Wind Power Prediction Based On Improved Density Clustering And Hybrid Models

Introduction: In recent years, with the increasing installed capacity of wind turbines, wind power has become one of the primary means of new energy generation. This article focuses on improving the accuracy of short-term power forecasting for wind farms. Methods: A wind farm power prediction model based on the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering model of the Snow Ablation Optimizer (SAO) algorithm and ARIMA-LSTM hybrid model is proposed. Firstly, SDBSCAN clustering is utilized to classify large wind farms into several sub-clusters. On this basis, the data of each sub-cluster is first input into the Autoregressive Integrated Moving Average Model (ARIMA) prediction model to get the preliminary prediction results and calculate the residual values. The residual values were input into the Long Short-Term Memory (LSTM) prediction model to obtain the secondary prediction results. Superimpose the two model prediction results to get the final prediction results. Results: After analyzing the results, it was found that all the wind farm power prediction results after clustering are better than those without using the clustering model. The ARIMA-LSTM hybrid prediction models used are all better than the single ARIMA and LSTM models. Discussion: This paper addresses the complexity of wind power data, which is influenced by the different output characteristics of other units in wind farms. SDBSCAN clustering and a hybrid prediction model are used for wind power prediction methods, respectively. It improves the accuracy of wind power prediction and reduces energy loss in real-world engineering applications. However, there is still the problem of data dimension limitation, and future research can introduce more meteorological data to improve the prediction ability of complex weather. Conclusion: Experimentally validated effectiveness of the proposed method for wind power prediction.