Research on Power Prediction of Photothermal Power Generation Systems in Remote Areas Based on a GA-LSTM Coupled Algorithm

For the problem of using a non-thermal energy storage dish-type concentrated solar power (CSP) system as an independent power supply scheme in remote areas under specific scenarios, a CSP power prediction model based on the theory of similar days and a coupled Genetic Algorithm-Long Short-Term Memory (GA-LSTM) algorithm is proposed. The model selects similar days for prediction using the GA-LSTM coupled algorithm and the gray relational degree, and establishes a GA-LSTM prediction model using similar day meteorological data and historical power generation data. Through case analysis, the GA-LSTM prediction model can effectively predict the output of the non-thermal energy storage dish-type CSP system in remote areas under specific scenarios. To analyze the reliability and accuracy of the similar day theory, compared with the LSTM control group without the similar day theory, the accuracy is improved by about 70%-75%. The case analysis shows that, compared to the unoptimized LSTM model, the GA-LSTM prediction model has increased the accuracy rate by 5.61% for clear weather types, by 4.44% for clear-to-cloudy weather types, by 18.46% for cloudy weather types, and by 21.32% for cloudy-to-clear weather types.