Probabilistic Multi-Layer Perceptrons for Wind Farm Condition Monitoring

TL;DR

The paper addresses wind-farm condition monitoring by learning normal behaviour with a probabilistic multi-layer perceptron (PMLP) that outputs a predictive density $y \sim \mathcal{N}(\mu(\mathbf{x}), \sigma^2(\mathbf{x}))$ from SCADA features. It introduces transfer learning via fine-tuning (LPMLP), training on farm-wide data to improve single-turbine predictions and handle missing data, with two architecture variants for mean and variance paths. Condition monitoring is performed with CUSUM control charts on standardized residuals, enabling probabilistic fault detection and alerting. On real Kelmarsh wind-farm data, LPMLP/PMLP achieve lower RMSE/MAE and higher coverage than sparse Gaussian processes and Bayesian NNs, demonstrating practical applicability and robust uncertainty quantification for proactive maintenance.

Abstract

We provide a condition monitoring system for wind farms, based on normal behaviour modelling using a probabilistic multi-layer perceptron with transfer learning via fine-tuning. The model predicts the output power of the wind turbine under normal behaviour based on features retrieved from supervisory control and data acquisition (SCADA) systems. Its advantages are that (i) it can be trained with SCADA data of at least a few years, (ii) it can incorporate all SCADA data of all wind turbines in a wind farm as features, (iii) it assumes that the output power follows a normal density with heteroscedastic variance and (iv) it can predict the output of one wind turbine by borrowing strength from the data of all other wind turbines in a farm. Probabilistic guidelines for condition monitoring are given via a cumulative sum (CUSUM) control chart, which is specifically designed based on a real-data classification exercise and, hence, is adapted to the needs of a wind farm. We illustrate the performance of our model in a real SCADA data example which provides evidence that it outperforms other probabilistic prediction models.

Figures (9)

• Figure 1: Overall flowchart of the proposed probabilistic condition monitoring system. LPMLP: Large probabilistic multi-layer perceptron presented in Section \ref{['prob-mlp-ft']}.
• Figure 2: Architecture A1 used in the application of Section \ref{['experiments']} for both PMLP and LPMPL with shared initial layers and independent branches for mean and standard deviation.
• Figure 3: Architecture A2 used in the application of Section \ref{['experiments']} for both PMLP and LPMPL with shared initial layers.
• Figure 4: Real data wind power curves for all six wind turbines at Kelmarsh wind farm before removing standbys and warnings existing in the operational status and event file.
• Figure 5: Real data wind power curves for all six wind turbines at Kelmarsh wind farm after removing standbys and warnings using the operational status and events file provided by the data provider data.