| SUN Dong-qin,TANG Zhan-jun,LI Ying-na,LU Peng.Prediction of the Corrosion Rate of Wind Turbine Blade Based on ISOA-KELM[J],51(11):271-278, 304 |
| Prediction of the Corrosion Rate of Wind Turbine Blade Based on ISOA-KELM |
| |
| View Full Text View/Add Comment Download reader |
| DOI：10.16490/j.cnki.issn.1001-3660.2022.11.025 |
| KeyWord:seagull optimization algorithm nuclear extreme learning machine wind turbine blade surface corrosion corrosion rate prediction |
| Author | Institution |
| SUN Dong-qin |
Kunming University of Science and Technology, College of Information Engineering and Automation Kunming |
| TANG Zhan-jun |
Kunming University of Science and Technology, College of Information Engineering and Automation Kunming |
| LI Ying-na |
Kunming University of Science and Technology, College of Information Engineering and Automation Kunming |
| LU Peng |
Yunnan Longyuan Wind Power Generation Limited Company Yunnan Qujing, |
|
| Hits: |
| Download times: |
| Abstract: |
| To scientifically stimulate the wind turbine blades maintenance plan and to protect the safety of wind farm personnel and property, the corrosion mechanism analysis of raw material for wind turbine blades was conducted. It was found that there are five main factors affecting the corrosion rate, which are temperature, external load, humidity, light, and the aging time of the material itself. Therefore, for the wind turbine blade in service, the influencing factors considered in this study are maximum temperature, average temperature, wind speed, humidity, precipitation, light intensity, and blade service time. Weekly maximum temperature, average temperature, average wind speed, average humidity, total precipitation, average light intensity, and service time of the wind turbine blades are obtained from the wind farm database and weather stations. These data are used to train the model to predict the corrosion rate of the wind turbine blades. The prediction model consists of a classifier and an optimization algorithm. A Kernel Extreme Learning Machine (KELM) was chosen as the classifier, and the hyper parameters of the KELM are optimized using an optimization algorithm to improve the classification performance. The corresponding improvement scheme is proposed to solve the problem that the SOA is easy to fall into local optimal. The method of randomly selecting the initial position of the seagull is replaced by the method of logistics chaotic mapping to improve the quality of the initial position of the seagull. The Levy flight strategy is introduced in the update method of seagull position, which makes the Seagull Optimization Algorithm have stronger global search ability. Metropolis criterion is adopted to make seagull individuals in poor positions have a certain probability to be accepted and improve the diversity of the population. The modified SOA is used to optimize the parameters of KELM, and establishes prediction model of corrosion rate on the surface of ISOA-KELM wind turbine blades. To verify the prediction performance of the ISOA-KELM model, the parameters of KELM were optimized using the basic seagull optimization algorithm (SOA), particle swarm optimization (PSO), and genetic algorithm (GA) to compare the prediction errors with SOA-KELM, PSO-KELM, and GA-KELM, respectively. The obtained data are divided into training and test sets in a ratio of 3:1, and the model is trained using the training set. The results show that optimizing KELM using ISOA improves the prediction accuracy of KELM, and the obtained values of Mean Absolute Error (MAE) of 0.457, Mean Square Error (MSE) of 0.280, and R-square of 0.959 are better than the above three comparison models. After a series of experiments, the R-square of ISOA-KELM model is higher than 0.95, which further proves that the model has good accuracy and robustness in predicting the corrosion rate of wind turbine blades. And the prediction accuracy of ISOA-KELM model is higher than the average when the weekly corrosion area is 1.5~3.2 cm2, and the corrosion rate in general is within this range, which shows that the prediction model can have good performance under normal circumstances. After obtaining the prediction model, the prediction experiment was conducted for the corrosion rate in January and February 2021. Calculate the average values of maximum temperature, temperature, wind speed, humidity, total precipitation, and light intensity in January and February of the past three years, and input the obtained average values and service time into the model to predict the corrosion rate of wind turbine blades in January and February of 2021. After 20 experiments, the average value of each index is obtained as MSE is 0.502, MAE is 0.531, R-square is 0.912. Because the influence of corrosion is the average of the past three years, so the prediction effect is not as good as the prediction effect of the model comparison for the test set, but still has high accuracy. It is proved that the model has good robustness and can provide decision suggestions for the maintenance plan of wind farms, so as to guarantee the safety of wind turbine blades. |
| Close |
|
|
|