提出一种短期负荷预测算法,用于解决对未来能耗周期能源使用的预测问题。首先介绍短期负荷特点,分析短期负荷运行规律,并采用零相滤波器对原始负荷曲线进行预处理,相除奇异点。其次,介绍BP神经网络基本结构,并针对BP神经网络容易陷入局部极小值的缺点,采用PSO算法确定网络训练初始权值。然后,设计一种基于PSO⁃BP神经网络的短期负荷预测算法,包括预滤波、训练样本集建立、神经网络输入/输出模式设计、神经网络结构确定等。最后,选择上海市武宁科技园区的电科商务大厦进行负荷预测,实验结果表明,与传统的BP神经网络相比,PSO⁃BP神经网络用于短期负荷预测算法的精度更高,预测负荷和实际负荷之间的平均绝对误差(MAE)小于1%。% An Algorithm of short⁃time load forecast(STLF)is proposed to solve the problem of forcasting the energy appli⁃cation within the power consumption period in the future. The characteristics of STLF are introduced. A zero phase filter is adop⁃ted to preprocess the original load curve for removing fault record. The basic structure of BP neural network is introduced. Ai⁃ming at BP neural network’s shortcomings that is easy to arrive to local minimum value,PSO is used to determine the initial weight of network training. A STLF algorithm based on PSO⁃BP neural network was designed,including pre⁃filtering,training sample set establishment , I/O pattern design of neural network , structure determination of neural network , etc. Electrical Science Building was chosen to test the precision of the proposed algorithm. The results of experiment show that STLF of PSO⁃BP neural network has higher accuracy than that of the traditional BP neural network,and the MAE between forecasted load and real load is less than 1%.
展开▼