Wind power forecast based on broad learning system and simplified long short term memory network

Abstract

Due to its strong randomness and volatility, the modes of wind power are complex. After decomposing wind power time series into three subseries, the complex modes are deconstructed and each subseries maintain unique characteristics. Aiming at the characteristics of each subseries, a wind power forecast method based on Broad Learning System (BLS) and Simplified Long Short Term Memory (SLSTM) is proposed. Firstly, the decomposed subseries is analysed. The first layer subseries reflects the main amplitude changes caused by ramp events and the second represents the local fluctuations accompanying ramp events. Due to the randomness of ramp events, new modes appear frequently in the two subseries. The third layer subseries consists of continuous irregular jitter. Then, the BLS is introduced to forecast the first and second layer subseries. By adjusting its structure dynamically, various modes of ramp events can be learned from amplitude change and accompanying fluctuation. Additionally, according to the characteristics of the third layer subseries, continuous jitter needn't be transmitted in long‐term memory. Setting the forget gate to 0, SLSTM is proposed to forecast the third layer subseries. Finally, the wind power data from the Belgian ELIA website is used to verify the superiority of the proposed method.

Recent Advances in Fault Diagnosis Techniques for Photovoltaic Systems: A Critical Review

“…LSTM neural networks [83] Add fault identification methods. ART neural networks [84] It enhanced signal extraction capabilities.…”

“…There are many methods for feature extraction, such as using fault tree analysis to obtain the minimum cut set of the fault tree by replacing the bottom event matrix with logic gates [73], and then using wavelet features to obtain representative high-frequency signals, along with string current and open circuit voltage, as inputs to the neural network [74]. Some methods use PVA output voltage, output current, output power, inverter output voltage, output current, and light intensity as inputs [75]. The most widely used feature quantities are the maximum power point voltage The number of nodes in the hidden layer is a crucial issue, and can be determined by either of the following two formulas:…”

“…I A PVA fault diagnosis model based on an LSTM neural network is proposed in [83], which studies the influence of different fault states on the output characteristics of PVS through simulation and then obtains fault characteristic parameters. Reference [84] develops a PV fault diagnosis method based on an improved adaptive resonance theory (ART) neural network, which characterizes different fault characteristics using a matrix, allowing for the accurate diagnosis of PVS faults.…”

Adversarial false data injection attacks on deep learning‐based short‐term wind speed forecasting

“…To address the challenges posed by wind energy's variable nature, various short‐term wind speed forecasting (WSF) methods have been developed [12]. With an accurate wind speed prediction, wind farm decisions and controls can be properly formulated and implemented so that the smart grid stability can be influenced by the wind power generated as small as possible [13]. This would ensure optimal energy production and utilization, as well as grid stability and resilience, thus guaranteeing uninterrupted energy supply to consumers.…”

Wind Speed Modeling for Wind Farms Based on Deterministic Broad Learning System

“…These models can predict the development trends of systems and effectively solve problems where the operating mechanisms of complex systems are difficult to analyze [2][3][4]. Based on the nature of time series data, the requirements of the system problem, and the suitability of algorithms, common processing algorithms for time series data include univariate processing algorithms, multivariate processing algorithms [5], flat processing algorithms, linear processing algorithms, nonlinear processing algorithms, and algorithms for handling different time scales [6]. Researchers aim to develop models that accurately predict future values and thus forecast the numerical values, trends, or patterns of future time points or periods based on existing historical data [7][8][9].…”