Online Condition Assessment of Wind Turbine Based On Integrated Fuzzy Theory and Evidence Theory GUO Peng*, CHI Bing † *North China Electric Power University, Bei Jing,China,102206 1572647494@qq.com Keywords: wind turbines; condition assessment; evidence theory; fuzzy theory. Abstract For the purpose of optimizing the maintenance strategy, the wind turbine real-time condition needs to be assessed accurately. Based on SCADA system data, this paper put forward the wind turbine running condition assessment model integrating the Fuzzy Theory with Evidential Theory. Firstly, the index system of wind turbines running condition assessment was constructed in this paper. The active curve of temperature indicators was obtained under bin method, and then the indicators were quantified under degradation degree concept; Secondly, the wind turbine running condition assessment model integrating the fuzzy theory with evidential theory was proposed as a resolution to the fuzzy uncertainty of wind turbines condition assessment indicators. In this model, concept of fuzzy sets and membership functions of the fuzzy theory were imported into the evidence theory to build reasonable belief function assignment. Finally, the proposed assessment model and method were proved to be correct and effective by the practical operational results based on the actual monitoring data of a 1.5MW WTGS within a certain period of time. 1 Introduction As a typical representative of new energy sources, wind power is getting people's attention with the intensification of energy crisis and the upgrading of human environmental awareness. However, the operating environment of wind turbine is harsh. The wind velocity is not controllable and the external environment of wind turbine has a large temperature difference. That harsh environment leads to the costs of operation and maintenance of the wind turbine up to 10% to 15%[1]. Therefore, wind turbine failure could be avoided effectively by monitoring and evaluating the unit timely and accurately. In the meantime, it provides a reliable basis for the state maintenance. It has an important practical significance for wind turbine operation. Large-scale grid-connected wind turbines start relatively late in China. Test Data of wind turbine unit is relative lack, such as the pre-test data during manufacturing or operation, etc. In recent years, most studies focus on an additional sensors or special online monitoring device. Assess the state of the subsystem of wind turbine such as generator, gearbox, pitch system and drive chain[2-4]. However, this state assessment method is only directed against a single component. It does not consider the coupling and interaction between the various components or subsystems so it is not possible to assess the operational status of the unit from the overall perspective [5]. Therefore, it is necessary to find an effective way to do the wind turbine operating state assessment by using the SCADA system data instead of unduly relying on other data. At present, there have been some research developed several wind turbine state assessment models based on neural networks, probability and statistics, matter element analysis or fuzzy comprehensive evaluation, etc. The basic idea is divided into two ways. One way is to extract fault characteristics by digging the related fault information from SCADA historical data[6,7]. The second way is to establish a healthy running wind turbine unit model by handling historical SCADA system data. The former way requires a great amount of fault data supportive so it is hard to achieve for those units whose running time is short. The latter way could avoid the above shortcomings. But if using intelligent algorithms such as neural networks, it is short for training samples as the lacking of studies on the state assessment of the overall operation wind turbine in the existing literature. Based on the above characteristics, this paper presents a wind turbine state assessment model based on fuzzy theory and evidential theory. 2 The index system established based on SCADA data Wind Turbine Generator System (WTGS) is a complex system. It can be divided into many subsystems and components in terms of different function and structure. Then, evaluation index system can be established for each part. It is shown in Figure 1. 3 Calculate of the degradation degree 3.1 Monitoring data processing The indicators Xijk in FIG.1 are divided into two categories. The first one is the type of the smaller the better indicators, taking temperature as a representative, which is expressed by g(1)Xijk. The second one is the type of intermediate indicators. It is other indicators except temperature, which is expressed with g(2)Xijk . It should be noted that the index of environment temperature is included in the second category indicators, and the index of active power is included in the first category indicators. Grid frequency Grid factors Reactive power Active power Fig 3 Curve of gearbox oil temperature Phase current Environmental factors Temp. of environment Control cabinet Temp. of IGBT Wind speed Temp. of control cabinet Therefore, the polynomial function of wind speed vs gearbox bearings temperature can be described with the following expression: 54. 6, 3 5 2 4 3 1. 6 10 6. 78 10 1. 0873 ( 1) ' 2 gXi j ( ) 8. 1015 25. 7974 83. 7263, 2. 9 103 3 0. 1438 2 2. 2891 75. 8789, 3 3 10 ( 1) 10 Where, g ( ) is the temperature of calculated gearbox bearing; ω is the wind speed. ( 1) ' Xi j Cabin vibration acceleration Cabin Framework for condition assessment of WTG Phase voltage Cabin location Gearbox Temp. of gearbox oil Temp. of gearbox bearing2 Temp. of gearbox bearing1 Generator Temp. of generator winding Generator rotational speed Temp. of generator bearing Figure1 Wind turbine index system of operation data Bin method is used for reference to draw the curve of indicators which belongs to the first kind. Then each index’s fitting function changing with wind speed is obtained by polynomial curve fitting. It is illustrated by the case of gearbox oil temperature and gearbox bearing temperature in Figure 2, 3. 3.2 Calculate of the degradation degree The temperature deviation between actual monitoring value and normal temperature can be used as basis for judging whether failure in one certain wind speed. The deviation degree can be expressed by formula (2): gX ij gX gX ij ' ( 2) ij The concept of degradation degree is introduced to characterize the relative degradation between normal running state and fault condition for a wind turbine. Its range is from 0 to 1. Different state evaluation indicators values reflect the degree of deterioration. The closer to 1 means the more serious degradation. The degradation degree of assessment indicators for the first category is represented by the formula (3): 0 g(X1) ( 1) ij m( gX ) ij 1 Where, m ( gX ( 1) ij gX ij gX ij gX ij ( 3) ) is the degradation degree of indicators which belongs to the first category, αis the ideal value of the indicator, β is the limit value of the index. The indicators of the second category, taking engine room location, environment temperature, engine vibration acceleration as example, is calculated by formula (4) as follows: Fig 2 Curve of gearbox bearing temperature 1 gX 1 ijk g( 2) Xi j 2 1 g Xi j 2 2 1 m( g(X2) ) 0 2 g Xi j 2 ( 4) ij ( 2) g Xi j 2 2 g x 1 ij 1 2 1 gX 1 ij Where, mg ( (X2) ) is the degradation degree of indicators which ij belong to the second category, α1 is the upper limit value of the index, β1 is the lower limit value of the index; α2 , β2 are permitted value under normal working conditions. They’re determined by wind farm’s inspection records, fault manual. 4 WTGS online assessment model based on Fuzzy Theory and Evidence Theory The assessment indicators are divided into six sub-body of evidence according to the indicators layer established in Section 1. They are generator (sub-evidence1), gearbox (subevidence2), cabin (sub-evidence3), control cabinet (subevidence4), power factor (sub- evidence5) and environmental factors (sub-evidence6). Each sub-body of evidence is constituted by a number of evaluation indicators (FIG. 1). The steps of establishing WTGS online assessment model based on Fuzzy Theory and Evidence Theory are as follows: 1) to get the actual value of each evaluating index and calculate the degree of degradation; 2) to calculate the membership of each level status for sub-body of evidence; 3) Evidence Theory is used to testify each son-body of evidence. Then an overall assessment of the wind turbine is obtained. The flow chart of WTGS Assessment Model is shown in Figure 4. g>0.9 Y ridge is selected in this paper to determine the level of evaluation indicators. The membership function is defined as follows: 1 a a2 1 1 v 1( g ) si n (g 1 ) 2 a2 a1 2 2 0 1 a a2 1 (g 1 ) si n 2 2 a2 a1 2 v 2( g ) 1 a a4 1 1 si n (g 3 ) 2 2 a a 2 3 4 1 1 si n 2 a4 2 v 3( g ) 1 1 si n 2 2 a6 g a1 g a2 a1 g a2 a1 g a2 a2 g a3 a3 g a4 a3 g a4 1 a4 g a5 a a6 (g 5 ) a5 2 a5 g a6 a3 (g a3 a4 ) 2 0 a a6 1 1 v 4( g ) si n (g 5 ) 2 2 a a 2 5 6 1 ( 5) ( 6) ( 7) g a5 g a6 a5 g ( 8) a6 In formula (5)-(8), g is the shorthand of degradation degree, v1 (g) ~ v4 (g) are membership of evaluation index X r for levels of l1 ~l4 , a1 ~a 6 are the boundary values between the different status levels. The level of wind turbines operation is represented as L={l1 l2 l3 l4 }={ Good Qualified Attention Serious}. Since there is no a priori knowledge, so the membership degree boundary for different states adopt differentiation processing method in this article. The range of state l1 ~l4 are from 0~0.3, 0.1~0.6, 0.4~0.9, 0.7~1, so a1~a6 take values of 1/10, 3/10, 4/10, 6/10, 7/10 and 9/10, as shown in Fig5. 1 0.5 Evaluation Result is “Serious” N Monito ring Data Deterioratio n Degree Weight Each Index Membership Confidence Coefficient Membership of Each Evidence Corre cted Credit Assign ment Values Corre cted Credit Assign ment Values 0 Assess ment Results Fig 4 Flow chart of WTGS Assessment Model 4.1 Membership degree of indicators In this paper, Fuzzy Evaluation Method is used to determine the status of each index. Compared with other method, Fuzzy Evaluation Method can deal with state evaluation of boundary treatment hardening too much. Fuzzy Evaluation Method is used to overcome the problem that. According to the theory of fuzzy mathematics, the information of different status levels can be described by the membership functions. The membership function which combined half ladder with half 0 0.2 0.4 0.6 0.8 1 Fig 5 Membership degree of valuation indicators 4.2 Membership degree of sub-evidences By formula (5) ~ (8), it can determined evaluation index Xr for the membership of four different state levels v1 (g) ~ v4 (g). The mathematical expression of sub-evidence’s membership is shown as follows: v i ( Ej ) Where, v i ( Ej m v ( g r 1 i r )( x r ) ( 9) ) is membership degree of state level i for sub-evidence j ; v i ( g r ) is membership degree of state level i about assessment indicator for sub-evidence j ; ( x r ) is the weight of evaluation indicators 𝑋𝑟 . The Experts-decision Method is used to determine the weight of indicators. WTGS states are divided into four levels. Therefore, the value of i is from 1 to 4. In the same way, there are six sub-evidences in the assessment model. Therefore, the value of j is from 1 to 6. “m” is the amount of indicators in six sub-evidences. We can obtain that the values of m is 3, 3, 2, 2, 2, 5from Figure 1. Weights are obtained by expert scoring. Weights of evidence are shown in Table 1. 4.3 Evidence Synthesis Theory Evidence Synthesis Theory is also called DS Theory. It is a kind of uncertainty reasoning and processing method. The merger of evidences and the update of trust function are the basis of Evidence Synthesis Theory. Its uncertainty is described by the concept of identifying the framework, basic trust, trust function, true degree, and the reliability interval. 4.3.1 Basic definition DS evidence theory is an extension method of Bayes, it can make the information clear in the case of the lack of information. It has stronger decision-making process capability. DS evidence theory define the frame of discernment Θ. All the possible sets of Θ can be described by power-set 2Θ . If polar function m: 2Θ → [0, 1], satisfy m(f) = 0 and ∑A⊆Θ m(A) = 1 , function m is called the basic probability assignment on Θ. ∀A ⊆ Θ, m(A) is called Mass function or basic credibility of A . For two evidences, the combine rule of Dempster is shown as follows: 0, [ m1 m2 ] ( A) m1( Ai )m2( Bj ) Ai Bj A , 1 K A f; ( 10) A f . 4.3.2 Evidence synthesis The concept of confidence degree α is introduced to correct the credit assignment considering that different importance of sub-evidences. m'( A) mA ( ) , m( ) 1 ( 11) Where, m( A) is belief function value with corrected; m( ) is reliability allocation of uncertain evidence. ' ,