You are on page 1of 1

Objective: To estimate the residual age of transformer.

Introduction: One of the costliest assets in an electrical network is transformer. Life of a transformer in service
is expected to be in range of 25-30 years. But in actual, factors such as high moisture ingress, harmonic distortion,
degradation of insulation both oil and paper, abnormal heating of transformer oil and frequent breakdowns
reduces the effective life of transformer.

Estimation of residual age of a transformer will help the utility to take better decision w.r.t CAPEX planning,
inventory management and reliability improvement of the network.

Proposed model to estimate remaining age of transformer:

Data Collection Training ML Model Calculating remaining age

1. Data Collection: Historical DT failure data needs to be collected:


i. Dissolved gases concentration, BDV, Furan, Physical parameters, IR, ER, Tan Delta values and age of
DT at time of failure.
ii. Number of breakdowns, no. of hotspots and count of various electrical faults etc. faced by DT till its
failure.

All these above parameters of DT will act as predictor/features for the model. Now, assuming that the
standard age of DT is 25 years:

% age of DT till failure (Y) = [(Age of DT at time of failure in yrs.) / 25] x 100

So, let’s suppose we have ‘n’ number of DT failure data records then for each of the DT failed we will
calculate % age of DT till failure hence,

Y = {y1, y2, y3, ……., yn} i.e., response

X is a (n x p) matrix where n = no. of records and p = no. of features

2. Training ML Model: As Y is continuous in nature, we have a regression problem at hand. There are various
supervised machine learning models to choose from such as Random Forest, Ada Boost or XG Boost etc.
After all the basic preprocessing of dataset and feature engineering, we will split our data (X,Y) in training
data (x_train, y_train) and testing data (x_test, y_test) in the ratio of 80:20.
x_train = 0.8(X) y_train = 0.8(Y) ; x_test = 0.2(X) y_test = 0.2(Y)

While training the ML model with (x_train as i/p and y_train as o/p), we tell the model that we have input
and corresponding output, and ask the model to find a relationship between input and output so that for
unseen inputs it can predict the output based on this relationship.

3. Calculating remaining age: After training of model, x_test which is nothing but 20% of total historical DT
failure data collected will be fed to the model.

x_test Trained Model y_pred

y_pred here is the prediction or % age of DT till failure for each of the record in x_test as predicted by the
model. And hence, estimated remaining age of DT can be calculated as:

Residual age (in years) = [(Standard age of 25 years) x (% age of DT till failure as predicted by model)] –
Current Age of DT

You might also like