基于PSO优化BiLSTM的联合循环电厂输出功率预测

Output Power Prediction of Combined Cycle Power Plant Based on PSO Optimized BiLSTM

针对联合循环电厂发电能力受环境温度、压力、相对湿度和电力需求等条件变化影响而造成对输出功率预测精度较差的问题,提出粒子群算法(PSO,particle swarm optimization)与BiLSTM(BiLSTM,bi-directional long short-term memory)相结合的预测模型PSO-Bi LSTM。利用PSO的寻优能力对BiLSTM的隐含层神经元个数、周期次数、学习率、批大小值等参数进行优化,不仅实现自动调参,而且进一步提高了BiLSTM模型在电厂输出功率预测性能。最后依据优化参数建立PSO-BiLSTM预测模型对UCI标准数据集进行预测。实验结果表明,PSO-Bi LSTM模型的均方根误差、平均绝对百分比误差等指标均优于所列举的典型算法以及优化组合算法,模型在预测电力负荷数据方面有较高的精度。

In view of the poor prediction accuracy of the power generation capacity of combined cycle power plant caused by the changes of ambient temperature, pressure, relative humidity and power demand, a prediction model PSO-BiLSTM is proposed which combines PSO and BiLSTM. The number of hidden layer neurons, cycle times, learning rate, batch size and other parameters of BiLSTM are optimized by using the optimization ability of PSO, which not only realizes automatic parameter adjustment, but also further improves the output power prediction performance of BiLSTM model in power plants. Finally, the PSO-BiLSTM prediction model is established according to the optimized parameters to predict the UCI standard dataset. The experimental results show that the root mean square error and mean absolute percentage error of PSO-BiLSTM model are better than the listed typical algorithms and optimal combination algorithms, and the model has high accuracy in predicting power load data.