收入
售电
预测
年度
deep reinforcement learning
income
selling electricity
forecast
annual
