ORC is a heat to power solution to convert low-grade thermal energy into electricity with relative low cost and adequate efficiency. The working of ORC relies on the liquid–vapor phase changes of certain organic fluid under different temperature and pressure. ORC is a well-established technology utilized in industry to recover industrial waste heat to electricity. However, the frequently varied temperature, pressure, and flow may raise difficulty to maintain a steady power generation from ORC. It is important to develop an effective prediction methodology for power generation in a stable grid system. This study proposes a methodology based on deep learning neural network to the predict power generation from ORC by 12 h in advance. The deep learning neural network is derived from long short-term memory network (LSTM), a type of recurrent neural network (RNN). A case study was conducted through analysis of ORC data from steel company. Different time series methodology including ARIMA and MLP were compared with LSTM in this study and shows the error rate decreased by 24% from LSTM. The proposed methodology can be used to effectively optimize the system warning threshold configuration for the early detection of abnormalities in power generators and a novel approach for early diagnosis in conventional industries.

中文翻译：

---

基于深度学习的热能转换发电预测

ORC 是一种热电解决方案，可将低品位热能转化为电能，成本相对较低，效率也足够高。ORC的工作依赖于某些有机流体在不同温度和压力下的液-汽相变化。ORC 是一种成熟的工业技术，用于将工业废热回收为电能。然而，频繁变化的温度、压力和流量可能会增加维持 ORC 稳定发电的难度。在稳定的电网系统中开发有效的发电预测方法非常重要。本研究提出了一种基于深度学习神经网络的方法，可提前 12 小时预测 ORC 的发电量。深度学习神经网络源自长短期记忆网络（LSTM），一种循环神经网络（RNN）。通过分析钢铁公司的 ORC 数据进行了案例研究。在本研究中，包括 ARIMA 和 MLP 在内的不同时间序列方法与 LSTM 进行了比较，结果表明错误率比 LSTM 降低了 24%。所提出的方法可用于有效优化系统预警阈值配置，用于发电机异常的早期检测和传统行业早期诊断的新方法。