International Conference on Civil, Transportation and Environent (ICCTE 206) An Approach of Converter Transforer Condition Evaluation Based on The Belief Rule Base Inference Methodology and Evidence Reasoning Qi Long,a, Yi Li2,b, Yong Sun3,c, Shaojun Yang,d, Qing Li3,e, Youping Fan2,f CSG EHV Power Transission Copany, Guangzhou 50620, China 2 School of Electrical Engineering, Wuhan University, Wuhan 430072, China 3 Maintenance& Test Center, CSG EHV Power Transission Copany, Guangzhou 50663, China a longqi@ehv.csg.cn, bdqliyi@whu.edu.cn, csunyong@ehv.csg.cn, dyangshaojun@ehv.csg.cn, e liqing@ehv.csg.cn, fypfan@whu.edu.cn Keywords: converter transforer; condition evaluation; belief rule base; evidential reasoning; RIMER Abstract. Various types of low-level indexes and uncertainty of evaluation inforation due to quantitative index and qualitative index exist in converter transforer condition evaluation. To solve this proble, a novel converter transforer condition evaluation approach based on the belief rule base inference ethodology and evidence reasoning (RIMER) is proposed. It firstly builds rule base with belief structure and transfors input inforation to a pre-defined for, then calculates the activation weight of input, after that using evidence reasoning for index aggregation to get the final evaluation distribution. At last, a converter transforer condition evaluation in the 500v converter station is investigated to illustrate the usage of the ethodology and its advantages. Introduction Converter transforer is the very iportant equipent in the DC transission syste. Besides, it is the core equipent connecting both the converter and inverter station in HVDC syste. Evaluating the converter transforer running state accurately and tiely can reduce the aintenance cost, shorten the outage tie, iprove equipent utilization and the reliability level of the transission syste, which is the ey to realize successfully converter transforer condition based aintenance. It has the iportant theory significance and practical value. In view of various types of low-level indexes and uncertainty of evaluation inforation in the converter transforer condition evaluation, a novel converter transforer condition evaluation ethodology based on the belief rule base and evidence reasoning (RIMER) is proposed in this article. It firstly builds rule base with belief structure and transfors input inforation to a pre-defined for, then calculates the activation weight of input, after that using evidence reasoning for index aggregation to get the final evaluation distribution. At last, a converter transforer condition evaluation in the 500v converter station is investigated to illustrate the usage of the ethodology and its advantages. Theory of RIMER RIMER has been put forward by professor Yang in 2006 as a ind of belief rule base and reasoning ethod based on evidential reasoning algorith, with the odeling ability of uncertainty and fuzzy uncertainty and a probability of nonlinear characteristics of data. RIMER ainly includes two aspects: one is the expression of nowledge, the other is the nowledge reasoning. Knowledge expression is achieved through BRB(belief rule base) syste while nowledge reasoning is through ER(evidential reasoning) algorith. Belief rule base. Belief rule base is a rule base which is coposed of IF-THEN rules and each rule is belief structure. Rules in the rule base rules represented by Eq.. R : If x = A x2 = A2 L xt = AT Then{(r, β, ),L, (rl, β L, )} () x is input vector, Ai is the reference value of the preise condition i in rule. β i is the degree of 206. The authors - Published by Atlantis Press 856
L confidence of conclusionr i in rule while preise condition was established. By the way, β, l= l. The relative weight of rule is θ,while relative weights between the preise conditions in this rule is δ δ 2 δ T. Evidence reasoning. When the input inforation x arrives, using the ER algorith to cobine the belief rules in BRB to get the final output of BRB syste, this is the basic idea of RIMER. RIMER ainly through the following three steps to ipleent the BRB syste of reasoning. The activation weight of input inforation x to rule is calculated through the forulas below ω M δi θ ( αi ) i= = L M l θ ( αi) l l= i= ω [0,],=,2,,L; α i (i=,2, M) is the degree of confidence of input x i to the reference value A i in rule, αi { α, α2, L, αij, } The specific generated ethod of α i j see Eq.6.Calculating correction factor shown as Eq.3~ Eq.4 T At φ(, t) αt, j t=, ct R µ =, φ(, t) T = (3) 0, ct R φ(, t) t= l, l, δi (2) β = β µ (4) µ donates for correction factor, the correction of the results. If the input inforation is coplete, A naely t and t α t,,there is β l, = βl,. T donates for the evaluation index nuber in rule. Using the ER algorith. After the rule is activated,there is ω 0. The cobination of BRB rules using the ER algorith is ainly coposed of the following two steps. Firstly, transfer the degree of confidence βl, of output part into basic probability for, as shown in Eq.5~Eq.8 = ωβ (5) l, l, = ω β (6) R, l= l, R, ω L = (7) % ω ( β ) (8) R, l l, L = = l, the basic probability for assigned to conclusion l in rule. R, is the basic probability for of that rule is assigned to no conclusion, apparently R, = R, + % R,.Then fuse the ultiple rules activated ( ω > 0) by the input. Might as well set the first S rules are activated,the calculation process is as Eq.9~ Eq.4 = K ( + + ) (9) le, ( + ) E ( + ) le, ( ) l, + le, ( ) R, + RE, ( ) l, + = K ( ) (0) RE, ( + ) E( + ) RE, ( ) R, + % = K ( % + % + % ) () RE, ( + ) E( + ) RE, ( ) R, + RE, ( ) R, + RE, ( ) R, + L L KE( + ) = le, ( ) t, + l= t=, t l (2) le, ( S) βl= (3) RE, ( + ) 857
β R % = RE, ( S) RE, ( + ) where β l donates for degree of confidence of the evaluation results, β Rdonates for degree of confidence of uncertainty of the evaluation results. le, ( ), RE, ( ), % RE, ( ) are results of the fusion of the first rules, and stipulate that l, E() = l,, RE, () = R,, % RE, () = % R,. The conversion of input inforation of different fors. The conversion of input inforation is to convert input inforation into a reliability data structure, and calculate the atching degree of input to the preise condition of rules, the input fors are as Eq.5 ( x, ε) ( x2, ε2) L ( x M, ε M ) (5) ε ii ( =,2, L, M) donates for the degree of confidence set to input, reflecting the uncertainty of the other i input, M is the nuber of inputs. Using fuzzy seantic value to describe quantitative input inforation. For the evaluation syste, evaluation index ostly has the following fors of quantitative inforation()deterinistic nuerical;(2)closed interval;(3)triangular fuzzy nuber;(4)the trapezoidal fuzzy nuber. In order to convert the input inforation x iof the above for into atching degree α jrelative to the fuzzy reference value A j, calculation forula is as Eq.6 ϕ( x α j) εi α Ji (6) ϕ( x, α ) α j [0,], ϕ( x i, α i, j ) is the siilar degree of xand A i j, called siilar function. The selection of ϕ is related to index type, specific discussion see the literature [-3] Use the subjective decision to describe the qualitative input inforation.the degree of confidence α jof input relative reference A j is given directly by the decision aers (or experts) according to the subjective. For exaple, if ε jis the degree of confidence relative to A jwhich is the experts offer for qualitative input inforation, then α j = ε j. i j (4) The evaluation odel based on RIMER The selection of evaluation index. Through statistical analysis and query converter transforer current rules, analysis of the experts' opinions and inducing 4 low-level evaluation indexes of converter transforer reflecting running state, as input inforation of the evaluation odel, as shown in Fig. 858
c c 2 c c 3 c 4 c 2 c 2 c 22 c 23 c 3 c 3 c 32 c 33 c 4 c 0 c 42 c 4 c 43 Fig. Evaluation indexes of converter transforer condition The evaluation indexes are ainly divided into 4 categories such as electrical test c, oil test c 2, defect repair records c 3 and operation conditions c 4. The evaluation indexes fro aspects of the equipent itself, test and operation coprehensively reflecting the service life of converter transforer such as insulation aging condition, internal health, electrical or echanical properties, and fro vertical and horizontal aspects to analyse the probles left over by history and defects of the faily. On account of that the paraeters reflecting the running state of the converter transforer are so any. In order to analyse it conveniently, this paper uses the index syste of the tower structure, which can ore easily grasp the relationship between the various indicators. The established status of the evaluation index syste is a hierarchy structure of three layers, including priary evaluation indexes, 4 secondary evaluation indexes and 4 tertiary evaluation indexes. The conversion input inforation. For quantitative test index inforation, using relative deterioration degree with ebership function ethod to deal with it. Relative deterioration degree is a scope for quantitative indexes of [0,],to characterize the deterioration degree of transforer current running status copared to the fault status. The specific calculation process see literature[4]. According to the degree of degradation, and then cobining with the relevant rules and expert experience to deterine electrical triangle and half trapezoid distribution function for three inds of state level (good, average, bad) fuzzy boundary range, establishing relative deterioration degree grade of ebership function for each state. The final results are presented as ebership degree R = [ r, r, r ], r [0,] ( j =,2,3) is representing ebership degree of three inds of state level set i i i2 i3 ij (Good, Average, Bad ) respectively. The specific calculation process see literature [5]. Using expert scoring ethod to deal with qualitative inforation such as defect repair record and operation condition inforation. First using expert survey for, that is, gives the evaluation objects and evaluation index according to the survey, by the experts estiate evaluation atrix, then cobined with analytic hierarchy process (AHP), calculating ebership degree of each the final indexes. The specific calculation process see literature [6-7].Transforation of input inforation degree of confidence is to transfor the input inforation into confidence data structure. And the input for corresponding to evaluation index is Eq.7 ( x, ε ) ( x, ε ) ( x, ε ) (7) 2 2 3 3 x i donates for the precondition of input attributes, ε i donates for degree of confidence of x i,taing c 44 859
biggest two degree of confidence of the preise properties as input inforation. Such as insulation resistance is 235.6 MΩ,after relative deterioration degree and ebership degree calculation,the results are R =[0.74,0.25,0].The input inforation of insulation resistance is (Good,0.74) (Average,0.25), siplified as c =(0.74,0.25,0). Establishing the initial belief rule base. Establishing an initial belief rule base according to the evaluation index preise attribute reference set by the above and expert experience. Assuing that the rules weights in rule base are equal, the relative weight of the preise of rules are also equal, naelyθ = θ l ( l), δ 2 = δ =L = δ T The initial belief rule base includes 4 tertiary evaluation indexes, 4 secondary evaluation indexes and priary evaluation index. Tertiary evaluation index is divided into three grades: good, average and bad. Priary evaluation index and secondary evaluation index is also divided into three levels: noral, warning and urgent. The initial belief rule base has 297 rules. Liited to the space, listing part of the, as Table. Table The initial belief rule base ID Preise condition Evaluation result ID Preise condition Evaluation result R R 2 R 3 R 4 R 5 c =A,c 2 =A,c 3 =A,c 4=A c =A,c 2 =A,c 3 =A,c 4=B c =A,c 2 =A,c 3 =A,c 4=C c =A,c 2 =A,c 3 =B,c 4=A c =A,c 2 =A,c 3 =B,c 4=B c ={(Nor.,),(War.,0),( Urg.,0)} c ={(Nor.,),(War.,0),( Urg.,0)} c ={(Nor.,0.2),(War.,0.5 ),(Urg.,0.3)} c ={(Nor.,0.7),(War.,0.3 ),(Urg.,0)} c ={(Nor.,0.4),(War.,0.6 ),(Urg.,0)} R 6 R 7 R 8 R 9 c =A,c 2 =A,c 3 =B,c 4=C c =A,c 2 =A,c 3 =C,c 4=A c =A,c 2 =A,c 3 =C,c 4=B c =A,c 2 =A,c 3 =C,c 4=C c ={(Nor.,0.),(War.,0.3 ),(Urg.,0.6)} c ={(Nor.,0.2),(War.,0.4 ),(Urg.,0.4)} c ={(Nor.,0.),(War.,0.2 ),(Urg.,0.7)} c ={(Nor,0),(War,0.2),( Urg,0.8)} A denotes for Good, B denotes for Average, C denotes for Bad, Nor. denotes for Noral, War. denotes for Warning, Urg. denotes for Urgent. Case study This paper studies a converter transforer of a 500V convertor station. According to consulting the history records of aintenance, we find that there was a sall defect record of the transforer few years ago and there is no faily defects or siilar equipent failure. In 203 and 205, there is a short-ter overload respectively. Two short circuit have occurred in 204 and there is no abnoral noise or operating over voltage. In a preventive test, the easured insulation resistance is ore than 0000 which is siilar to the last test, the absorption ratio is.25, the oil dielectric loss is 0.53, the axiu change rate of winding DC resistance phase is.35%, the winding dielectric loss is 0.32, the oil icro water is 5.6g/l, and the oil breadown voltage is 48v. After calculating the relative deterioration degree and ebership degree of the evaluation index, and then the reliability is converted, the calculation results are as follows:c =(,0,0), c 2 =(,0,0), c 3 =(0.44,0.56,0),c 4 =(0.38,0.62,0),c 2 =(,0,0),c 22 =(0.88,0.,0),c 23 =(0.75,0.24,0),c 3 =(,0,0),c 32 =(,0,0),c 33 =(,0,0),c 4 =(0.33,0.66,0) c 42 =(0.45,0.55,0),c 43 =(,0,0),c 44 =(,0,0) First, calculating the electrical test c, and c =(,0,0),c 2 =(,0,0),c 3 =(0.44,0.56,0),c 4 =(0.38, 0.62,0) are nown. The input activates the rule R, R 2, R 4, R 5 in belief rule base, according to the Eq.2,Eq.3 and Eq.6, ω =0.2, ω 2 =0.9, ω 4 =0.26, ω 5 =0.43, µ =, µ 2 =, µ 4 =, µ 5 =. According to the Eq.5~Eq.8,(, 2, 3, R, R, % R )=(0.08,0.04, 0,0.88,0.88,0),( 2, 22, 32, R2, R2, % R2 )=(0.07,0.2,0,0.8,0.8,0),( 4, 24, 34, R4 R4, % R4 )=(0.2,0.4, 0,0.74,0.74,0),( 5, 25, 35, R5, R5, % R5 )=(0.5, 0.28,0,0.57,0.57,0). According to the Eq.9~Eq.4, c (Noral, Warning, Urgent) =(0.53, 0.47, 0). 860
Siilarly, followed by calculation of the oil test c 2, defect repair record c 3 and operating condition of the c 4, and the value of converter transforer operation condition c 0. In the calculation process fro the botto, the index is upper, the nuber of rules activated is ore, so in the MATLAB prograing the above algorith is achieved according to the Eq.2~Eq.6. The results are as follows: c 2 =(0.84,0.5,0), c 3 =(,0,0), c 4 =(0.4,0.58,0), c 0 = (0.49,0.5,0).Fro the calculation results, it can be seen that the running state of converter transforer is between "Noral" and "Warning", and the degree of confidence of the "Warning" is higher. Aong the, degree of confidence of the " Warning" in the evaluation results of electrical test c and operation condition c 4 are both high, which should be paid attention to. The actual situation is that the converter transforer has ran for a long tie and there were twice over load, leading to partial aging of internal insulation, the overall status is safety, and the transforer can continue to run, but it needs to be observed, which is consistent with the evaluation result. Conclusions The approach of converter transforer condition evaluation based on the belief rule base inference ethodology and evidence reasoning (RIMER) is a coprehensive evaluation ethod which cobine qualitative with quantitative and calculation results with expert experience. Although the evaluation index type is of diversification and evaluation inforation is incoplete, this ethod provides a probability distribution of the overall equipent operation status. It better solved the proble to evaluate converter transforer condition accurately due to the inforation uncertainty lead by the cobination of quantitative index and qualitative index. The case shows that the converter transforer condition evaluation approach based on the belief rule base inference ethodology and evidence reasoning (RIMER) can well reflect the actual running status of equipent and can provide the scientific decision-aing of aintenance for the next step. How to establish a belief rule base appropriately and confir values of preise weights precisely is the ey point to this evaluation approach and also the future research direction of the ethod. References [] Yang Jianbo. European Journal of Operational Research,200, 3():3-6. [2] Bustince H, Burillo P. Fuzzy Sets and Systes,2000. 3: 205-29. [3] Chen S M, Hsiao W H. Fuzzy Sets and Systes, 2000, 3: 85-203. [4] Viasne S, Wets G, Vanthienen J. Fuzzy Sets and Systes,2000, 3: 253-265. [5] Wang Qian, Study of the Coprehensive Assessent Method for The Power Transforer Condition in service with Fuzzy theory[d]. Chongqing University, 2005. (In Chinese) [6] Zhang Deng, Research on Condition Assessent Method for Power Transforer[D]. Huazhong University of Science and Technology, 203. (In Chinese) [7] Zhang Biao, Wang Yuhong, Li Xingyuan, etc.high Voltage Apparatus, 205, 5, (9), 60-65. (In Chinese) 86