改进Elman神经网络在短期热负荷预测中的应用

热电厂的短期热负荷预测在城市集中供暖中起着至关重要的作用,直接影响热电厂的经济效益和热能利用率。电厂的短期热负荷一般采用神经网络预测模型进行预测,而BP神经网络应用最为广泛。Elman神经网络算法在BP神经网络基础上加入了承接层,作为一步延时算子,实现记忆能力,使系统具备适应时变能力,增强系统全局稳定性。但Elman神经网络算法模型的构造依然需要大量样本的支撑,而且输入层的变量多,导致预测时间依然很长,收敛速度慢。该文在Elman神经网络预测前,进行了相关系数预处理和对样本中异常值的平均化预处理,通过数据归一化运算,使Elman神经网络输入层变量大幅减少。仿真实验表明,改进的Elman神经网络算法使预测模型快速寻优,减少预测时间的同时明显提高预测精度。

Application of improved Elman neural network in short-term thermal load forecasting

The short-term thermal load forecasting of thermal power plants plays a vital role in urban central heating,directly affecting the economic benefits and thermal energy utilization of thermal power plants.The short-term thermal load forecasting of power plants is generally predicted by neural network prediction models,while BP neural networks are the most widely used.The Elman neural network algorithm adds a receiving layer to the BP neural network as a one-step delay operator to realize the memory ability,so that the system can adapt to the time-varying ability and enhance the global stability of the system.However,the construction of the Elman neural network algorithm model still needs a large number of samples to support,and the input layer has many variables,resulting in a long prediction time and a slow convergence rate.In this paper,before the Elman neural network prediction,the correlation coefficient preprocessing and the averaging preprocessing of the outliers in the sample are performed.Through the data normalization operation,the input layer variables of the Elman neural network are greatly reduced.Simulation experiments show that the improved Elman neural network algorithm makes the prediction model fast optimization,reduces the prediction time,and significantly improves the prediction accuracy.