Source-linked AI summary
Wind speed prediction using a hybrid model of the multi-layer perceptron and whale optimization algorithm
Saeed Samadianfard, Sajjad Hashemi, Katayoun Kargar, Mojtaba Izadyar, Ali Mostafaeipour, Amir Mosavi, Narjes Nabipour, Shahaboddin Shamshirband
TL;DR
Accurate wind-speed prediction is needed to support renewable wind-power use, but variability makes forecasting difficult and traditional approaches can perform poorly. The study combines a multi-layer perceptron with the Whale Optimization Algorithm and evaluates it across ten stations in northern Iran using reference-station data. The hybrid model outperforms standalone MLP models across the stations, with lower RMSE, SI, and RE and higher NSE, WI, and KGE values.
Problem
Wind-speed prediction is difficult because wind has high stochastic variability, while some existing approaches have short-term limitations and costly, intricate calculations.
Method
The study develops an MLP-WOA model that uses reference-station data, without atmospheric datasets, to predict wind speed at ten target stations in northern Iran.
Results
The MLP-WOA model outperforms standalone MLP models across all ten stations, producing lower RMSE, SI, and RE and higher NSE, WI, and KGE values.
Takeaways & Limitations
The study concludes that WOA can improve MLP prediction accuracy and may support accurate wind-speed forecasting.
Takeaways & Limitations
The study’s physical-model comparison notes that such approaches involve intricate calculations, high costs, and poor short-term performance, despite potentially better long-term accuracy.
Abstract
from arXiv · showhide
Wind power as a renewable source of energy, has numerous economic, environmental and social benefits. In order to enhance and control renewable wind power, it is vital to utilize models that predict wind speed with high accuracy. Due to neglecting of requirement and significance of data preprocessing and disregarding the inadequacy of using a single predicting model, many traditional models have poor performance in wind speed prediction. In the current study, for predicting wind speed at target stations in the north of Iran, the combination of a multi-layer perceptron model (MLP) with the Whale Optimization Algorithm (WOA) used to build new method (MLP-WOA) with a limited set of data (2004-2014). Then, the MLP-WOA model was utilized at each of the ten target stations, with the nine stations for training and tenth station for testing (namely: Astara, Bandar-E-Anzali, Rasht, Manjil, Jirandeh, Talesh, Kiyashahr, Lahijan, Masuleh, and Deylaman) to increase the accuracy of the subsequent hybrid model. The capability of the hybrid model in wind speed forecasting at each target station was compared with the MLP model without the WOA optimizer. To determine definite results, numerous statistical performances were utilized. For all ten target stations, the MLP-WOA model had precise outcomes than the standalone MLP model. The hybrid model had acceptable performances with lower amounts of the RMSE, SI and RE parameters and higher values of NSE, WI, and KGE parameters. It was concluded that the WOA optimization algorithm can improve the prediction accuracy of MLP model and may be recommended for accurate wind speed prediction.
1. Introduction
Wind-speed prediction is important for using renewable wind energy, but its stochastic variability and local environmental influences make accurate forecasting difficult. The paper develops an MLP-WOA hybrid that uses reference-station data and WOA optimization to predict wind speed across ten stations in northern Iran.
- Motivation: Wind energy offers renewable benefits, but its irregular nature complicates reliable electricity supply and motivates accurate wind-speed prediction.Local climate, topography, and obstacles such as buildings also affect wind energy.
- Existing approaches: Wind-power prediction depends on wind-speed estimation, with physical, statistical, and intelligence-learning models used for forecasting.Physical approaches use atmospheric and meteorological information, whereas statistical and learning approaches are commonly applied to short-term prediction.
- Existing approaches: Physical approaches involve intricate calculations and high costs and have shown poor short-term performance, although they can be more accurate for long-term prediction.The stated comparison is specific to short-term versus long-term forecasting horizons.
- Existing approaches: Prior studies report that hybrid models can outperform corresponding single models by combining statistical, intelligent, filtering, or decomposition methods.Examples include KF-ANN, ANN-ARIMA, and decomposition-based combinations with ANFIS or MLP.
- Study contribution: The study develops an MLP-WOA hybrid for wind-speed prediction without atmospheric datasets, using reference-station data to estimate target-station values.WOA is used to improve the model’s precision.
- Study contribution: The proposed model is investigated across ten spatially scattered stations in northern Iran using reference-station data.The study specifically examines wind-speed forecasting for the stated station network.
2. Methods and materials
The study uses a multilayer perceptron to model nonlinear wind-speed relationships, with nine neighboring stations as inputs and the target station as output. Whale Optimization Algorithm mechanisms optimize the model, and multiple statistical criteria assess prediction accuracy.
- The multilayer perceptron models nonlinear prediction problems by learning implicit relationships from training data and calculating outputs from those relationships.
- The network uses input, hidden, and output layers, with nine neighboring stations providing inputs and the target station providing the output.
- The optimized network uses 9 input neurons, 8 hidden neurons, and an output layer, with sigmoid tangent and linear functions trained using the Levenberg–Marquardt algorithm.
- WOA searches for improved solutions through exploitation and exploration, including encircling, spiral-shaped movement, and random prey selection.
- WOA updates candidate locations relative to the best solution using coefficient vectors, random factors, shrinking encirclement, and spiral movement.
- The study evaluates prediction accuracy using R2, RMSE, RE, WI, SI, NSE, and KGE.
O P NSE
The study uses wind-speed observations from ten stations in Gilan province to develop and evaluate prediction models. Stations serve alternately as targets, with the remaining nine used as reference stations for MLP and MLP-WOA estimation.
- Study data: Monthly mean wind-speed data were collected for ten Gilan stations over 2004–2014.The stations include Astara, Bandar-E-Anzali, Rasht, Manjil, Jirandeh, Talesh, Kiyashahr, Lahijan, Masuleh, and Deylaman.
- Study area: Station elevations range from −23.6 to 1581.4 m above sea level, while locations span 36º42′–38º21′ N and 48º51′–50º00′ E.
- Wind-speed characteristics: Lahijan has the lowest mean wind speed at approximately 1.46 ms-1, whereas Jirandeh has the highest at 5.25 ms-1.
- Wind-speed characteristics: Jirandeh records the maximum wind speed among the reported stations, reaching 25.6 ms-1.
- Model development: The study reports pairwise wind-speed correlation coefficients among the studied stations and uses them to characterize relationships between target and reference stations.
- Model development: Each target station is predicted using MLP and MLP-WOA models, with the other nine stations designated as reference stations.
3. Result and discussion
The study evaluates MLP and WOA-optimized MLP models for wind-speed prediction using data from neighboring stations in northern Iran. Across the ten stations, MLP-WOA generally outperformed standalone MLP on the reported statistical measures.
- Study design: The study compares MLP with WOA-optimized MLP models for wind-speed prediction using data from nine neighboring sites in northern Iran.The target station is estimated with both models, using the remaining stations as reference data.
- Model development: 70% of the 3,611 observations was used for training and 30% for testing, with random resampling repeated across model-selection runs.Model conditions were selected using R2 and RMSE before applying the WOA process.
- Comparative results: WOA-optimized models had lower RMSE and SI than standalone MLP models during training and better testing performance across the stations.Standalone MLP RMSE ranged from 0.57 to 1.18, while MLP-WOA RMSE ranged from 0.52 to 1.09.
- Station-level results: Kiyashahr and Rasht were the most accurate stations for both model types, while Manjil and Jirandeh had higher RMSE values.For classical MLP, Kiyashahr and Rasht had RMSE values of 0.57 and 0.62; for WOA-MLP, they had 0.54 and 0.52.
- Statistical assessment: WOA-optimized models also showed higher WI, NS, and KGE and lower RE than classical MLP models.The reported comparisons include both error-based and agreement-based statistical parameters.
4. Conclusion
The study presents MLP-WOA for forecasting daily wind speed at ten locations using neighboring stations’ data. WOA optimization improved the MLP’s wind-speed prediction performance, while atmospheric information was not used.
- WOA addresses the challenge of selecting suitable weights in artificial-intelligence algorithm layers.
- The study forecasts wind speed at each target location using input data from neighboring reference locations.
- MLP-WOA combines the Whale Optimization Algorithm with standalone MLP models to predict daily wind speed at each of ten target stations.
- The method does not use atmospheric information for wind-speed prediction.
- Multiple statistical indices were used to evaluate the MLP-WOA model’s performance.
- The hybrid MLP-WOA model achieved high wind-speed estimation accuracy at almost all stations.