Multi-wind Field Output Power Prediction Method based on Energy Internet and DBPSO-LSSVM

Similar documents
Wind Turbine Gearbox Temperature Monitor Based On CITDMFPA-WNN Baoyi Wanga, Wuchao Liub, Shaomin Zhangc

Journal of Engineering Science and Technology Review 7 (1) (2014)

Weighted Fuzzy Time Series Model for Load Forecasting

Dynamic Data Modeling of SCR De-NOx System Based on NARX Neural Network Wen-jie ZHAO * and Kai ZHANG

Short-Term Wind Speed Forecasting Using Regularization Extreme Learning Machine Da-cheng XING 1, Ben-shuang QIN 1,* and Cheng-gang LI 2

Fault prediction of power system distribution equipment based on support vector machine

A Hybrid Time-delay Prediction Method for Networked Control System

A Wavelet Neural Network Forecasting Model Based On ARIMA

Data Mining Based Anomaly Detection In PMU Measurements And Event Detection

An Improved Method of Power System Short Term Load Forecasting Based on Neural Network

Heat Load Forecasting of District Heating System Based on Numerical Weather Prediction Model

Experimental Study on the Cyclic Ampacity and Its Factor of 10 kv XLPE Cable

Reading, UK 1 2 Abstract

Wind Speed Forecasting in China: A Review

Electric Load Forecasting Using Wavelet Transform and Extreme Learning Machine

PREDICTING SOLAR GENERATION FROM WEATHER FORECASTS. Chenlin Wu Yuhan Lou

WIND POWER FORECASTING: A SURVEY

Application of Swarm Intelligent Algorithm Optimization Neural Network in Network Security Hui Xia1

Short-term water demand forecast based on deep neural network ABSTRACT

Wind Speed Forecasting Using Back Propagation Artificial Neural Networks in North of Iran

Research Article A PLS-Based Weighted Artificial Neural Network Approach for Alpha Radioactivity Prediction inside Contaminated Pipes

WIND POWER IS A mature renewable energy technology

Open Access Permanent Magnet Synchronous Motor Vector Control Based on Weighted Integral Gain of Sliding Mode Variable Structure

Bearing fault diagnosis based on Shannon entropy and wavelet package decomposition

Research of concrete cracking propagation based on information entropy evolution

Two-stage Pedestrian Detection Based on Multiple Features and Machine Learning

Experimental and numerical simulation studies of the squeezing dynamics of the UBVT system with a hole-plug device

Big Data Analysis in Wind Power Forecasting

A Spatial Load Forecasting Method Based on the Theory of Clustering Analysis

Information. A Fitting Approach. 1. Introduction

Open Access Combined Prediction of Wind Power with Chaotic Time Series Analysis

Application Research of ARIMA Model in Rainfall Prediction in Central Henan Province

Research on fault prediction method of power electronic circuits based on least squares support vector machine

ANN and Statistical Theory Based Forecasting and Analysis of Power System Variables

Method for Recognizing Mechanical Status of Container Crane Motor Based on SOM Neural Network

The Research of Urban Rail Transit Sectional Passenger Flow Prediction Method

About Nnergix +2, More than 2,5 GW forecasted. Forecasting in 5 countries. 4 predictive technologies. More than power facilities

WIND SPEED ESTIMATION IN SAUDI ARABIA USING THE PARTICLE SWARM OPTIMIZATION (PSO)

Feature Selection Criterion for Gravity Matching Navigation

Wind Power Forecasting using Artificial Neural Networks

Journal of South China University of Technology Natural Science Edition

The Influence Degree Determining of Disturbance Factors in Manufacturing Enterprise Scheduling

Stress Analysis of Tensioning Bolt based on MATLAB

A Fuzzy Logic Based Short Term Load Forecast for the Holidays

A Support Vector Regression Model for Forecasting Rainfall

GL Garrad Hassan Short term power forecasts for large offshore wind turbine arrays

Research Article Data-Driven Fault Diagnosis Method for Power Transformers Using Modified Kriging Model

Spinning Reserve Capacity Optimization of a Power System When Considering Wind Speed Correlation

An artificial neural network hybrid with wavelet transform for short-term wind speed forecasting: A preliminary case study

Bearing fault diagnosis based on TEO and SVM

Journal of Chemical and Pharmaceutical Research, 2014, 6(3): Research Article

Gravitational Search Algorithm with Dynamic Learning Strategy

A Hybrid Model of Wavelet and Neural Network for Short Term Load Forecasting

Construction of Emergency Services Platform for Coal Mining Accidents by Integrating Multi Source Datasets

An Improved Quantum Evolutionary Algorithm with 2-Crossovers

Research on robust control of nonlinear singular systems. XuYuting,HuZhen

This paper presents the

Modelling residual wind farm variability using HMMs

Predict Time Series with Multiple Artificial Neural Networks

Power Load Forecasting based on Multi-task Gaussian Process

A Sliding Mode Control based on Nonlinear Disturbance Observer for the Mobile Manipulator

Probabilistic Energy Forecasting

RBF Neural Network Combined with Knowledge Mining Based on Environment Simulation Applied for Photovoltaic Generation Forecasting

Open Access Research on Data Processing Method of High Altitude Meteorological Parameters Based on Neural Network

Analysis on the Operating Characteristic of UHVDC New Hierarchical Connection Mode to AC System

A New Robust Decentralized Control Method for Interconnected Nonlinear Systems Based on State Extension and Adaptive Tracking

Prediction of gas emission quantity using artificial neural networks

ARX System identification of a positioning stage based on theoretical modeling and an ARX model WANG Jing-shu 1 GUO Jie 2 ZHU Chang-an 2

Experimental Research of Optimal Die-forging Technological Schemes Based on Orthogonal Plan

Condition Parameter Modeling for Anomaly Detection in Wind Turbines

Keywords:radar cross section (RCS), statistical model, fitting technique, test hypothesis.

No. 6 Determining the input dimension of a To model a nonlinear time series with the widely used feed-forward neural network means to fit the a

Soft Sensor Modelling based on Just-in-Time Learning and Bagging-PLS for Fermentation Processes

Predicting Roadway Surrounding Rock Deformation and Plastic Zone. Depth Using BP Neural Network

Bayesian Network-Based Road Traffic Accident Causality Analysis

Research Article Finite Element Analysis of Flat Spiral Spring on Mechanical Elastic Energy Storage Technology

Research on State-of-Charge (SOC) estimation using current integration based on temperature compensation

1983. Gearbox fault diagnosis based on local mean decomposition, permutation entropy and extreme learning machine

Controlled Quantum Teleportation via Four Particle Asymmetric Entangled State *

Study on Coal Methane Adsorption Behavior Under Variation Temperature and Pressure-Taking Xia-Yu-Kou Coal for Example

Analysis on Climate Change of Guangzhou in Nearly 65 Years

Long Term Load Forecasting Using SA-ANN Model: a Comparative Analysis on Real Case Khorasan Regional Load

Key words: insulator, the equivalent diameter, natural icing, icing thickness, icing degree, characterization method

Study on form distribution of soil iron in western Jilin and its correlation with soil properties

Numerical Interpolation of Aircraft Aerodynamic Data through Regression Approach on MLP Network

A research on the reservoir prediction methods based on several kinds of seismic attributes analysis

Integrating Geographic Information System and Building Information Model for Real Estate Valuation

RESEARCH ON AEROCRAFT ATTITUDE TESTING TECHNOLOGY BASED ON THE BP ANN

APPLICATION OF RADIAL BASIS FUNCTION NEURAL NETWORK, TO ESTIMATE THE STATE OF HEALTH FOR LFP BATTERY

A Fast Method for Embattling Optimization of Ground-Based Radar Surveillance Network

Distributed Clustering and Local Regression for Knowledge Discovery in Multiple Spatial Databases

A new method for short-term load forecasting based on chaotic time series and neural network

THE DESIGN AND IMPLEMENTATION OF A WEB SERVICES-BASED APPLICATION FRAMEWORK FOR SEA SURFACE TEMPERATURE INFORMATION

One-Hour-Ahead Load Forecasting Using Neural Network

A Novel Analytical Expressions Model for Corona Currents Based on Curve Fitting Method Using Artificial Neural Network

The Fault extent recognition method of rolling bearing based on orthogonal matching pursuit and Lempel-Ziv complexity

Grid component outage probability model considering weather and. aging factors

WEATHER DEPENENT ELECTRICITY MARKET FORECASTING WITH NEURAL NETWORKS, WAVELET AND DATA MINING TECHNIQUES. Z.Y. Dong X. Li Z. Xu K. L.

Forecast daily indices of solar activity, F10.7, using support vector regression method

In-Situ Fabrication of CoS and NiS Nanomaterials Anchored on. Reduced Graphene Oxide for Reversible Lithium Storage

Transcription:

, pp.128-133 http://dx.doi.org/1.14257/astl.16.138.27 Multi-wind Field Output Power Prediction Method based on Energy Internet and DBPSO-LSSVM *Jianlou Lou 1, Hui Cao 1, Bin Song 2, Jizhe Xiao 1 1 School of Information Engineering, Northeast DianLi University, Jilin 1312, China 2 State Grid Jilin Electric Power Co.Ltd., Jilin power Supply Company Abstract: Traditional wind power prediction method only applicable to a single wind farm. In view of this isolated prediction model. In this paper, we propose a multi-wind farm prediction method based on DBPSO-LSSVM and energy Internet. Firstly, based on DBSCAN algorithm to detect outliers and select multi-wind field training samples. And to search the optimal input parameters of LSSVM based on particle swarm algorithm. Secondly, to predict multiple wind field s power combined with numerical weather prediction system. The method we proposed can be used to make the scheduling plan in advance to solve a large number of abandoned wind power rationing problem every year. In experiment, the method has lowest error rate compares to LSSVM and BP neural network and it is more suitable to predict wind fields in different areas. Keywords: Multiple wind farms power prediction; Energy Internet; Least square support vector machine; particle swarm; wind power utilization 1 Introduction Wind energy as a renewable energy has been paid more and more attention. However, wind power itself has inherent shortcomings such as intermittent and randomness. The prediction model is isolated and the capacity of wind power utilization is limited in some areas. So every year there are a serious of problems of abandoned wind power rationing. This waste a lot of wind power resource. At present, the following methods are proposed to predict the wind farm power: In paper [1], predict wind farm power based on component Bayesian method. In paper [2], predict wind farm power based on error stack correction. In paper [3], predict wind farm power based on Particle Swarm Optimization for nuclear extreme learning machine model. In paper [4], predict wind farm power based on statistical method. In paper [5], predict wind farm power based on Markoff chain principle and Correlation vector. In paper [6], propose to use (OS-ELS) learning machine to modify the wind speed and achieve the short-term wind prediction. In paper [7], propose to use Tuple vector time warping to predict wind farm power. However, the current method can only predict a single wind farm.with the rapid development of energy Internet, the efficient and fast data interconnection, information sharing platform has a great impact on the traditional isolated operation mode [8],[9]. In this paper, we propose a multiple wind farms prediction method. ISSN: 2287-1233 ASTL Copyright 16 SERSC

Firstly, we introduce some relate works. Secondly, we introduce the prediction process. Finally, the effectiveness of method is proved by experiment. 2 Relate Work 2.1 Relate Technologies and Characteristics based on Energy Internet Firstly, the method we propose based on the characteristics such as renewable energy sources, information openness and sharing, advanced information processing technology, energy free transmission and advanced storage technology of Energy Internet to predict multiple wind fields. In order to make wind power scheduling plan in advance, to resolve the wind power utilization limited [1-11]. 2.2 DBSACN Outlier Detection Theory DBSCAN algorithm is a well-known clustering algorithm that according to two parameters Eps and MinPts to identify a cluster [12]. Due to the noise of real data is very large, it will affect the accuracy of the prediction model. So we remove outliers based on DBSCAN algorithm. The steps as follows: (1) Remove the point of less than or equal to zero to exclude human factors; (2) Set the parameters Eps and MinPts of DBSCAN algorithms and to cluster; (3) Among the clusters output, the number of data points in the Eps radius is larger than the MinPts labeled as the training sample data, the others are labeled as outliers; (4) Remove outliers and the remaining data are the training sample data. 2.3 PSO-LSSVM Regression Theory LS-SVM is an improvement of SVM and maps the training sample data to high dimensional feature space by a nonlinear mapping[13].and then, based on Particle Swarm Optimization(PSO)[14] to select parameters γ and σ2 of LSSVM, Specific steps are as follows: (1) Initial accelerate constant, inertia weight, population size,maximum number of evolution, parameters γ and σ2 and the parameters are mapped into random particles. (2) According to current position to calculate optimal position of current point and group.update the speed and position of each particle, and generate new population. (3) Calculate new location and adaptive value of each particle, compare its best position and optimal location in history. If better, replace it, otherwise, not change. (4) Check whether to meet the end of the optimization conditions, if satisfy, output the best nuclear parameters of LS-SVM, otherwise, turn to step (2). Copyright 16 SERSC 129

3 Multi-wind Farm Output Power Prediction Method based on Energy Internet and DBPSO-LSSVM 3.1 Selection and Normalization of Input Parameters The wind speed, air density and wind direction are parameters which have great influence on output power of wind turbine [15], [16]. So the data corresponding to the three attributes and power attribute are used as the initial input data in this paper. In order to eliminate the influence of the volume, the input data normalized to the [, 1] range, the normalization formula is as follows: x * x (1) In formula: x* is normalized initial training sample data; u is mean of all sample data; σ is Standard deviation of all sample data. 3.2 Establishment of Prediction Model based on Energy Internet and DBPSO-LSSVM The specific steps of DBPSO-LSSVM are as follows: (1) Collect multiple wind field SCADA data sets under energy Internet environment based on the features in 2.1 chart. And Normalization. (2) Multiple wind field initial sample data sets are input DBSCAN algorithm. Based on the 2.2 step to detect outliers, and select the training sample data sets. (3) To establish PSO-LSSVM models based on 2.3 step. (4) At last, multiple wind farm prediction models are established and combined with NWP system to predict. Wind field 1 data Wind field 2 data Wind field n data Wind field 1 data Wind field 2 data Wind field n data Initialize particle swarm Based on DBSCAN clustering Adaptive weight Energy Internet Based on DBPSO-LSSVM regression prediction Greater than the parameter Minpts Y Training sample N outliers Update speed and position Meet the end conditions Y Output optimized LSSVM parameters N NWP System Wind field 1 model Wind field 2 model Wind field n model Wind field 1 training samples Wind field 2 training samples Initialize LSSVM parameters Wind field n training samples Re training Scheduling personnel Predictive value 1 Predictive value 2 Predictive value n Wind field 1 prediction model Wind field 2 prediction model Wind field n prediction model Fig.1. Integrated prediction From figure 1, the method we propose to share the multiple wind farms SCADA data in different regions under the environment of energy Internet. And when the SCADA data of n wind fields are input, the n regression models will be output. This 13 Copyright 16 SERSC

is due to the method will be selected different training samples and different optimal input parameters among different wind fields. At last, it will be output different prediction models aim at different data distribution of wind farms. 4 Example analysis In experiment, we predict two wind farms in North and South China based on the method we propose with MATLAB simulation technology. The value of wind speed, wind direction and air density are as the input data based on numerical weather prediction system to provide. The output data are predict values each wind field. 4.1 Comparison of Prediction Results of Wind Power Based on the method we propose to predict two wind farms power compare to BP neural network, LSSVM and actual output power, the result as shown in figure 2: 9 Actual power BP-neural network DBPSO-LSSVM 9 Actual power LSSVM DBPSO-LSSVM 7 7 5 3 5 3 1 1 1 Actual power BP-neural network DBPSO-LSSVM 1 Actual power LSSVM DBPSO-LSSVM 1 1 1 1 Fig.2. Comparison of DBPSO-LSSVM,LSSVM,BP neural network, actual output power From figure 2, the method we propose has the highest prediction accuracy and the predict values close to actual power mostly between two wind farms in different areas. Remove outliers of SCADA data and select the optimal input parameters can improve the precision of forecasting when establish model based on LSSVM. 4.2 Comparison of Output Power Prediction Error In this paper, the mean square error (MSE) and the mean absolute percentage error (MAPE) are used to evaluate the prediction results. Copyright 16 SERSC 131

n 2 Pt Pt 1 E MSE (2) n t1 n 1 Pt Pt E MAPE (3) n P t1 t In formula: P is the real value of the forecast; P^ is predictive value of prediction; n is data quantity. Table 1. Main error indicators of the prediction models of A and B wind fields Algorithm type(a) MSE MAPE Algorithm type(b) MSE MAPE BP-neural network 7.24.1724 BP-neural network 4.98.1582 LSSVM 2..1396 LSSVM 1.93.1297 DBPSO-LSSVM.4151.79 DBPSO-LSSVM.3782.694 From Table 1: through the MSE and the MAPE two common error indicators calculated, the method we propose in this paper has the smallest error index. And the fluctuation range of error index between two wind fields is the smallest of all. So the proposed method is stable to different data distribution and more suitable than the other two methods to predict the output power of different wind fields. 5 Conclusion In this paper, based on the information sharing and interconnection of energy Internet, we propose a multiple wind fields prediction method based on energy Internet and DBPSO-LSSVM. The method we propose takes account into data distribution of wind farm in different areas. And to establish the best forecasting models for different wind fields. It has the smallest error compared to LSSVM, BP neural network. The fluctuation range of MSE and MAPE between two wind fields are.4 and.7 up and down. It is more stable. So the method provides a basic research for future to realize multiple wind farms prediction under the energy Internet environment. References 1. Ming, Y., Shu, F., Xueshan, H.: Probabilistic forecasting method for output power of wind farm based on component sparse Bayesian learning [J]. Automation of electric power system, 12, 36(14): 125-13. 2. Mao, M., Cao, Y., Zhou, S.: Improved short term wind power prediction method based on error stack correction [J]. Automation of electric power system, 13, 37(23): 34G38. 132 Copyright 16 SERSC

3. Yang, X., Guan, W., Liu, Y.: Wind power interval prediction method based on Particle Swarm Optimization for nuclear extreme learning machine model [J]. Chinese Journal of Electrical Engineering, 15, 1. 4. Chen, Y., Sun, R., Wu, Z.: Power prediction of regional wind farm group based on statistical method [J]. Automation of electric power system, 13, 37(7): 1-5. 5. Yang, X., Wang, B., Lan, H.: Short term power prediction of wind farm [J].Electric power system and its automation, 15, 27(9): 85-9. 6. Wang, Y., Wang, Z., Huang, M.: Ultra short term wind power prediction based on OS-ELM and Bootstrap method [J]. Automation of electric power system, 14, 6: 4. 7. Liu, Y., Liu, C., Li, W.: Multi time scale power prediction of wind power generation based on the matching of output mode and power mode matching for wind power generation [J]. Chinese Journal of Electrical Engineering, 14, 34(25): 435-4358. 8. Dong, Z., Zhao, J., Wen, F.: From smart grid to energy Internet: basic concepts and research framework [J].Automation of electric power system, 14, 38(15): 1-11. 9. Sun, H., Guo, Q., Pan, Z.: Energy Internet: philosophy, architecture and frontier outlook [J]. Automation of electric power system, 15, 19: 1. 1. Yan, T., Cheng, H., Zeng, P.: Energy Internet architecture and key technologies [J]. Power grid technology, 16, (1): 15-113. 11. Zeng, M., Yang, Y., Liu, D.: Coordinated optimization operation mode and key technology of "source network, load and storage" of energy Internet [J]. Power grid technology, 16, (1): 114-124. 12. Ester, M., Kriegel, H. P., Sander, J.: A density-based algorithm for discovering clusters in large spatial databases with noise[c]//kdd. 1996, 96(34): 226-231. 13. Gu, Y., Zhao, W., Wu, Z.: Research on robust regression algorithm based on least squares support vector machine [J].Journal of Tsinghua University: Natural Science Edition, 15 (4): 396-2. 14. Liu, Z., Zhang, W., Wang, Z.: Optimization layout of charging stations of urban electric vehicles based on quantum particle swarm optimization algorithm [J]. Chinese Journal of Electrical Engineering, 12, 32(22): 39-45. 15. Zhang, Z., Zhou, Q., Kusiak, A.: Optimization of wind power and its variability with a computational intelligence approach [J]. Sustainable Energy, IEEE Transactions on, 14, 5(1): 228-236. 16. Fan, G., Wang, W., Liu, C.: Short term wind power forecasting system based on artificial neural network [J].Power grid technology, 8, 32(22): 72-76. Copyright 16 SERSC 133