Deep Learning Method For Load Forecast In Smart Solar System Based Long Short-Term Memory

Predicting the load power of a solar power plant is crucial for maximizing profits from available solar energy. This research introduces a deep learning approach using the Long Short-Term Memory (LSTM) model to forecast load power. The developed method improves the performance and accuracy of hourly predictions for the full load power of commercial buildings integrated with a solar power plant. Load operations are typically affected by weather parameters, such as temperature, dew point, humidity, pressure, wind direction, and wind speed, which all influence load power and are considered key factors. Additionally, an advanced metering infrastructure collects internal parameters of solar microgrids over the span of a year, including battery discharge power. These additional parameters were also used as input features in the proposed method. The LSTM approach is specifically utilized to forecast the full load power of the building. The performance and accuracy of the LSTM model were compared with other deep learning methods using MSE and MAE benchmarks. The comparative outcomes show that the LSTM approach achieved higher performance, with maximum improvements of 39.64% in MSE and 27.05% in MAE for the training data, and 12.58% in MSE and 19.94% in MAE during the validation process.