CN103886374A - Cable joint wire temperature prediction method based on RBF neural network - Google Patents

Abstract

The invention relates to a cable joint wire temperature prediction method based on an RBF neural network. The cable joint wire temperature prediction method based on the RBF neural network mainly includes the first step of collection of sample data, wherein factors such as environment humidity, environment temperature, sheath temperature, joint insulation temperature, contact temperature and various surface temperatures which are related with the cable joint wire temperature are measured in real time; the second step of network training, wherein the collected data in the step (1) are preprocessed, training data and prediction data are divided, various parameters are set, a network is built, and the data are predicted finally. As the neural network technology is applied to predication of the cable joint wire temperature, the cable joint wire temperature can be well monitored in a real-time and on-line mode, and faults can be well analyzed.

Description

A kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network

Technical field

The invention belongs to temperature of cable junction value prediction technology, particularly a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network.

Background technology

Power cable is very important equipment in electric system, once break down, littlely causes user to have a power failure for a long time, greatly may cause cable line associate device chain-react and break down, and even causes the paralysis of distribution system part.Cable splice is the critical piece that cable line breaks down, therefore, very necessary to the research of cable splice, the Various types of data gathering for temperature monitoring terminal, we are necessary to analyze the duty of cable splice, and judge whether certain cable splice has occurred fault, to find the potential risk of cable and make timely and processing, reduce the loss causing due to cable fault, improve the reliability of power supply.

But high-tension cable on-line monitoring research mainly concentrates on the aspects such as the aging and defect detection of cable insulation, its Main Means has two kinds of Partial Discharge Detection and on-line temperature monitorings.Wherein, the gordian technique difficult point that implement Partial Discharge Detection is to exist how at the scene the sensitivity that improves Partial Discharge Detection equipment in the situation of larger electromagnetic field.Generally speaking, in order to meet actual requirement of engineering, need to adopt high-precision high frequency partial electric discharge and very high frequency(VHF) part to put detector, its core technology is not yet broken through at home, therefore corresponding checkout equipment all need to be from external import, and equipment purchase expense and following maintenance cost are all quite high.Except discharge examination, temperature detection is the most approved important means in cable monitoring both at home and abroad at present, because no matter be that the aging of cable causes Leakage Current increase or overload to cause loss to increase, the form that capital raises by temperature embodies, and temperature rising is also the key factor that causes cable fault.Therefore,, if can implement online detection to cable operating temperature, can accomplish real-time monitoring to the safe operation of cable completely.

Artificial neural network is a kind of animal nerve network behavior feature of imitating, carry out the algorithm mathematics model of distributed parallel information processing, there is powerful pattern-recognition and data fitting ability, dissimilar neural network is suitable for processing different problems, for example, self-organizing network is applicable to solve clustering problem, and generalized regression network is applicable to fitting problems.This network relies on the complexity of system, by adjusting interconnective relation between inner great deal of nodes, thereby reaches the object of process information.

RBF network be a kind of simple in structure, fast convergence rate, can Approximation of Arbitrary Nonlinear Function network.Broomhead in 1988 and Lowe have the principle of local acknowledgement according to biological neuron, radial basis function is introduced in neural network.RBF network is proved to be has uniformly approximated performance to nonlinear network, has now been widely used in the fields such as time series analysis, pattern-recognition, nonlinear Control and image processing.

It is the non-linear process of a more complicated that temperature of cable junction value changes, and artificial neural network is showing good performance aspect processing nonlinear pattern recognition.Multilayer perceptron is most widely used one wherein, but the error backpropagation algorithm that multilayer perceptron adopts is easily absorbed in local minimum, and training time that need to be longer, its transition function can only be exported two kinds of possible values, the linear operation rule of multilayer perceptron has determined that it can only solve the problem of linear separability, has significant limitation while therefore processing problem.By contrast, radial basis function network has the non-linear mapping capability comparing favourably with multilayer perceptron, and has higher convergence precision and travelling speed.

Summary of the invention

The present invention is mainly that the existing laying of solution prior art is difficult, cost is high, Maintenance Difficulty, and most detection can not accurately judge the technical matters of heating cause and cable splice running status etc.; A kind of energy real-time estimate cable splice conductor temperature is provided, and can be by the variation tendency of conductor temperature being gone to judge a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network of cable splice in the working condition in following moment.

Above-mentioned technical matters of the present invention is mainly solved by following technical proposals:

A cable splice conductor temperature Forecasting Methodology based on RBF neural network, is characterized in that, comprises the following steps:

Step 1, respectively between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along at least three temp probes are circumferentially evenly set; Edge, external contacts place at cable splice circumferentially evenly arranges at least two temp probes; And an ambient humidity meter and an environment temperature; Then use data collection station according to the time interval collecting temperature data of setting; Obtain 11 kinds of temperature variables that affect cable splice conductor temperature; Respectively: temperature, cable skin low order end temperature between temperature, cable skin high order end temperature, cable skin middle part temperature, cable skin low order end and middle part between ambient humidity, right sheath temperature, joint insulation course temperature, left sheath temperature, environment temperature, external contacts temperature, cable skin high order end and middle part;

Step 2, the data that gather based on step 1 are carried out establishment and the prediction of RBF network, specifically: in the data importing matlab first step 1 being gathered, carry out pre-service, then data after pre-service are divided into two large divisions, a part is training data, another part is predicted data, and described training data and predicted data comprise and comprise respectively some groups of data, and every group of data comprise 11 kinds of temperature variables that affect cable splice conductor temperature; Training data is input in three layers of RBF neural network and trains the RBF network that obtains training, then predicted data is input in RBF network and is predicted, obtain the data of prediction.

In above-mentioned a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network, describedly training data be input to the concrete steps of training in three layers of RBF neural network be:

Step 1, asks for Basis Function Center t based on K-means clustering algorithm _i(n); Be set with I cluster centre, i cluster centre of the n time iteration is t _i(n), i=1,2 ..., I, I value is rule of thumb to determine here, then carries out the operation of following sub-step:

Step 1.1, netinit.From the input sample data that comprises 11 temperature variable values, the random individual different sample of I of selecting is as initial cluster centre t _i(0) (i=1,2 ..., I).

Step 1.2, the training sample set of input is divided into groups by arest neighbors rule: calculate this training sample nearest apart from which cluster centre, just it is classified as to the same class of this cluster centre, calculate:

i(X _p)＝argmin||x _p-t _i(0)||

Find corresponding i value, by X _pbe classified as i class.

In formula,

be p input sample; P=1,2,3 ... P, P is total sample number; T _ifor the center of network hidden layer node.

Step 1.3, upgrades cluster centre.Calculate each cluster set θ _pthe mean value of middle training sample, i.e. new cluster centre t _i:

In formula, η is Learning Step, 0< η <1.

In the time that cluster centre no longer changes, the t obtaining _ibe the final Basis Function Center of RBF neural network, continue iteration otherwise return to (2).

Step 2, solves variances sigma _i; The basis function of the RBF neural network that should predict based on cable splice conductor temperature is Gaussian function, variances sigma _ican solve as follows:

${\mathrm{\&sigma;}}_I=\frac{c_{\max }}{2I},(i=\mathrm{1,2},...,I)$

In formula, c _maxthe ultimate range between selected center.

Step 3, calculates the RBF god based on cable splice who creates by the weights between the hidden layer of network and output layer.

Hidden layer can directly calculate by least square method to neuronic connection weights between output layer, and computing formula is as follows:

$\mathrm{\&omega;}=\exp \left(\frac{I}{c_{\max }^2}{\left|\right|x_p-t_i\left|\right|}^2\right),i=\mathrm{1,2},...,I;p=\mathrm{1,2},...,P$

Step 4, center and the variance of trying to achieve based on step 1, step 2 are expressed radial basis function:

$R(x_p-t_i)=\exp (-\frac{1}{2{\mathrm{\&sigma;}}^2}{\left|\right|x_p-t_i\left|\right|}^2)$

In formula, || x _p-t _i|| be European norm; t _ifor the center of Gaussian function; σ is the variance of Gaussian function.The known network of structural drawing by RBF neural network is output as

$y_i=\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_{\mathrm{ij}}(-\frac{1}{2{\mathrm{\&sigma;}}^2}{\left|\right|x_p-t_i\left|\right|}^2),j=\mathrm{1,2},...,n$

In formula, ω _ijfor hidden layer is to the connection weights of output layer; I=1,2,3...I, I is hidden layer node number; y _ifor with the actual output of j output node of the network of input sample, i.e. the conductor temperature value of cable splice.

In above-mentioned a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network, respectively between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along three temp probes are circumferentially evenly set; Edge, external contacts place at cable splice circumferentially evenly arranges two temp probes; And an ambient humidity meter and an environment temperature; Then use data collection station according to the time interval collecting temperature data of setting; Obtain 11 kinds of temperature variables that affect cable splice conductor temperature; Respectively: temperature, cable skin low order end temperature between temperature, cable skin high order end temperature, cable skin middle part temperature, cable skin low order end and middle part between ambient humidity, right sheath temperature, joint insulation course temperature, left sheath temperature, environment temperature, external contacts temperature, cable skin high order end and middle part.

In above-mentioned a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network, in described step 2, image data imports carries out pretreated concrete grammar in matlab and is:

For the data of Real-time Collection, first remove more above owing to operating, measuring the inaccurate misdata of collecting temperature value former thereby that cause;

Then, between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along three temp probes are circumferentially evenly set, be that synchronization same position has gathered three temperature values, for reducing measuring error, by average data identical monitoring location, be about to mutually in the same time, three temperature values of same position are added and again divided by 3; Edge, external contacts place at cable splice circumferentially evenly arranges two temp probes; Be that synchronization same position has gathered three temperature values, for reducing measuring error, by average data identical monitoring location, be about to mutually in the same time, three temperature values of same position are added and again divided by 2.

Therefore, tool of the present invention has the following advantages: 1 the inventive method does not need to solve complicated pluralism likelihood equations; 2 the inventive method can directly obtain conductor temperature value, have avoided, by various approximate treatment and the error of deriving and producing, having improved precision of prediction; 3 the inventive method, compared with existing BP Neural Network Temperature value prediction, have significantly improved precision of prediction.

Embodiment

Below by embodiment, technical scheme of the present invention is described in further detail.

Embodiment:

The present invention mainly comprises the following steps:

Step 1, respectively between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along three temp probes are circumferentially evenly set; Edge, external contacts place at cable splice circumferentially evenly arranges two temp probes; And an ambient humidity meter and an environment temperature; Then use data collection station according to the time interval collecting temperature data of setting; Obtain 11 kinds of temperature variables that affect cable splice conductor temperature; Respectively: temperature, cable skin low order end temperature between temperature, cable skin high order end temperature, cable skin middle part temperature, cable skin low order end and middle part between ambient humidity, right sheath temperature, joint insulation course temperature, left sheath temperature, environment temperature, external contacts temperature, cable skin high order end and middle part;

Step 2, the data that gather based on step 1 are carried out establishment and the prediction of RBF network, specifically: in the data importing matlab first step 1 being gathered, carry out pre-service, then data after pre-service are divided into two large divisions, a part is training data, another part is predicted data, and described training data and predicted data comprise and comprise respectively some groups of data, and every group of data comprise 11 kinds of temperature variables that affect cable splice conductor temperature; Training data is input in three layers of RBF neural network and trains the RBF network that obtains training, then predicted data is input in RBF network and is predicted, obtain the data of prediction;

Pretreated method for predicting is: for the data of Real-time Collection, first remove more above owing to operating, measuring the inaccurate misdata of collecting temperature value former thereby that cause;

Then, between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along three temp probes are circumferentially evenly set, be that synchronization same position has gathered three temperature values, for reducing measuring error, by average data identical monitoring location, be about to mutually in the same time, three temperature values of same position are added and again divided by 3; Edge, external contacts place at cable splice circumferentially evenly arranges two temp probes; Be that synchronization same position has gathered three temperature values, for reducing measuring error, by average data identical monitoring location, be about to mutually in the same time, three temperature values of same position are added and again divided by 2.

Training data is input to the concrete steps of training in three layers of RBF neural network is:

Step 1, asks for Basis Function Center t based on K-means clustering algorithm _i(n); Be set with I cluster centre, i cluster centre of the n time iteration is t _i(n), i=1,2 ..., I, I value is rule of thumb to determine here, then carries out the operation of following sub-step:

Step 1.1, netinit.From the input sample data that comprises 11 temperature variable values, the random individual different sample of I of selecting is as initial cluster centre t _i(0) (i=1,2 ..., I).

Step 1.2, the training sample set of input is divided into groups by arest neighbors rule: calculate this training sample nearest apart from which cluster centre, just it is classified as to the same class of this cluster centre, calculate:

i(X _p)＝argmin||x _p-t _i(0)||

Find corresponding i value, by X _pbe classified as i class.

In formula,

be p input sample; P=1,2,3 ... P, P is total sample number; Ti is the center of network hidden layer node.

Step 1.3, upgrades cluster centre.Calculate each cluster set θ _pthe mean value of middle training sample, i.e. new cluster centre t _i:

In formula, η is Learning Step, 0< η <1.

In the time that cluster centre no longer changes, the t obtaining _ibe the final Basis Function Center of RBF neural network, continue iteration otherwise return to (2).

Step 2, solves variances sigma _i; The basis function of the RBF neural network that should predict based on cable splice conductor temperature is Gaussian function, variances sigma _ican solve as follows:

${\mathrm{\&sigma;}}_I=\frac{c_{\max }}{2I},(i=\mathrm{1,2},...,I)$

In formula, c _maxthe ultimate range between selected center.

Step 3, calculates the RBF god based on cable splice who creates by the weights between the hidden layer of network and output layer.

Hidden layer can directly calculate by least square method to neuronic connection weights between output layer, and computing formula is as follows:

$\mathrm{\&omega;}=\exp \left(\frac{I}{c_{\max }^2}{\left|\right|x_p-t_i\left|\right|}^2\right),i=\mathrm{1,2},...,I;p=\mathrm{1,2},...,P$

Step 4, center and the variance of trying to achieve based on step 1, step 2 are expressed radial basis function:

$R(x_p-t_i)=\exp (-\frac{1}{2{\mathrm{\&sigma;}}^2}{\left|\right|x_p-t_i\left|\right|}^2)$

In formula, || x _p-t _i|| be European norm; t _ifor the center of Gaussian function; σ is the variance of Gaussian function.The known network of structural drawing by RBF neural network is output as

$y_i=\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_{\mathrm{ij}}(-\frac{1}{2{\mathrm{\&sigma;}}^2}{\left|\right|x_p-t_i\left|\right|}^2),j=\mathrm{1,2},...,n$

In formula, ω _ijfor hidden layer is to the connection weights of output layer; I=1,2,3...I, I is hidden layer node number; y _ifor with the actual output of j output node of the network of input sample, i.e. the conductor temperature value of cable splice.

Finally, can utilize the cable temperature value in above-mentioned data, use the method for the conductor temperature value in 6 these moment of conductor temperature value prediction before in certain moment, create Elman network prediction phase conductor temperature value in the same time, prove that the predicted value of above-mentioned RBF network is correct.

The object of the present invention is to provide a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network, and use Elman neural net method to prove the correctness of said method predicted data, solve Traditional calculating methods setting up very complicated practical problems in model and solving equation group, solve that the predicated error existing in the cable splice conductor temperature Forecasting Methodology based on BP neural network is large, speed of convergence slowly, is easily absorbed in the problems such as local minimum.

Be prerequisite mainly with wire transmission greatly to the monitoring method of cable temperature both at home and abroad at present, all have the problems such as laying is difficult, cost is high, Maintenance Difficulty, and detection can not accurately judge heating cause and cable splice running status mostly.Energy real-time estimate cable splice conductor temperature of the present invention, and can be by the variation tendency of conductor temperature being gone to judge that cable splice is in the working condition in following moment, this has important realistic meaning for improving cable safety in operation and stability.

The impact of random initializtion network weight and threshold value when the precision of prediction of BP neural network is wound establishing network, it is all random making the precision that network predicts the outcome at every turn, even if determined a better BP network structure of a prediction effect, the result of prediction also likely can be very poor next time; RBF neural network just has greatly improved in this respect, and simple in structure, fast convergence rate, and precision of prediction is higher than BP network.

The present invention can utilize another kind of neural network---and Elman neural network proves the correctness of RBF neural network prediction conductor temperature.Elman neural network is a kind of typical Feedback Neural Network.The output of feedforward neural network only determines by the weights of current input and network, and the output of Feedback Neural Network is except with current input is relevant with network weight, also relevant with the input before network.Generally, Feedback Neural Network has the computing power stronger than feedforward neural network, and its most outstanding advantage is to have very strong associative memory and optimize computing function.

In the present invention, the principle that Elman neural network checking RBF neural network forecast conductor temperature value is correct is as follows: only utilize conductor temperature data to carry out data pre-service, use wire moment 6 temperature values before to predict that current conductor temperature creates Elman network, and predict the conductor temperature value of the time period of RBF neural network forecast, the conductor temperature value of two kinds of neural network predictions is compared, find that the conductor temperature value variation tendency that both predict is consistent, and differ very little, within the error range allowing.

The present invention provides the theoretical foundation of practice for the technical research of cable project and Analysis on Fault Diagnosis, simultaneously, as other temperature data of continuous collecting in cable actual motion, form new database, can directly revise the dimension of input quantity and improve, add other sign amount, can effectively judge the operation conditions of cable splice, thereby the practice function of expanding RBF neural network, makes it in practice, bring into play larger effect.

Specific embodiment described herein is only to the explanation for example of the present invention's spirit.Those skilled in the art can make various modifications or supplement or adopt similar mode to substitute described specific embodiment, but can't depart from spirit of the present invention or surmount the defined scope of appended claims.

Claims (4)

1. the cable splice conductor temperature Forecasting Methodology based on RBF neural network, is characterized in that, comprises the following steps:

Step 1, respectively between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along at least three temp probes are circumferentially evenly set; Edge, external contacts place at cable splice circumferentially evenly arranges at least two temp probes; And an ambient humidity meter and an environment temperature meter; Then use data collection station according to the time interval collecting temperature data of setting; Obtain 11 kinds of temperature variables that affect cable splice conductor temperature; Respectively: temperature, cable skin low order end temperature between temperature, cable skin high order end temperature, cable skin middle part temperature, cable skin low order end and middle part between ambient humidity, right sheath temperature, joint insulation course temperature, left sheath temperature, environment temperature, external contacts temperature, cable skin high order end and middle part;

Step 2, the data that gather based on step 1 are carried out establishment and the prediction of RBF network, specifically: in the data importing matlab first step 1 being gathered, carry out pre-service, then data after pre-service are divided into two large divisions, a part is training data, another part is predicted data, and described training data and predicted data comprise and comprise respectively some groups of data, and every group of data comprise 11 kinds of temperature variables that affect cable splice conductor temperature; Training data is input in three layers of RBF neural network and trains the RBF network that obtains training, then predicted data is input in RBF network and is predicted, obtain the data of prediction.

2. a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network according to claim 1, is characterized in that, describedly training data is input to the concrete steps of training in three layers of RBF neural network is:

Step 1, asks for Basis Function Center t based on K-means clustering algorithm _i(n); Be set with I cluster centre, i cluster centre of the n time iteration is t _i(n), i=1,2 ..., I, I value is rule of thumb to determine here, then carries out the operation of following sub-step:

Step 1.1, netinit; From the input sample data that comprises 11 temperature variable values, the random individual different sample of I of selecting is as initial cluster centre t _i(0) (i=1,2 ..., I);

Step 1.2, the training sample set of input is divided into groups by arest neighbors rule: calculate this training sample nearest apart from which cluster centre, just it is classified as to the same class of this cluster centre, calculate:

i(X _p)＝argmin||x _p-t _i(0)||

Find corresponding i value, by X _pbe classified as i class;

In formula,

be p input sample; P=1,2,3 ... P, P is total sample number; T _ifor the center of network hidden layer node;

Step 1.3, upgrades cluster centre; Calculate each cluster set θ _pthe mean value of middle training sample, i.e. new cluster centre t _i:

In formula, η is Learning Step, 0< η <1;

In the time that cluster centre no longer changes, the t obtaining _ibe the final Basis Function Center of RBF neural network, continue iteration otherwise return to (2);

Step 2, solves variances sigma _i; The basis function of the RBF neural network that should predict based on cable splice conductor temperature is Gaussian function, variances sigma _ican solve as follows:

${\mathrm{\&sigma;}}_I=\frac{c_{\max }}{2I},(i=\mathrm{1,2},...,I)$

In formula, c _maxthe ultimate range between selected center;

Step 3, calculates the RBF god based on cable splice who creates by the weights between the hidden layer of network and output layer;

Hidden layer can directly calculate by least square method to neuronic connection weights between output layer, and computing formula is as follows:

$\mathrm{\&omega;}=\exp \left(\frac{I}{c_{\max }^2}{\left|\right|x_p-t_i\left|\right|}^2\right),i=\mathrm{1,2},...,I;p=\mathrm{1,2},...,P$

Step 4, center and the variance of trying to achieve based on step 1, step 2 are expressed radial basis function:

$R(x_p-t_i)=\exp (-\frac{1}{2{\mathrm{\&sigma;}}^2}{\left|\right|x_p-t_i\left|\right|}^2)$

In formula, || x _p-t _i|| be European norm; t _ifor the center of Gaussian function; σ is the variance of Gaussian function; The known network of structural drawing by RBF neural network is output as

$y_i=\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_{\mathrm{ij}}(-\frac{1}{2{\mathrm{\&sigma;}}^2}{\left|\right|x_p-t_i\left|\right|}^2),j=\mathrm{1,2},...,n$

In formula, ω _ijfor hidden layer is to the connection weights of output layer; I=1,2,3...I, I is hidden layer node number; y _ifor with the actual output of j output node of the network of input sample, i.e. the conductor temperature value of cable splice.

3. a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network according to claim 1, it is characterized in that, respectively between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along three temp probes are circumferentially evenly set; Edge, external contacts place at cable splice circumferentially evenly arranges two temp probes; And an ambient humidity meter and an environment temperature; Then use data collection station according to the time interval collecting temperature data of setting; Obtain 11 kinds of temperature variables that affect cable splice conductor temperature; Respectively: temperature, cable skin low order end temperature between temperature, cable skin high order end temperature, cable skin middle part temperature, cable skin low order end and middle part between ambient humidity, right sheath temperature, joint insulation course temperature, left sheath temperature, environment temperature, external contacts temperature, cable skin high order end and middle part.

4. a kind of cable splice conductor temperature Forecasting Methodology based on RBF neural network according to claim 1, is characterized in that, in described step 2, image data imports carries out pretreated concrete grammar in matlab and be:

For the data of Real-time Collection, first remove more above owing to operating, measuring the inaccurate misdata of collecting temperature value former thereby that cause;

Then, between the left sheath of cable splice, right sheath, joint insulation course, cable skin high order end and middle part, between cable skin high order end, cable skin middle part, cable skin low order end and middle part, cable skin low order end place is along three temp probes are circumferentially evenly set, be that synchronization same position has gathered three temperature values, for reducing measuring error, by average data identical monitoring location, be about to mutually in the same time, three temperature values of same position are added and again divided by 3; Edge, external contacts place at cable splice circumferentially evenly arranges two temp probes; Be that synchronization same position has gathered three temperature values, for reducing measuring error, by average data identical monitoring location, be about to mutually in the same time, three temperature values of same position are added and again divided by 2.