Short-term prediction of concentrating solar power based on FCM–LSTM

In China, the development of concentrated solar power has gained momentum to harness the country’s abundant solar energy resources. Predicting the short-term power generation capacity of concentrated solar power stations is crucial for mitigating the impact of the randomness and volatility of solar energy and facilitating effective grid dispatching. To solve this problem, this study presents a short-term concentrated solar power prediction combination model based on fuzzy C-means clustering. Fuzzy C-means clustering is an objective function–based fuzzy clustering algorithm that yields more flexible clustering results by incorporating fuzzy theory. Using a concentrated solar power station in Qinghai as an example, this study employs cubic spline interpolation to preprocess experimental data and divide the data into training and testing sets. Subsequently, a fuzzy c-means clustering algorithm is used to classify the preprocessed data. Different forecast scenarios are established, enhancing the precision of the prediction model. The relationship between the data is fully explored by calculating the Pearson correlation coefficient between meteorological factors and each factor in the output data under different types. Based on the degree of correlation between the factors, the input variables of different prediction submodels are determined. The influence of various meteorological factors on the prediction model under different scenarios was fully considered. Additionally, the neural network prediction model of long short-term memory in different scenarios is constructed. The test set is used to evaluate the accuracy of the combined model, and the membership degree of each sample group is determined by calculating their distance from different cluster centers to divide the test data and classify them into different scenarios. Consequently, the combined prediction model is tested. To fully confirm the feasibility and accuracy of the combined model, the test results are compared with the prediction results of the traditional long short-term memory neural network model, BP neural network model, support vector machines, and random forest. Results demonstrate that the long short-term memory neural network prediction model based on fuzzy C-means clustering has a good effect, which considerably reduces prediction error and closely aligns with actual output compared to the other two prediction models. Therefore, this model can provide a reference for power grid dispatching, effectively capturing the influence between weather factors and concentrated solar power and proving the applicability and effectiveness of the combined prediction model in different scenarios.