Renewable energy resources have gathered substantial interest, and several nations are striving to use them as the dominant power resource. However, the power output from these energy sources is inherently uncertain due to their reliance on natural forces like wind, sunlight, tides, geothermal, etc. An accurate estimation of expected consumer load demand can assist with scheduling and coordination between various generating units, ensuring a consistent supply of power to consumers. Internet of Things (IoT) devices are becoming ubiquitous in all technological domains and making different kinds of data readily available. This data from heterogenous IoT sources can be combined and applied towards rapid, short-term load forecasting. This work proposes a Long Short-Term Memory (LSTM) based load prediction model that combines weather data, historical and current load demand to project the hour-ahead load demand. LSTMs are excellent for picking out patterns in time series data and learning long-term dependencies, allowing them to predict over a prolonged period. Using our LSTM model, we obtained a Mean Absolute Percentage Error (MAPE) of 0.62% on the hour-ahead forecast. We further enhanced this model using Wavelet Transforms (WT-LSTM) and observed an improvement of 16% over LSTM model. Both models performed significantly better than their equivalent Artificial Neural Network (ANN) model counterparts, with LSTM and WT-LSTM outperforming the ANN and WT-ANN by 50%, respectively. Short term load forecasts from models predicting on such streaming data from IoT sensors can be used to do rapid generator balancing, thus making the grid more reactive to changes and capable of providing a reliable power supply.

可再生能源引起了极大的兴趣，一些国家正在努力将其作为主要的电力资源。然而，由于这些能源依赖于自然力，如风、阳光、潮汐、地热等，因此它们的功率输出本质上是不确定的。准确估计预期的消费者负荷需求可以帮助各种发电机组之间的调度和协调，确保为消费者提供稳定的电力供应。物联网 (IoT) 设备在所有技术领域变得无处不在，并使不同类型的数据随时可用。这些来自异构物联网来源的数据可以组合起来，用于快速、短期的负荷预测。这项工作提出了一种基于长短期记忆（LSTM）的负载预测模型，该模型结合了天气数据，历史和当前负载需求来预测提前一小时的负载需求。LSTM 非常适合在时间序列数据中挑选模式和学习长期依赖关系，从而使它们能够在很长一段时间内进行预测。使用我们的 LSTM 模型，我们在提前一小时的预测中获得了 0.62% 的平均绝对百分比误差 (MAPE)。我们使用小波变换 (WT-LSTM) 进一步增强了这个模型，并观察到比 LSTM 模型提高了 16%。两种模型的性能都明显优于等效的人工神经网络 (ANN) 模型，LSTM 和 WT-LSTM 的性能分别优于 ANN 和 WT-ANN 50%。来自物联网传感器的此类流数据预测模型的短期负载预测可用于快速发电机平衡，