Power Prediction Based on Improved SSA-CNN-BiLSTM

For power systems to operate in a safe, stable, and cost-effective manner, photovoltaic (PV) power-generating accuracy is essential. A sensible mix of the most recent swarm intelligence techniques and machine learning algorithms will yield accurate prediction results. These algorithms are constantly being proposed. This paper presents the establishment of a power prediction model (SCSSA-CNN-BiLSTM) for the problem of poor accuracy and easy fall into the local minima of convolutional neural network (CNN) in the power prediction problem of photovoltaic power generation. The model is based on the optimized CNN combined with bidirectional long and short-term memory neural networks by SCSSA. Before training the data and selecting the prediction model, the dataset is preprocessed by splitting it into training and test sets. The model that produces the best optimization result is then chosen, and the predictions made by the model are compared to those made by the CNN-BiLSTM model. The experimental results demonstrate that the SCSSA-CNN-BiLSTM model outperforms the conventional CNN-BiLSTM model in short-term PV power prediction, with MAE of 0.6459, MSE of 11.5716, and 0.99867.