Short-term power forecasting of photovoltaic generation based on CFOA-CNN-BiLSTM-Attention

Under the goal of 'double carbon', the penetration of photovoltaic (PV) power generation in the power system is increasing, and in view of the strong volatility and high stochasticity of PV power output, reliable PV power prediction can provide a reference for the development of scheduling plans and improve the stability and reliability of power grid operation. Traditional deep neural networks are prone to problems such as local optimality, slow convergence speed, and poor prediction results due to insufficient feature extraction capability. In order to improve the prediction accuracy, a deep neural network photovoltaic power generation short-term prediction model integrating the capture optimisation algorithm (CFOA), convolutional neural network (CNN), bidirectional long and short-term memory network (BiLSTM), and attention mechanism (AM) is proposed. Firstly, the spatial features of the data are extracted using the CNN method and input to the next layer, and the temporal features implicit in the spatial feature information are extracted using the BiLSTM method and the extracted spatial and temporal features are input to the next layer; then, the self-attention mechanism is incorporated to define the relative importance in order to capture the long-term dependency relationship between each of the input elements, and the weights of extracted input features are automatically assigned. After that, the CFOA optimisation algorithm is introduced for model hyper-parameter optimisation, and the prediction model is built to obtain the predicted values of PV power generation; finally, the model is validated using actual data from a PV power station. The results show that the proposed combined prediction method has better prediction stability and accuracy in short-term PV power prediction.