Multi-dimensional and Multi-level Optimization Technology for Station-level Power Prediction of Mountain Wind Farms

To improve the accuracy of the wind power prediction system for mountain wind farms, this paper conducts a characteristic study on multi-dimensional indicators such as wind speed, direction, and terrain that affect wind power at multiple levels. Levels include regions, stations, sectors, and single unit. It is conducted using the historical operating data of multiple wind farms for many years. According to the spatial characteristics of mountainous areas, the corresponding relationship of regional wind resources is explored. Based on the massive historical data, the station-level domain modeling, sector optimization, power relationship optimization of single-turbine, and secondary correction of the system deviation inertia are carried out to improve the prediction accuracy of station-level wind power. Through independent innovation of multi-dimensional and multi-level power prediction optimization methods, the corresponding relationship model of provincial-level and regional wind resources is developed. Through station-level domain modeling and sector modeling, the influence of atmospheric flow field and systematic error of predicting are reduced. Various power prediction models are integrated. Deep learning modeling is proceeded to realize the integration and complementary advantages of multiple models. Through the independent development of a secondary correction model of wind power prediction, the systematic inertia error of predicting is reduced. After multi-dimensional and multi-level optimization, the prediction of the accuracy of the mountain areas reaches 87.95%, and the preparation rate increased by 3.78 percentage points.