2236 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO.

5, OCTOBER 2018

Conductor Temperature Estimation and Prediction

at Thermal Transient State in Dynamic
Line Rating Application
David L. Alvarez , Student Member, IEEE, Filipe Faria da Silva, Member, IEEE,
Enrique E. Mombello , Senior Member, IEEE, Claus Leth Bak , Senior Member, IEEE,
and Javier A. Rosero , Member, IEEE

Abstract—The traditional methodology for defining the ampac- I. INTRODUCTION

ity of overhead lines is based on conservative criteria regarding the
ITH the constant increase in power consumption, an
operating conditions of the line, leading to the so-called static line
rating. Although this procedure has been considered satisfactory
for decades, it is nowadays sensible to account for more realistic
W upgrade and update of current assets are necessary for
control and operation of existing power networks. As a result of
line operating conditions when calculating its dynamic ampacity. the advances in renewable power generation, such as wind and
Dynamic line rating is a technology used to improve the ampac-
ity of overhead transmission lines based on the assumption that solar energy, there exists a constant growth in new power plants.
ampacity is not a static value but a function of weather and line’s Therefore, bottlenecks are arising in transmission level, mainly
operating conditions. In order to apply this technology, it is nec- in overhead lines (OHLs), which are facing economic, social,
essary to monitor and predict the temperature of the conductor political and implementation time challenges. In order to reduce
over time by direct or indirect measurements. This paper presents both congestion and face these challenges, different techniques
an algorithm to estimate and predict the temperature in overhead
line conductors using an Extended Kalman Filter, with the aim of [1] can be used depending on the characteristics of the line.
minimizing the mean square error in the current and subsequent Among these solutions is monitoring the line state allowing the
states (temperature) of the conductor. The proposed algorithm as- assessment of thermal limits and the application of DRL [2],
sumes both actual weather and current intensity flowing along the as long as the ampacity is limited by the sag of the catenary.
conductor as control variables. The temperature of the conductor, As the forerunner of DLR, OHL’s ampacity by probabilistic
mechanical tension and sag of the catenary are used as measure-
ments because the common practice is to measure these values with methods was introduced using seasonal atmospheric conditions
dynamic line rating hardware. The algorithm has been validated [3]. Subsequently, the monitoring of OHL’s thermal state was
by both simulations and measurements. The results of this study reached using information technologies.
conclude that it is possible to implement the algorithm into Dy- Because of only one span in an OHL can limit the ampacity
namic Line Rating systems, leading to a more accurate estimation and its behavior depends on the adjacent suspension spans, this
and prediction of temperature.
set of spans is assumed monitored for DLR. This set is com-
Index Terms—Dynamic Line Rating (DLR), dynamic state monly known as the critical stringing section. However, this
estimation, extended kalman filter (EKF), overhead line (OHL). section can change over time as a result of weather variations.
Consequently, different methodologies can be used to identify
critical stringing sections and to define DLR devices location
[4], [5].
In OHLs, two types of thermal limits are defined. The first
Manuscript received July 7, 2017; revised February 1, 2018; accepted April 7, one is related to thermal equilibrium (steady state) and used for
2018. Date of publication April 27, 2018; date of current version September 25, planning and control. The second one is related to transient state
2018. This work was supported by the Colombian Department of Science, Tech- and given by a relationship between current intensity and time;
nology and Innovation (Colciencias) under Project 617 — National Doctorates.
Paper no. TPWRD-00894-2017. (Corresponding author: David L. Alvarez.) this limit is used for contingencies assessment during operation.
D. L. Alvarez and J. A. Rosero are with the Departamento de Inge- Using DLR, the data required to define these two limits are
nierı́a Eléctrica y Electrónica, Facultad de Ingenierı́a, Universidad Nacional historic reports of weather or low scale atmospheric models
de Colombia, Bogotá 111321, Colombia (e-mail:, [email protected];
[email protected]). based on local measurements [6] and direct measurements in
F. Faria da Silva and C. L. Bak are with the Department of Energy Tech- critical stringing sections whether of sag length, mechanical
nology, Aalborg University, Aalborg 9100, Denmark (e-mail:, [email protected]; tension, inclination, clearance, among others [7].
[email protected]).
E. E. Mombello is with the Instituto de Energı́a Eléctrica, CONICET, The ampacity limit at steady state using DLR can estimate
Universidad Nacional de San Juan, San Juan J5400ARL, Argentina (e-mail:, with weather forecast [8]. On the other hand, to compute dy-
[email protected]). namic limits (at thermal transient state) with DLR, it is necessary
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. to know on-line both the conductor temperature and the atmo-
Digital Object Identifier 10.1109/TPWRD.2018.2831080 spheric conditions. To this end, some techniques are proposed,

0885-8977 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 ALVAREZ et al.: CONDUCTOR TEMPERATURE ESTIMATION AND PREDICTION AT THERMAL TRANSIENT STATE 2237

such as computing wind speed from direct measurements [9] the thermal, electrical, and mechanical characteristics of OHLs.
or including a weather station together with the direct monitor- Consequently, it is possible to estimate the thermal state of the
ing device [10]. However, given the nature of the atmospheric conductor by monitoring these physical changes.
conditions, which vary in space and time and the uncertainty
in the parameters of the OHL, an inaccuracy is obtained in the A. Heat Transfer at Transient State
ampacity when the temperature is used [11].
Heat transfer in OHL conductors is a well-known process
The variation in the atmospheric conditions along a stringing
[22] and is described in standards and guides [21], [23]. The
section can be modeled by the average conductor temperature
main equation for this process is
with an effective wind speed in order to avoid spot tempera-
ture [12]. The impact of data uncertainty is addressed in the dTS ik m 2 R (TS ) + qs − qc (TS ) − qr (TS )
literature. For instance, the uncertainties in input data as well = (1)
dt mCp
as in the parameters used in heat transfer models are addressed
by affine arithmetic in [13]. Similarly, robust corrective control where TS is the temperature of the conductor at the surface, ik m
measures considering the weather forecast uncertainty is used is the current intensity, R is the ac electrical resistance per unit
in reference [14]. The impact of the uncertainty in both the cate- length, qs is the solar heating, qc is the convective cooling, qr
nary parameters and temperature in the calculation of the sag is is the radiative cooling, m is the mass per unit length and Cp
analyzed in [15]. An enhanced methodology is presented in [16] is the specific heat capacity of the conductor. (1) can solve by
using on-line information of a self-organized sensor network. numerical integration by using
This network uses temperature sensors and has the ability to ik m 2 R (TS ) + qs − qc (TS ) − qr (TS )
predict, estimate and validate information used for DLR. In this ΔTS = Δt (2)
mCp
way, this paper presents a state estimation algorithm for DLR
at thermal transient state which allows to estimate and predict taking time intervals Δt, provided that the initial temperature,
the average conductor temperature of stringing sections. The the thermal parameters of the conductor and the atmospheric
algorithm is based on an Extended Kalman Filter (EKF), and conditions along the integration time are known. The computing
it has the advantage of using available DLR systems. To im- time to calculate temperature by this numerical method is not a
plement the EKF, it is provided that the set of critical stringing problem with modern computers, because under contingencies
sections are monitored by DLR hardware and their atmospheric or normal operation the thermal constant of the conductors is
conditions are known. higher (commonly 15 min) than the processing time used to
The motivation to propose this algorithm is that currently used solve it (less than 1 s).
methods to minimize errors in the estimation of temperature in The maximum current intensity (|ik m |m ax ) at conductor
OHLs [17]–[19] are probably not the best choice for on-line reach the thermal equilibrium can compute using the maximum
dynamic state estimation during thermal transients. With the allowable conductor temperature (TS m a x ) as follow
EKF, estimation and prediction of both states and parameters 
of nonlinear dynamic systems is reached [20]. Additionally, the qc (TS m a x ) + qr (TS m a x ) − qs
|ik m |m ax = (3)
uncertainties in the atmospheric conditions, the current inten- R (TS m a x )
sity and the direct measurements are considered by the proposed
Thus, OHL’s ampacity can be estimated using both static or dy-
EKF with the computing of covariance propagation matrix and
namic line ratings. For contingencies management, commonly
the Kalman gain. The state variables of the proposed EKF are
the maximum current intensity vs time plot is computed solving
the average conductor temperature, the average effective wind
(1) until the conductor reach the maximum allowable tempera-
speed, the emissivity and the solar absorptivity of conductor sur-
ture for different values of ik m .
face. The average temperature was chosen because it is possible
to estimate the OHL ampacity with this value. The consideration
of additional parameters leads to improvements in temperature B. Direct Measurements for DLR
prediction, since wind speed has the greatest impact on cooling Although (1) correctly models the behavior of temperature in
[12], and emissivity and absorptivity commonly present a high OHL’s conductors, there exist uncertainties in the computing re-
uncertainty [21]. sults because of inaccuracies in the inputs and parameters. Thus,
This paper is organized as follows: Section II provides a brief direct measurements for DLR are required in critical stringing
introduction to heat transfer in OHL’s conductor and to direct sections to enhance the accuracy. With these measurements,
measurements used in DLR. Section III introduces the algorithm the thermal state is measured discretely by taking samples be-
developed. Section IV presents a case study and experimental tween 1 min, and 10 min [12], allowing a thermal monitoring.
test carried out with the aim of evaluating the performance of Direct measurements are classified by CIGRE [7] into temper-
the algorithm. Finally, conclusions are presented in Section V. ature, sag and mechanical tension. From measurements of these
variables, the temperature of the conductor is computed using
known relationships, such as the state equation (temperature
II. BACKGROUND related to tension) and the catenary equation (related tension
Thermal behavior of OHLs is determined by heat transfer as with sag). Although the conductor temperature can be directly
a result of heat gains and heat losses. This phenomenon affects monitored, the monitoring system has errors produced whether

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 2238 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

modeled by
⎡ ⎤
f (x, u, w, t)
⎢ ⎥
⎢ 0 ⎥
⎢ ⎥
ẋ = ⎢ ⎥
⎢ 0 ⎥
⎣ ⎦
0
zk = h (xk , vk )
w (t) ∼ (0, Q)
vk ∼ (0, Rk ) (4)
where f is the function (1), x is the state vector, u is the vector of
Fig. 1. Proposed EKF model to estimate and predict thermal states on OHL control variables, t is the time and w are the errors in the system.
conductors. T
The state vector is x = TS |ϑ| εs αs . The control variables
selected are |ik m |, the ambient temperature (Ta ), the wind attack
angle (δ) and the solar radiation (S), i.e., u = |ik m | Ta δ S .
by changes in the temperature along the span, the influence of
Finally, the state variables are related to a set of measurements
measurement devices over the spot where the reading is taken,
zk at time k by means of measurement functions h (xk , vk ),
uncertainties in the catenary parameters such as the mechanical
which have errors vk . The errors vk and w are assumed to have
tension reference, conductor creep, among others. Finally, an
a normal probability distribution with mean zero and covariance
error propagation occurs in the prediction of temperature during
Q and Rk .
a thermal transient given the uncertainty of atmospheric con-
ditions. For instance, when the temperature in the conductor
reaches the steady state after a thermal transient, it is not af- B. Prediction of Future States
fected by the initial value of temperature, but only atmospheric  a system error w = 0 is assumed
To predict a future state,
conditions and current intensity. and a state prediction x̂−k is carried out at time t by
⎡  + ⎤
f x̂k −1 , u, 0, t
III. PROPOSED DYNAMIC STATE ESTIMATOR ⎢ ⎥
⎢ 0 ⎥
⎢ ⎥
A hybrid EKF algorithm is proposed in this paper since the x̂˙ − = ⎢ ⎥
k
⎢ 0 ⎥
heat transfer phenomenon in conductors is continuous in time, ⎣ ⎦
and measurements for DLR are commonly taken at discrete 0
times. With this algorithm is aimed to estimate and predict the
temperature in OHL’s conductors by using both direct measure- Ṗ−
k = FP++ P+
k −1 k −1 F + LQL
T T
(5)
ments of DLR and the atmospheric conditions. To implement an  +  −
taking the estimation of the current states x̂k −1 . x̂˙ k is com-
 to model the system (ẋ = f (x,
EKF is necessary  .+. .)), predict puted by numerical integration (2). P is the covariance of the
future states x̂− and update the current states x̂k with new
k estimation error, F is the Jacobian of the model with respect to
measurements (zk ). The proposed EKF for DLR is described
state variables (F = ∂f /∂x) calculated by
by means of Fig. 1 as follows:
⎡ df df df df ⎤
⎢ dTS dϑ dεs dαs ⎥
A. System Modeling ⎢ ⎥
⎢ 0 0 0 0 ⎥
F= ⎢
⎢
⎥
⎥ (6)
Significant errors in the prediction of the temperature dur- ⎢ 0
ing a thermal transient occur mainly because of inaccuracies ⎣ 0 0 0 ⎥⎦
in the value of wind speed under forced cooling. Addition- 0 0 0 0 x̂,u
ally, the values of emissivity (εs ) and solar absorptivity (αs )
of conductor surface can vary between 0.2 and 0.9 [21], de- and L is the Jacobian of the model with respect to control
pending on the environmental conditions and time. Hence, this variable errors (L = ∂f /∂w) computed using
paper proposes to consider (εs, αs ), along with the effective ⎡ df df df df ⎤
wind speed (|ϑ|) and the average conductor temperature as state ⎢ dwi k m dwT a dwδ dwS ⎥
variables. By assuming that OHL’s thermal constant is in a ⎢ ⎥
⎢ 0 0 0 0 ⎥
time interval of 5–15 min and based on CIGRE recom- L= ⎢
⎢
⎥
⎥ (7)
mendations [12], |ϑ| can consider as the average effective wind ⎢ 0 0 0 0 ⎥
⎣ ⎦
speed during this period and therefore assumed constant. Like-
wise, εs and αs are assumed constant. Thus, the system can be 0 0 0 0 x̂,u

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 ALVAREZ et al.: CONDUCTOR TEMPERATURE ESTIMATION AND PREDICTION AT THERMAL TRANSIENT STATE 2239

TABLE I
Algorithm 1: Proposed Algorithm for DLR Dynamic SE. ATMOSPHERIC CONDITIONS TAKEN FROM CIGRE GUIDE [21]
1: procedure Hybrid EKF(zk , uk , x̂+ +
k −1 , Pk −1 , Δt, tk ,
Q, Rk )
+
2: x̂−
k ← x̂k −1
3: Pk ← P+
−
k −1
4: for j ← Δt to tk step Δtdo  Predict
5: x̂˙ −
k ← f x̂ −
k , u, 0, Δt
˙− Temperature Estimation: Each measurement sample was pro-
6: x̂− −
k ← x̂k + x̂k cessed using the Algorithm 1, where x̂+ +
k and Pk are updated,
7: F ← ∂f /∂x|x̂ −k ,u k
and used as inputs for the next estimation, as shown in Fig. 1.
8: L ← ∂f /∂w|x̂ −k ,u k Thus, the ability of the EKF to use the information of previous
9: Ṗ− − − T
k ← FPk + Pk F + LQL
T
measurements is used.
− − −
10: Pk ← Pk + Ṗk Temperature Prediction: It is performed to obtain the pre-
11: end for dicted value of temperature at time k + ΔtC , where ΔtC is the
12: Hk ← ∂h/∂x|x̂ −k assumed duration of a contingency. Thus, a temperature predic-
13: Mk ← ∂h/∂v|x̂ −k tion is performed by means of (1) at time k + ΔtC using the
 
T −1 estimated values at time k.
14: Kk ← P− k Hk H
T − T
k Pk Hk + Mk Rk Mk
  − 
15: x̂+ −
k ← x̂k + Kk zk − h x̂k  Update
A. Simulation Results
16: P+k ← (I − K k H k ) P−
k (I − K k H k )T
+
Kk Mk Rk M  k +Kk + 
T T To test the algorithm with simulations, the data for temper-
17: return x̂k , Pk ature tracking calculation given in [21] was used, assuming a
18: end procedure span with a length of 300 [m], having a horizontal component
of conductor tension of 24.2 [kN] at 20 [◦ C]. To simulate the
measurements and the control variables random errors (vk , w)
C. Update of Current States were added to the assumed Theor. values, as shown in Fig. 1.
Normal distributions of the error with mean zero and a stan-
The system update is performed by
dard deviation (σz k ) considered as the third part of the accu-
 
T −1
Kk = P− − T
k Hk Hk Pk Hk + Mk Rk Mk
T racy were assumed; therefore, the variances are computed as
  −  var (zk ) = σz k 2 . A typical accuracy of ±1.5 [K] [7] in the
x̂+ −
k = x̂k + Kk zk − h x̂k measurements of conductor temperature was used. Hence, if a
maximum conductor operating temperature of 75 [◦ C] is used, a
P+ − T
k = (I − Kk Hk ) Pk (I − Kk Hk ) + Kk Mk Rk Mk Kk
T T
standard deviation of σT S = 1.5/3 [K] in temperature measure-
(8)
ments is equivalent to σD = 5.5/3 [cm] in measurements of sag
using the measurements recorded at time k, where K is the and to σH = 100/3 [N] in measurements of mechanical tension.
Kalman gain, H is the Jacobian of the measurement functions Finally, simulations were run with a ΔtC = 15 min and direct
respect to state variables (H = ∂h/∂x) calculated with measurements recorded at time samples of tk = 1 min.
  1) Thermal Transient System Description: A thermal transient
dh
H= 0 0 0 (9) with measurements of current intensity and atmospheric condi-
dTS x̂ tions provided every 10 min is assumed, as shown in Table I
and M is the Jacobian of the measurement functions respect to [21]. The conductor DRAKE ACSR (aluminum (a) and steel
measurement errors (M = ∂h/∂v) computed by (s)) was used to simulate the span. Its properties are shown in
Table II, where A is the cross-sectional area, d is the conductor
dh (TS ) diameter, RT r e f is the conductor AC resistance at temperature
M= (10)
dv Tr ef , β is the linear temperature resistance coefficient and α is
x̂
the coefficient of linear thermal elongation.
The expressions used for computing the partial derivatives of
2) Temperature Estimation and Prediction: Known examples
matrices F, L, H, M can be found in [19]. Finally, the proposed
were used to assess the performance of the algorithm assuming
EKF for DLR estimation is shown in algorithm 1.
direct measurements. For simulating direct measurements, the
theoretical temperature (TS -Theor.) during transient state was
IV. ALGORITHM VALIDATION computed with the values of Tables I–II by applying numeri-
In this section, both simulations and an experimental test are cal integration (2) to the heat transfer equation (1). Equivalent
performed to validate the effectiveness of the proposed EKF. values of sag and tension were computed using the values of TS -
Algorithm 1 was implemented in Matlab with time steps Δt = Theor. and measurement functions [19]. Then, direct measure-
0.1 [s]. The EKF was evaluated in the estimation of temperature ments were simulated with the Matlab function randn, adding
for real time monitoring and in the prediction of temperature for values of normal random errors with mean zero and equiva-
contingencies management as follows: lent standard deviation to the sag, tension and temperature. The

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 2240 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

TABLE II
CONDUCTORS USED FOR SIMULATIONS AND LABORATORY TEST

Fig. 3. Theoretical wind speed |ϑ| and estimated wind speed ϑ̂ using the
proposed EKF for the assumed critical scenarios.

Fig. 2. Both theoretical and limits of temperature of the conductor during

transient state, computed with the scenarios: lower limit - |ϑ k | = |ϑ k |Theor. +
0.5 [m|/s] and upper limit - |ϑ k | = |ϑ k |Theor. − 0.5 [m|/s].

simulated temperature measurement zk =0 was selected to be

the value of x̂+ +
k =0 , and for the covariance the value P̂k =0 = σ
2

was assumed. Since in these simulations the aim is to analyze

the performance of the algorithm provided that direct measure-
Fig. 4. Temperature of the conductor (T S -Theor.), simulations of measure- 
ments of conductor temperature (T S Simulated), estimated temperatures TS
ments are available, the values of atmospheric conditions except and maximum allowable current intensity (|ik m |m a x ) at steady state using the
for the wind were assumed without errors, that is w (t) = 0. proposed algorithm during the thermal transient.
For the forecast average effective wind speed an uncertainty of
±0.5 [m/s] was assumed (σ|ϑ| = 0.5/3 [m/s]).
To test the algorithm, two critical scenarios were mod- is modeled discreetly in this example, both |ϑ̂k | = |ϑk | and
eled. The first considering a wind speed with the tempera- P̂k (2, 2) = σ|ϑ| 2 must be reset during the run of the algorithm
ture lower limit |ϑk | = |ϑk |Theor. + 0.5 [m/s] and the second at each time k in which the wind changes.
one using a wind speed with the temperature upper limit a) Simulations Assuming Temperature Measurements: For
|ϑk | = |ϑk |Theor. − 0.5 [m/s]. Fig. 2 shows the values of tem- simulation of direct temperature measurements, normal random
perature during the thermal transient of TS −Theor. and the errors with mean zero and σ = 1.5/3 [K] were added to the TS -
ones for the critical scenarios. The shaded area shows the ob- Theor. Fig. 3 shows wind speed estimated for both critical cases.
tained uncertainty limits. To compute the root-mean-square er- Fig. 4 shows the values of TS -Theor., simulations of measured
ror (RMSe), for the lower limit was 4.02 [K], and for the up- temperature and estimated temperature using the proposed al-
per limit of 6.73 [K]. Finally, provided that the effective wind gorithm with the two critical cases. The error e is computed with

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 ALVAREZ et al.: CONDUCTOR TEMPERATURE ESTIMATION AND PREDICTION AT THERMAL TRANSIENT STATE 2241

 6. Simulation of mechanical tension (H ), and estimated temperature

Fig.
TS using the proposed algorithm simulating measurements of tension on the
conductor.

Fig. 5. Temperature of the conductor (T S -Theor.), and predictions 15 min,

before of temperature and maximum allowable current intensity during a con-
tingency computed with the proposed algorithm, simulating measurements of
conductor temperature.

respect to the TS -Theor. The RMSe for the estimated temper-

ature was 0.34 [K] taking the lower limit, and 0.54 [K] taking
the upper limit. The RMSe using simulations of direct temper-
ature measurements was 0.6 [K]. Additionally, Fig. 4 shows the
maximum allowable current intensity at steady state for a tem-
perature of 75 [◦ C]. The current intensity was computed with
(3) using the conditions of instant k for each scenario. Fig. 5
shows predicted values both of temperature and of maximum
current allowable during a contingency, obtained a RMSe for
the lower limit of 1.6 [K], and 2.2 [K] for the upper limit.
b) Simulations Assuming Tension Measurements: For simu-
lating tension measurements, errors were added as done in the  
Fig. 7. Simulation of sag length (D), and estimated temperature TS using
previous simulation using an accuracy of 100 [N]. Fig. 6 shows
the proposed algorithm simulating measurements of sag on the catenary.
the simulations of measured tension (H), TS -Theor., and the
estimated temperature with the proposed algorithm. The RMSe TABLE III
of both estimated and predicted temperature for the lower limit COMPARISON PERFORMANCE BETWEEN THE THREE KINDS OF DIRECT
were 0.18 [K] and 1.5 [K], and for the upper limit were 0.22 [K] MEASUREMENTS FOR 1000 RANDOM CASES
and 2 [K] respectively.
c) Simulations Assuming Sag Measurements: As in the case
of tension measurements, Fig. 7 shows the performance of the
algorithm when sag measurements are available. The RMSe of
both estimated and predicted temperature for the lower and the
upper limits were 0.28 [K] and 1.5 [K], and 0.29 [K] and 1.5 [K]
respectively. RMSe and the average computing time to run Algorithm 1. As
Finally, 1000 simulations for each one of the three direct result, the proposed algorithm showed stability, convergence
measurements were performed. To simulate a more realistic and speed, and it reached a smaller error in temperature values
case, errors were added on control variables, since these are than in the case of using only records of direct measurements.
commonly measured or assumed. Thus, normal random errors Although the standard deviation for δ was taken using typical
with mean zero were added to current intensity (|ik m |) with anemometers accuracy, this value is unrealistic, because of wind
σ = 5/3 [A], to ambient temperature (Ta ) with σ = 1/3 [K], turbulence along the stringing section. However, to estimate the
and to wind attack angle (δ) with σ = 12.5/3 [◦ ]. These standard average effective wind speed instead of spot values, the effect
deviations were taken from [19]. Table III shows the average of wind turbulence is considered [12].

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 2242 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

Fig. 10. Experimental test setup.

Fig. 8. Current intensity planned (|ik m |) and forecasted wind speed (|ϑ|)
used in the test.

Fig. 9. Circuit diagram of the experimental test.

Fig. 11. Measurements of current intensity (|ik m |), temperature of the con-
B. Experimental Results ductor (T S ) and ambient temperature (T a ) recorded every 30 [s].
A laboratory setup was designed to evaluate the algorithm.
The setup consisted of controllably injecting a current inten-
sity through the OHL conductor Linnet and measuring its tem- 2) Test Results: The setup was initially energized with 300
perature. The properties of the conductor are shown in Ta- [A], and when the conductor reached the thermal steady state,
ble II. To carry out the validation, an ambient temperature of the planned conditions (Fig. 8) were controlled and the variables
Ta = 19 [◦ C], and the planned current intensity (|ik m |) and the |ik m |, TS and Ta were measured and recorded every 30[s] with
wind (|ϑ|) shown in Fig. 8 were assumed as forecast values an accuracy of ±5 [A] and ±1.5 [K]. Measurements are shown
throughout the test. An auto-transformer and a fan were used to in Fig. 11. Since the test was carried out indoors, the solar
control both |ik m | and |ϑ|. As in the simulations, the two critical radiation was assumed to be S = 0.
cases in the estimation and prediction of the temperature were Fig. 12 shows the values of temperature computed using (2)
used. Additionally, a value of emissivity s = 0.9 was used as for the three cases and the temperature measured. The Root
initial parameter for the lower limit and a value of emissivity mean square residuals (RMS ) obtained were RMS = 2.4 [K]
s = 0.2 for the upper limit. Thus, three different cases were for case 1, RMS = 5.7 [K] case 2 and RMS = 5.5 [K] for
analyzed: case 1 using the assumed planned and forecasted val- case 3.
ues, case 2 using the upper limits, and case 3 using the lower 3) Estimation of Average Temperature: The values estimated
limits. both of effective wind speed and emissivity of the conductor
1) Test Setup: Considering the laboratory atmospheric con- using the proposed EKF in each case are shown in Fig. 13. The
ditions, the conductor under test theoretically reaches 75 [◦ C] estimated average conductor temperature for the case with the
with an |ik m | of almost 500 [A]. Taking the limitations of the highest RMS (case 2) is shown in Fig. 14. A RMS = 1.5 [K]
short circuit current of the laboratory into account, a special was obtained with this estimated temperature.
three-winding three-phase distribution transformer (HV-LV-LV) 4) Prediction of Temperature: Taking the case 2, the tem-
with open ends was used to reach this current, as shown in perature predicted 15 min before is shown in Fig. 15. In this
Fig. 9. The setup used is shown in Fig. 10. To reduce the temperature prediction, a RM S = 2.5 [K] was obtained. Ad-
influence of loop impedance the leads were located almost ditionally, Fig. 15 shows the maximum current intensity allow-
perpendicular to the conductor and the transformer was about able until the conductor reaches 75 [◦ C]. This current intensity
1 [m] away. was predicted 15 min before.

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 ALVAREZ et al.: CONDUCTOR TEMPERATURE ESTIMATION AND PREDICTION AT THERMAL TRANSIENT STATE 2243

Fig. 14. Comparison of estimated averaged temperature and measured tem-

perature, and comparison of residual of estimated and computed temperature in
each case.
Fig. 12. Comparison of temperature measured and computed with planned
and forecasted conditions, upper limit and lower limit.

 
Fig. 13. Estimated effective wind speed ϑ̂ and emissivity (ˆs ) for each
case.

V. CONCLUSIONS Fig. 15. Comparison between the average temperature predicted 15 min be-
fore and temperature measured, comparison between residual of predicted aver-
This paper presents an algorithm to estimate and predict age temperature and temperature computed in each case, and maximum current
thermal transient states in OHL conductors and addresses its intensity allowable until the conductor reaches 75 [◦ C].
implementation. This algorithm uses an EKF based on the heat
transfer equation, using atmospheric conditions, current inten- Nevertheless, models of wind behavior for long time emergency
sity, conductor parameters and direct measurements as inputs. could be included in future studies.
The uncertainty in these values was considered. To simulate The algorithm assessment showed a reduction in the
and test the EKF, the algorithm estimated and predicted values RMEe and RMS when thermal estimation and prediction are
of average conductor temperature, with processing times lower carried out by the proposed EKF, allowing to increase the re-
than the time spent between measurement samples, showing liability in the thermal monitoring of OHLs. For instance, de-
computational efficiency and stability. The algorithm can be spite using the most critical case, the RMSe obtained using the
directly implemented on current DLR systems in a fast and algorithm to estimate and predict the average conductor tem-
cost-effective way. perature was less than the RMSe obtained in all cases, both
Average effective wind speed, emissivity and solar absorptiv- the simulations and the experiment. The algorithm validation
ity were chosen as parameters to be estimated, due to the im- was performed using low wind speeds which is considered a
pact of their uncertainty on heat transfer. Effective wind speed critical scenario. In the cases of higher wind speeds and low
was assumed constant during a typical time of contingency. current intensities where the conductor temperature is close to

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 2244 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 33, NO. 5, OCTOBER 2018

the ambient temperature, a field test validation is necessary. Fi- [18] E. Carlini, C. Pisani, A. Vaccaro, and D. Villacci, “A reliable com-
nally, further analysis should be carried out using data validation puting framework for dynamic line rating of overhead lines,” Elect.
Power Syst. Res., vol. 132, pp. 1–8, Mar. 2016. [Online]. Available:
techniques. http://linkinghub.elsevier.com/retrieve/pii/S0378779615003302
[19] D. L. Alvarez, F. Faria da Silva, E. E. Mombello, C. L. Bak, J. A. Rosero,
and D. L. Ólason, “An approach to dynamic line rating state estimation at
REFERENCES thermal steady state using direct and indirect measurements,” Elect. Power
Syst. Res., vol. 163, Part B, pp. 599–611, Oct. 2018. [Online]. Available:
[1] CIGRE WG B2.13, “Guidelines for increased utilization of existing over- https://www.sciencedirect.com/science/article/pii/S0378779617304595
head transmission lines,” CIGRE, Paris, CIGRE WG B2.13, Technical [20] D. Simon, “Nonlinear Kalman filtering,” in Opt. State Estimation
Brochure 353, 2008. Kalman, H Infinity, Nonlinear Approaches. Hoboken, NJ, USA: Wi-
[2] D. Douglass et al., “Real-Time overhead transmission-line monitor- ley, May 2006, ch. 13, pp. 395–426. [Online]. Available: http://doi.
ing for dynamic rating,” IEEE Trans. Power Del., vol. 31, no. 3, wiley.com/10.1002/0470045345
pp. 921–927, Jun. 2016. [Online]. Available: http://ieeexplore.ieee. [21] CIGRE WG B2.42, “Guide for thermal rating calculations of overhead
org/document/6991585/ lines,” CIGRE, Paris, CIGRE WG B2.42, Technical Brochure 601, 2014.
[3] D. Koval and R. Billinton, “Determination of transmission line ampac- [22] V. T. Morgan, Thermal Behaviour of Electrical Conductors: Steady, Dy-
ities by probability and numerical methods,” IEEE Trans. Power App. namic, and Fault-current Ratings. Taunton, Somerset, England: Research
Syst., vol. PAS-89, no. 7, pp. 1485–1492, Sep. 1970. [Online]. Available: Studies Press, 1991.
http://ieeexplore.ieee.org/document/4074224/ [23] IEEE Standard for Calculating the Current-Temperature of Bare Over-
[4] I. Cotton and J. Teh, “Critical span identification model for dynamic head Conductors, IEEE Std 738-2006, 2007.
thermal rating system placement,” IET Generation, Transmiss. Dis-
trib., vol. 9, no. 16, pp. 2644–2652, Dec. 2015. [Online]. Available:
http://digital-library.theiet.org/content/journals/10.1049/iet-gtd.2015
.0601 David L. Alvarez was born in El Cocuy, Boyaca,
[5] M. Matus et al., “Identification of critical spans for monitoring sys- Colombia, in 1987. He received the B.Sc, M.Sc. and
tems in dynamic thermal rating,” IEEE Trans. Power Del., vol. 27, Ph.D. degrees in electrical engineering from the Uni-
no. 2, pp. 1002–1009, Apr. 2012. [Online]. Available: http://ieeexplore. versidad Nacional de Colombia, Bogotá, Colombia,
ieee.org/document/6163401/ in 2009, 2014, and 2017, respectively. From 2009 to
[6] J. L. Aznarte and N. Siebert, “Dynamic line rating using numerical 2012, he was with Siemens Manufacturing, Tenjo,
weather predictions and machine learning: A case study,” IEEE Trans. Colombia developing new technologies and tools for
Power Del., vol. 32, no. 1, pp. 335–343, Feb. 2017. [Online]. Available: distribution transformers. He is currently an inde-
http://ieeexplore.ieee.org/document/7442844/ pendent Consultant. His research interests include
[7] CIGRE WG B2.36, “Guide for application of direct real-time monitoring transformers, computational electromagnetics, and
systems,” CIGRE, Paris, CIGRE WG B2.36, Technical Brochure 498, dynamic line rating.
2012.
[8] A. Michiorri et al., “Forecasting for dynamic line rating,” Renewable. Sus-
tain. Energy Rev., vol. 52, pp. 1713–1730, Dec. 2015. [Online]. Available: Filipe Faria da Silva was born in Portugal in 1985.
http://linkinghub.elsevier.com/retrieve/pii/S1364032115007819 He received the M.Sc. degree in electrical and com-
[9] I. Albizu, E. Fernandez, P. Eguia, E. Torres, and A. J. Mazon, “Ten- puters engineering from Instituto Superior Técnico,
sion and ampacity monitoring system for overhead lines,” IEEE Trans. Lisbon, Portugal, in 2008, and the Ph.D. degree in
Power Del., vol. 28, no. 1, pp. 3–10, Jan. 2013. [Online]. Available: http:// electric power systems from Aalborg University, Aal-
ieeexplore.ieee.org/document/6313952/ borg, Denmark, in 2011. From 2008, he was with
[10] C. R. Black and W. A. Chisholm, “Key considerations for the se- EDP-Labelec and from 2008 to 2011 was with the
lection of dynamic thermal line rating systems,” IEEE Trans. Power Danish TSO Energinet. He is currently an Associate
Del., vol. 30, no. 5, pp. 2154–2162, Oct. 2015. [Online]. Available: Professor with the Department of Energy Technology,
http://ieeexplore.ieee.org/document/6967802/ Aalborg University, where he is also Semester Coor-
[11] I. Albizu, E. Fernandez, A. Mazon, and J. Bengoechea, “Influ- dinator for the Electrical Power System and High
ence of the conductor temperature error on the overhead line am- Voltage Engineering master program and Vice Leader of the Modern Power
pacity monitoring systems,” IET Generation, Transmiss. Distrib., Transmission Systems research program. He has supervised 8+ Ph.D. students
vol. 5, no. 4, p. 440, 2011. [Online]. Available: http://digital-library. and dozens of Master students in these areas, as well as authored more than
theiet.org/content/journals/10.1049/iet-gtd.2010 .0470 100 articles and a book (as of April 2018). His research interests include power
[12] CIGRE WG B2.12, “Guide for selection of weather parameters for bare cables, electromagnetic transients, system modelling, network stability, HVDC
overhead conductor ratings,” CIGRE, Paris, CIGRE WG B2.12, Technical transmission and HV phenomena. He is an active Member of Cigré, currently
Brochure 299, 2006. the head of Denmarks IEEE Power and Energy Society, the Danish Representa-
[13] A. Piccolo, A. Vaccaro, and D. Villacci, “Thermal rating assessment of tive for Cigré SC C4 System Technical Performance, and the Convener of Cigré
overhead lines by Affine Arithmetic,” Elect. Power Syst. Res., vol. 71, WG C4.46.
no. 3, pp. 275–283, Nov. 2004. [Online]. Available: http://linkinghub.
elsevier.com/retrieve/pii/S0378779604000677
[14] M. A. Bucher and G. Andersson, “Robust corrective control mea-
sures in power systems with dynamic line rating,” IEEE Trans. Power Enrique E. Mombello (M’95–SM’00) was born in
Syst., vol. 31, no. 3, pp. 2034–2043, May 2016. [Online]. Available: Buenos Aires, Argentina, in 1957. He received the
http://ieeexplore.ieee.org/document/7163367/ B.S. and Ph.D. degrees in electrical engineering from
[15] A. Polevoy, “Impact of data errors on sag calculation accuracy Universidad Nacional de San Juan, San Juan, Ar-
for overhead transmission line,” IEEE Trans. Power Del., vol. 29, gentina, in 1982 and 1998, respectively. From 1989 to
no. 5, pp. 2040–2045, Oct. 2014. [Online]. Available: http://ieeexplore. 1991, He was with the High Voltage Institute, RWTH
ieee.org/document/6828800/ Aachen University, Aachen, Germany. He is currently
[16] E. Carlini, G. Giannuzzi, C. Pisani, A. Vaccaro, and D. Villacci, “Ex- a Principal Researcher with the National Council of
perimental deployment of a self-organizing sensors network for dynamic Technical and Scientific Research, Buenos Aires, Ar-
thermal rating assessment of overhead lines,” Elect. Power Syst. Res., gentina, and a Lecturer with the Instituto de Energa
vol. 157, pp. 59–69, Apr. 2018. [Online]. Available: http://linkinghub. Elctrica, National University of San Juan, Buenos
elsevier.com/retrieve/pii/S0378779617304789 Aires, Argentina. He has more than 30 years of experience in research projects.
[17] A. Michiorri, P. C. Taylor, and S. C. E. Jupe, “Overhead line His research interests include the development of high-frequency models of
real-time rating estimation algorithm: Description and validation,” transformers, electrical transients and resonance processes within transformers,
Proc. Institution Mech. Eng., A, J. Power Energy, vol. 224, design and diagnostics of power transformers, asset management, transformer
no. 3, pp. 293–304, May 2010. [Online]. Available: http://journals. life management, electromagnetic transients in power systems, corona losses in
sagepub.com/doi/10.1243/09576509JPE859 overhead transmission lines, and low-frequency electromagnetic fields.

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.
 ALVAREZ et al.: CONDUCTOR TEMPERATURE ESTIMATION AND PREDICTION AT THERMAL TRANSIENT STATE 2245

Claus Leth Bak was born in Rhus, Denmark, on Javier A. Rosero was born in Potosi, Colombia, in
April 13, 1965. He received the B.Sc. degree (honors) 1978. He received the B.Sc degree in electrical engi-
in electrical power engineering, in 1992, the M.Sc. neering from the Universidad of Valle, Cali, Colom-
degree in electrical power engineering from the De- bia, in 2002, and the Ph.D. degree from the Technical
partment of Energy Technology, Aalborg University, University of Catalonia, Barcelona, Spain, in 2007.
Aalborg, Denmark, in 1994, and the Ph.D. degree in From 2002 to 2004, he worked for construction and
2015 with the thesis EHV/HV underground cables in maintenance of power systems and substations, Bo-
the transmission system. After his studies, he worked gota, Colombia. From 2007 to 2009, he worked with
as a Professional Engineer in electric power trans- Asea Brown Boveri, Barcelona, Spain providing ser-
mission and substations with specializations within vice and technical support for drives for industrial
the area of power system protection with NV Net and windmill systems. He is currently an Associate
Transmission Company. In 1999, he was an Assistant Professor with the De- Professor with the Department of Electrical and Electronic Engineering, Uni-
partment of Energy Technology, Aalborg University, Aalborg, Denmark, where versidad Nacional de Colombias, Bogota, Colombia. He has published more
he is currently a Full Professor. He has supervised/cosupervised +35 Ph.D. and than 50 referred journal and conference papers. His research interests include
+50 M.Sc. thesis. He is the author/coauthor of approximately 240 publications. the areas of modelling, diagnosis and control of electrical machines and drives,
His research interests include corona phenomena on overhead lines, power sys- electric mobility, and smart grids. He was the recipient of the IEEE AESS Harry
tem modeling and transient simulations, underground cable transmission, power Rowe Mimno Award for excellence in technical communications for 2007 from
system harmonics, power system protection, and HVDC-VSC offshore trans- the Aerospace and Electronic Systems Society IEEE 2007. He is a Member of
mission networks. He is a Member of CIGRE JWG C4-B4.38, CIGRE SC C4, the Institute of Electrical and Electronics Engineers, International Society of
and SC B5 study committees, and Danish CIGRE National Committee. He was Automation, and ISA.
the recipient of the DPSP 2014 best Paper Award and the PEDG 2016 best
Paper Award. He was the Head of the Energy Technology Ph.D. program (+100
Ph.D.s) and the Head of the Section of Electric Power Systems and High Voltage
in AAU, and a Member of the Ph.D. Board at the Faculty of Engineering and
Science.

Authorized licensed use limited to: ULAKBIM UASL - Suleyman Demirel Universitesi. Downloaded on December 17,2024 at 10:27:30 UTC from IEEE Xplore. Restrictions apply.