时间卷积网络在热负荷预测中的应用

Application of temporal convolutional network in heat load prediction

针对热力站优化控制中,因热负荷数据不准确,热负荷预测准确度较低,导致难以实现热力站热量按需分配的问题,本文提出了采用面积法拟合热负荷数据。该方法通过推导公式建立热负荷与室外温度的数学关系,得到热负荷曲线的趋势;采用积分运算的方式,使热负荷曲线面积与瞬时热量曲线面积相等,计算出热负荷曲线应整体移动的尺度,最终得到拟合的热负荷曲线。搭建时间卷积网络(Temporal Convolutional Network,TCN)模型,对拟合后的热力站热负荷数据预测,并与长短期记忆网络(Long Short-term Memory,LSTM)、反向传播神经网络(Back Propagation,BP)对比。实验结果证明,TCN相较于其它算法误差更小,可以实现精准热负荷预测。

To address the problem of difficulty in realizing heat distribution on demand at heat stations caused by the low accuracy of heat load prediction due to inaccurate heat load data in realizing optimal control of heat stations,the article uses the area method to fit the heat load data.The mathematical relationship between heat load and outdoor temperature is established by deriving formulas to obtain the trend of heat load curve.The integral operation is used to make the area of the heat load curve equal to the area of the instantaneous heat curve,the scale at which the heat load curve should move as a whole is calculated,and the fitted heat load curve is finally obtained.A Temporal Convolutional Network(TCN)model is built to predict the heat load data of the fitted heat station and compared with long short-term memory(LSTM)and back propagation neural network(BP).The experimental results prove that TCN has smaller error compared with other algorithms and can achieve accurate heat load prediction.