A novel PV power prediction method with TCN-Wpsformer model considering data repair and FCM cluster

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: April 6, 2025

Abstract Short-term day-ahead photovoltaic power prediction is of great significance for system dispatch plan formulation. In this work, to improve the accuracy prediction, a TCN-Wpsformer (temporal convolutional network-window probability sparse Transformer) model based on combining data restoration and FCM (fuzzy C means) cluster proposed. The time code dataset obtained after clustering was spliced with location code. A temporal neural network introduced extract segment features incorporate self-attention mechanism. short-term outputted by window Transformer in multiple steps. Compared original model, uses It captures long-term dependencies while filtering out relatively high importance computation, which improves reduces computational cost. computing reduced 68.83% R squared improved 5.3% compared Transformer. comparison made through 11 models, above 99% different volume station data. proves that stability cross scene generalisation ability well. Meanwhile, it can also provide more accurate confidence intervals basis point has certain application value.

Language: Английский

Research on Real-Time Prediction Method of Photovoltaic Power Time Series Utilizing Improved Grey Wolf Optimization and Long Short-Term Memory Neural Network

Processes, Journal Year: 2024, Volume and Issue: 12(8), P. 1578 - 1578

Published: July 28, 2024

This paper proposes a novel method for the real-time prediction of photovoltaic (PV) power output by integrating phase space reconstruction (PSR), improved grey wolf optimization (GWO), and long short-term memory (LSTM) neural networks. The proposed consists three main steps. First, historical data are denoised features extracted using singular spectrum analysis (SSA) complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Second, (GWO) is employed to optimize key parameters (PSR) Third, predictions made LSTM networks, dynamic updates training model parameters. Experimental results demonstrate that has significant advantages in both accuracy speed. Specifically, achieves mean absolute percentage error (MAPE) 3.45%, significantly outperforming traditional machine learning models other network-based approaches. Compared seven alternative methods, our improves 15% 25% computational speed 20% 30%. Additionally, exhibits excellent stability adaptability, effectively handling nonlinear chaotic characteristics PV power.

Language: Английский