CN107193992A - A kind of 220kV main transformer condition evaluation prediction methods based on decision Tree algorithms - Google Patents

Abstract

A kind of 220kV main transformer condition evaluation prediction methods based on decision Tree algorithms, using online monitoring data, decision making algorithm training, main transformer state estimation with judging, pass through data fusion, using decision making algorithm technology, the various online monitoring datas of 220KV main transformers are concentrated and merged, collect to form valuable data, strong prediction data and means support are provided for 220kV main transformer state estimations, the security reliability of lifting means brings the progress of main transformer condition evaluation prediction mode.

Description

A kind of 220kV main transformer condition evaluation prediction methods based on decision Tree algorithms

Technical field

It is a kind of the present invention relates to the main transformer condition evaluation prediction in power network and the technical method of monitoring real time data fusion To the method for power network 220kV main transformer condition evaluation predictions.

Background technology

At present, in the online monitoring data that 220KV main transformers equipment is gathered, substantial amounts of equipment operation real time information fails Utilize very well, for the real time data of these equipment, operation maintenance personnel (or monitoring personnel) grasps less than or grasped inaccurate, to this The grasp of a little fine or not situations of equipment running status, mainly by operation maintenance personnel (or monitoring personnel) artificial experience, to unwatched Equipment state, can neither accomplish monitoring in real time, substantial amounts of people's material resources are spent again to on-site verification.So research sets up special 220KV main transformer device status datas, decision Tree algorithms technology is truly realizing the prediction of 220KV main transformer equipment states Management.

The content of the invention

The purpose of the present invention is precisely in order to overcoming the defect of above-mentioned prior art presence and providing one kind and determined based on big data The 220kV main transformer condition evaluation prediction methods of plan algorithm, collection of the present invention based on on-line monitoring information, integrate and deposit Storage.

From the point of view of the collection of information, database needs to access different online monitoring datas.It is right after data access database The data of access will be analyzed, screened, being cleaned, characteristic processing etc., so as to be the standard of the foundation of algorithm and model, algorithm computing True property provides genuine and believable data.To achieve these goals, the present invention provides following technical scheme：

A kind of 220kV main transformer condition evaluation prediction methods based on decision Tree algorithms of the present invention, comprise the following steps：

(1) number, is monitored on-line using the oil chromatography hydrogen components content online monitoring data and oil temperature of 220KV main transformers According to storing to database, the basis of decision making algorithm is used as；

(2) the main transformer online monitoring data that taken at regular intervals is arrived, is stored using cycle timing method, decision making algorithm is used as Training sample set, is trained using decision tree；

(3), the result of training is the state of next timing cycles of 220KV main transformers, i.e., whether normal judgement.

In above-mentioned technical proposal：

(1) data sample is acquired, screens, analyzes, handled

Using existing on-Line Monitor Device gathered data, extracted with fixed interval (such as 15 minutes being), screen hydrogen Content (H2), oil temperature, as training sample basic data, while adding each time interval with interval time (such as 15 minutes) Main transformer state (normal or abnormal, just commonly using 1 and representing, abnormal to be represented with 0) information to training sample database, with hydrogen content (H2), Oil temperature together constitutes the basic sample data of Algorithm for Training (such as 100 sample or more).Then to training sample data at Reason, handles exceptional value, and outlier processing method uses average value complementing method.Flow is shown in Fig. 1；

(2) samples decision Tree algorithms

1) sample data

Sample data after treatment, is formed the sample data that can be calculated with decision Tree algorithms, including Oil temperature height (H), low (L) data value, hydrogen content (H2) data value, equipment state (normal 1, abnormal 0) value.Sample data form Such as table 1 below：

Name variable

Data 1

Data 2

Data 3

Data 4

…

Data 100

Oil temperature (H, L)

H

L

L

L

…

L

H2 contents

183.3

201

196.2

229

190.4

Equipment state (1,0)

0

1

1

0

…

1

Table I sample data form

2) decision Tree algorithms model

Decision tree is a kind of simple but widely used sorting algorithm.Decision tree is built by training data, efficiently Unknown data are classified.Decision-making number has two big advantages：

3) decision Tree algorithms step

The step of decision tree builds：

When a. starting, the record (oil temperature, hydrogen content (H2)) of all variables is regarded as a node；

B. the best cut-point of each variable (hydrogen content (H2), oil temperature) is found；

C. two nodes are divided into：Left and right；

D. two nodes of left and right are performed respectively, untill each node meets requirement.

4) decision Tree algorithms stop condition

The structure of decision tree is a recursive process, it is thus necessary to determine that stop condition, else process will not terminate.To a left side Performed respectively with right node, untill each node meets requirement.Intuitive way is when each child node only has a species Stop during the record of type.

(3), 220KV Transformer State Assessments are predicted

Made using on-Line Monitor Device collection with the hydrogen content (H2) of fixed interval (such as 15 minutes), oil temperature data For sample, calculating is trained to sample data with the recursive procedure of decision Tree algorithms.Until showing that the correct division of sample is (straight The mode of sight is stopped when each child node only has the final record of a type).The unique type of this node is exactly The state of 220kV main transformers, normal (1) or abnormal (0).By decision Tree algorithms, 220kV main transformer state estimations are lifted Real-time and foresight.

The beneficial effects of the invention are as follows change 220kV main transformers state estimation and rely on traditional data statistical analysis and people The deficiency of work empirical method, changes the information delay situation of 220kV main transformer equipment state ex-post analyses.Utilize in-site collecting 220kV main transformer online monitoring datas, timely anticipation 220kV main transformer running statuses support 220kV main transformer states The real-time and validity of assessment are realized, lift the security reliability of 220kV main transformers equipment operation.

Brief description of the drawings

Fig. 1 is flow chart of data processing figure of the present invention；

Fig. 2 is decision-tree model figure.

Embodiment

See Fig. 1, Fig. 2, a kind of 220kV main transformer condition evaluation prediction methods based on decision Tree algorithms of the present invention, bag Include following steps：

(1) number, is monitored on-line using the oil chromatography hydrogen components content online monitoring data and oil temperature of 220KV main transformers According to storing to database, the basis of decision making algorithm is used as；

(2) the main transformer online monitoring data that taken at regular intervals is arrived, is stored using cycle timing method, decision making algorithm is used as Training sample set, is trained using decision tree；

(3), the result of training is the state of next timing cycles of 220KV main transformers, i.e., whether normal judgement.

Embodiment

1) the 220kV main transformer hydrogen contents of fixed interval (such as 15 minutes) are obtained using on-Line Monitor Device (H2), oil temperature, state (1 or 0) information, are used as training sample basic data.

2) abnormality processing is carried out to training sample data using average value complementing method.Obtain that decision Tree algorithms can be inputted The authentic data calculated, including oil temperature high (H), low (L) data value, hydrogen content (H2) data value, equipment state are (normal 1, abnormal 0) value.

3) sample data input decision tree is subjected to recursive calculation, left and right node performed respectively, until each node Meet untill requiring.Intuitive way is stopped when each child node only has a type of record.

4) unique type of node is obtained to the state of 220kV main transformers, normal (1) or abnormal (0) as output.

Claims (1)

1. a kind of 220kV main transformer condition evaluation prediction methods based on decision Tree algorithms, its feature is being, including following step Suddenly：

(1), deposited using the oil chromatography hydrogen components content online monitoring data and oil temperature online monitoring data of 220KV main transformers Storage is used as the basis of decision making algorithm to database；

(2) the main transformer online monitoring data that taken at regular intervals is arrived, is stored using cycle timing method, the training of decision making algorithm is used as Sample set, is trained using decision tree；

(3), the result of training is the state of next timing cycles of 220KV main transformers, i.e., whether normal judgement.