CN114169237B - Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm - Google Patents
Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm Download PDFInfo
- Publication number
- CN114169237B CN114169237B CN202111473398.7A CN202111473398A CN114169237B CN 114169237 B CN114169237 B CN 114169237B CN 202111473398 A CN202111473398 A CN 202111473398A CN 114169237 B CN114169237 B CN 114169237B
- Authority
- CN
- China
- Prior art keywords
- temperature
- joint
- lstm
- sequence
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 title claims abstract description 23
- 230000005856 abnormality Effects 0.000 title claims abstract description 20
- 238000001514 detection method Methods 0.000 claims abstract description 48
- 230000002159 abnormal effect Effects 0.000 claims abstract description 16
- 238000012544 monitoring process Methods 0.000 claims abstract description 16
- 230000008859 change Effects 0.000 claims abstract description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 27
- 238000012360 testing method Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 11
- 230000009466 transformation Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 238000013145 classification model Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 239000002245 particle Substances 0.000 claims description 3
- 230000010259 detection of temperature stimulus Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 229920003020 cross-linked polyethylene Polymers 0.000 description 2
- 239000004703 cross-linked polyethylene Substances 0.000 description 2
- 210000001503 joint Anatomy 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing Relating To Insulation (AREA)
Abstract
The invention provides a power cable joint temperature abnormality early warning method combining an EEMD-LSTM and an isolated forest algorithm, which aims at a temperature hot spot power cable joint of a medium-low voltage distribution network to perform temperature abnormality early warning; establishing an EEMD-LSTM temperature prediction model according to the joint historical temperature monitoring data; using EEMD to decompose the original temperature data sequence into a plurality of subsequences with smaller orders of magnitude, and extracting joint temperature change trend information; predicting each subsequence through LSTM, reconstructing the prediction result of the subsequence, and outputting the temperature prediction value of the joint temperature at the future moment; adopting an isolated forest algorithm to detect temperature abnormality of a temperature index predicted by EEMD-LSTM; establishing a plurality of sub-detection classifiers, and combining the surface temperature of the joint, the core temperature and the relative temperature difference two by two to obtain three groups of classifiers; and finally, realizing joint temperature early warning. The method provided by the invention can be used for timely pre-judging the potential abnormal temperature of the connector and carrying out temperature pre-warning on the cable connector with overhigh temperature and faster temperature rise.
Description
Technical Field
The invention relates to the field of power cable line fault detection, in particular to a power cable joint temperature abnormality early warning method combining an EEMD-LSTM and an isolated forest algorithm.
Background
Along with the acceleration of industrialization and towns in China, the proportion of power cable lines in a power transmission and distribution network is continuously increased, and crosslinked polyethylene (XLPE) cables with voltage levels of 35kV and below are increasingly widely used.
The safe and stable operation of the power cable line is critical, when the cable line fails, power failure can occur, and when the cable line is severe, joint explosion can occur, so that fire disaster and casualties are caused. The cable intermediate joint is used as one of main accessories of a cable line and is used for connecting cables at two ends, and is the weakest link of the cable line. Through accident statistical analysis, the occurrence of the middle joint fault is not sudden like the line short-circuit fault, but is a continuously accumulated process. The faults that are commonly present accumulate a significant amount of joule heat, causing the cable temperature to rise continuously. When the system is in overload operation for a long time and the applied load exceeds the current-carrying capacity of the cable, the temperature at the joint can be increased, and potential safety hazards exist. The temperature is used as a non-electric quantity, can only reflect the insulation performance of the cable and the running load state of the system, is easy to measure and characterize, and is considered as an important cable condition detection means at home and abroad. The existing temperature detection technologies such as a digital temperature sensor, a fiber bragg grating and the like can accurately detect the temperature of the cable. And by combining a computer technology and an electronic technology, the cable temperature online monitoring system is developed and obtained, and the cable joint temperature and other indexes are monitored in real time. Meanwhile, the problems of joint temperature data transmission lag and insufficient data utilization exist, and the problem of overhigh maintenance cost caused by the complex cable line of the actual distribution network is solved.
Disclosure of Invention
The invention aims to solve the technical problems of misjudgment and missed judgment according to related standards and detection experience and by considering single threshold early warning; based on the distribution network equipment state evaluation guideline (Q/GDW 645-2011), the problems of excessive indexes and high index acquisition difficulty are solved. The above problems are solved by combining LSTM neural network and isolated forest algorithm.
A power cable joint temperature abnormality early warning method combining EEMD-LSTM and an isolated forest algorithm comprises the following steps:
Step 1, monitoring temperature information of a corresponding tag connector, including connector surface temperature, core temperature and internal and external relative temperature difference, based on an online monitoring system, aiming at a cable line in a 10kV distribution network cable trench in a certain area to obtain an initial temperature sequence;
Step2, decomposing the initial temperature sequence by adopting integrated empirical mode decomposition (EEMD), gradually decomposing a plurality of subsequences, and processing a non-stationary sequence;
step 3, establishing an LSTM network unit, optimizing LSTM network unit parameters, predicting each subsequence obtained by decomposition by using the LSTM network unit, reconstructing a subsequence prediction result, and outputting a joint temperature value at a future moment;
Step 4, extracting a joint temperature index data set, performing feature transformation on the original data set by using haar wavelet transformation, and extracting temperature change trend information to obtain a temperature detection sample;
Step 5, setting attention values and early warning values corresponding to the temperature indexes of the joint;
Step 6, detecting temperature abnormality by adopting an isolated forest algorithm, dividing a detection sample into a training set and a testing set, carrying out abnormality marking on the training set, and establishing a detection classifier of joint surface temperature-wire core temperature, surface temperature-relative temperature difference and wire core temperature-relative temperature difference;
and 7, detecting temperature abnormality of the test set by the established detection model according to the steps 3, 4, 5 and 6, and realizing joint temperature early warning.
Further, the integrated empirical mode decomposition described in step 2 is specifically expressed as:
The process of processing the adaptor temperature sequence using EEMD is specifically shown as follows:
(1) Setting zero-mean Gaussian white noise with the amplitude value of k and the decomposition repetition number of M, wherein the noise is subjected to normal distribution;
(2) Adding the Gaussian white noise n (t) into the cable joint temperature initial sequence x (t) to carry out m-th EMD decomposition;
1) Searching all local minimum points and maximum points in the sequence added with noise, using a 3-time spline interpolation function to fit and obtain an upper envelope sequence u (t) and a lower envelope sequence v (t), and taking the average value of the upper envelope sequence u (t) and the lower envelope sequence v (t) to obtain an average value sequence m (t) = (u (t) +v (t))/2;
2) Subtracting the average value sequence from the sequence added with noise to obtain a middle sequence h (t) =x (t) -m (t);
3) Judging whether h (t) meets an eigenmode function (IMF), if so, defining the eigenmode function as IMF1, otherwise, considering the eigenmode function as new x (t), and repeating the steps 1) -2). By such pushing, the mth EMD decomposition is completed;
4) The residual component r (t) =x (t) -h (t) is calculated. Repeating the steps 1) -3) for the residual components, performing the same stripping, and stopping when the decomposition termination condition is met;
(3) And carrying out aggregate average on IMFs obtained by M times of EMD decomposition to offset the amplitude influence caused by the added white noise.
Further, the establishing of the LSTM network element in step 3 is expressed as:
And (3) predicting the subsequence decomposed in the step (2) by using an LSTM network element. LSTM network element parameters are optimized. In LSTM network structure parameters, the single-step prediction node of the output layer is 1, and the window length is 24; the number of three-step predictive nodes is 3 and the window length is 72. The input window length is the historical sequence length used to make the joint temperature prediction. The use of a single LSTM layer can control the computational cost within a reasonable range while meeting network performance requirements, so the number of hidden layers is 1. At each training time, after determining the input and output layer lengths, a particle swarm algorithm is used to find the hidden layer neuron dimension cell_num, the net learning rate factor c, and the batch size S batch. And predicting each subsequence obtained by decomposition by using the LSTM network unit, reconstructing a subsequence prediction result, and outputting a joint temperature value at a future moment.
Further, in the step 4, the characteristic transformation is performed on the original data set by using haar wavelet transformation, and the temperature change trend information is extracted to obtain a temperature detection sample, which specifically comprises the following steps:
And extracting the data of the temperature index in the monitoring sample, the joint wire core temperature T jc, the surface temperature T bc and the internal and external temperature difference delta T c to obtain respective one-dimensional matrixes.
Tjc＝[Tjc1 Tjc2 … Tjci … Tjcn]
Tbc＝[Tbc1 Tbc2 … Tbci … Tbcn]
ΔTc＝[Tc1 Tc2 … Tci … Tcn]
(1) The average value a jc、Abc、Ac of the neighboring data is calculated.
(2) And calculating the difference between the original data and the average value. And calculating an average value, wherein part of information is lost, and the size of the matrix is half of that of the original matrix. To store the missing part of the detail information, the calculated average value is subtracted from the previous data of the adjacent data pair to obtain a difference D jc、Dbc、Dc. When the temperature acquisition period is short, this part of the difference can be regarded as the rate of temperature rise. Storing the calculated mean value and the difference value to obtain a new transformed matrix, wherein the new transformed matrix is used as a characteristic vector of joint temperature and expressed as follows:
Hjc＝[Ajc1 Ajc2 … Ajci … Ajc(n+1)/2 Djc1 Djc2 … Djci … Djc(n+1)/2]
Hbc＝[Abc1 Abc2 … Abci … Abc(n+1)/2 Dbc1 Dbc2 … Dbci … Dbc(n+1)/2]
Hc＝[Ac1 Ac2 … Aci … Ac(n+1)/2 Dc1 Dc2 … Dci … Dc(n+1)/2]
combining the original data, the final detection vector group is:
Xjc＝[Tjc Hjc]T
Xbc＝[Tbc Hbc]T
Xc＝[ΔTc Hc]T
further, the attention value and the early warning value corresponding to each temperature index of the set joint in step 5 are specifically expressed as:
According to the operation experience, attention values and early warning values are set here:
Further, in the step 6, the detection of the temperature anomaly by using the isolated forest algorithm is specifically expressed as follows:
The isolated forest algorithm is set as a standard parameter: the sample size Num sub is 256; the depth HEIGHTLIMIT of the tree is 8; the number Num tree of the isolated forest trees is 100; the maximum feature number is 1. After the setting of the index threshold is completed, different classification models are selected. And carrying out anomaly detection test on the converted complete sample, and the sample obtained by combining the temperature of the joint wire core, the temperature of the joint surface and the internal and external relative temperature difference.
Further, in the step 7, the detecting module to be built detects the abnormal temperature of the test set, so as to realize the early warning of the joint temperature, and the method includes:
The test performance was judged by ROC (Receiver Operating Characteristic) curve. The curve is plotted with false positive rate (false positive rate, FPR) as the abscissa and the detection rate (true positive rate, TPR) as the ordinate. The calculation formulas of TPR and FPR are as follows:
TPR＝TP/(TP+FN)
FPR＝FP/(FP+TN)
TP (true positive) indicates true positive, i.e., the actual joint temperature is abnormal and the detection result is abnormal, FN (false negative) indicates false negative, i.e., the actual joint temperature is abnormal and the detection result is normal, FP (false positive) indicates false positive, i.e., the actual joint temperature is normal and the detection result is abnormal, TN (true negative) indicates true negative, i.e., the actual joint temperature is normal and the detection result is normal.
The area under the ROC curve is called AUC (Area under the Curve), and when comparing different classification models, the AUC of each model can be compared as an index of the model quality.
And inputting the temperature test set to be detected into the selected detection classifier for detection.
The beneficial effects of the invention are as follows:
And the historical monitoring data source is fully utilized, and the integrated empirical mode decomposition (EEMD) is used for processing and analyzing the joint temperature sequence to obtain a plurality of subsequences with smaller orders of magnitude. And each subsequence is respectively predicted by adopting an LSTM network, so that the joint temperature trend can be predicted. Using an isolated forest algorithm, multi-index detection can be considered simultaneously. Compared with single index detection, the reliability is higher. The method provided by the invention can be used for timely pre-judging the potential abnormal temperature of the connector and carrying out temperature pre-warning on the cable connector with overhigh temperature and faster temperature rise.
Drawings
FIG. 1 is a flow chart of a power cable joint temperature anomaly early warning method combining EEMD-LSTM and an isolated forest algorithm;
FIG. 2 is joint initial temperature monitoring data;
FIG. 3 is a graph of joint temperature trend prediction;
FIG. 4 is an isolated forest algorithm detecting ROC curves;
Fig. 5 shows detection of temperature abnormality of the tag 13 A3.
Detailed Description
The power cable joint temperature abnormality early warning method combining EEMD-LSTM and an isolated forest algorithm is described in detail below with reference to the embodiment and the accompanying drawings. As shown in FIG. 1, the power cable joint temperature abnormality early warning method combining EEMD-LSTM and an isolated forest algorithm comprises the following steps:
(1) For a cable line in a 10kV distribution network cable duct in a certain area, monitoring temperature information of a corresponding tag connector based on an online monitoring system, wherein the temperature information comprises the surface temperature of the connector, the temperature of a wire core and the internal and external relative temperature difference, so as to obtain an initial temperature sequence;
(2) Decomposing the initial temperature sequence by adopting integrated empirical mode decomposition (EEMD), gradually decomposing a plurality of subsequences, and processing a non-stationary sequence;
The process of processing the adaptor temperature sequence using EEMD is specifically shown as follows:
1) Setting zero-mean Gaussian white noise with the amplitude value of k and the decomposition repetition number of M, wherein the noise is subjected to normal distribution;
2) Adding the Gaussian white noise n (t) into the cable joint temperature initial sequence x (t) to carry out m-th EMD decomposition;
① Searching all local minimum points and maximum points in the sequence added with noise, using a 3-time spline interpolation function to fit and obtain an upper envelope sequence u (t) and a lower envelope sequence v (t), and taking the average value of the upper envelope sequence u (t) and the lower envelope sequence v (t) to obtain an average value sequence m (t) = (u (t) +v (t))/2;
② Subtracting the average value sequence from the sequence added with noise to obtain a middle sequence h (t) =x (t) -m (t);
③ Judging whether h (t) meets an eigenmode function (IMF), if so, defining the eigenmode function as IMF1, otherwise, considering the eigenmode function as new x (t), and repeating the step ①-②. By such pushing, the mth EMD decomposition is completed;
④ The residual component r (t) =x (t) -h (t) is calculated. Repeating the step ①-③ for the residual component, performing the same stripping, and stopping when the decomposition termination condition is met;
3) And carrying out aggregate average on IMFs obtained by M times of EMD decomposition to offset the amplitude influence caused by the added white noise.
(3) Establishing an LSTM network unit, optimizing LSTM network unit parameters, predicting each subsequence obtained by decomposition by using the LSTM network unit, reconstructing a subsequence prediction result, and outputting a joint temperature value at a future moment;
And (3) predicting the subsequence decomposed in the step (2) by using an LSTM network element. LSTM network element parameters are optimized. In LSTM network structure parameters, the single-step prediction node of the output layer is 1, and the window length is 24; the number of three-step predictive nodes is 3 and the window length is 72. The input window length is the historical sequence length used to make the joint temperature prediction. The use of a single LSTM layer can control the computational cost within a reasonable range while meeting network performance requirements, so the number of hidden layers is 1. At each training time, after determining the input and output layer lengths, a particle swarm algorithm is used to find the hidden layer neuron dimension cell_num, the net learning rate factor c, and the batch size S batch. And predicting each subsequence obtained by decomposition by using the LSTM network unit, reconstructing a subsequence prediction result, and outputting a joint temperature value at a future moment.
(4) Extracting a joint temperature index data set, performing feature transformation on the original data set by using haar wavelet transformation, and extracting temperature change trend information to obtain a temperature detection sample;
And extracting the data of the temperature index in the monitoring sample, the joint wire core temperature T jc, the surface temperature T bc and the internal and external temperature difference delta T c to obtain respective one-dimensional matrixes.
Tjc＝[Tjc1 Tjc2 … Tjci … Tjcn]
Tbc＝[Tbc1 Tbc2 … Tbci … Tbcn]
ΔTc＝[Tc1 Tc2 … Tci … Tcn]
1) The average value a jc、Abc、Ac of the neighboring data is calculated.
2) And calculating the difference between the original data and the average value. And calculating an average value, wherein part of information is lost, and the size of the matrix is half of that of the original matrix. To store the missing part of the detail information, the calculated average value is subtracted from the previous data of the adjacent data pair to obtain a difference D jc、Dbc、Dc. When the temperature acquisition period is short, this part of the difference can be regarded as the rate of temperature rise. Storing the calculated mean value and the difference value to obtain a new transformed matrix, wherein the new transformed matrix is used as a characteristic vector of joint temperature and expressed as follows:
Hjc＝[Ajc1 Ajc2 … Ajci … Ajc(n+1)/2 Djc1 Djc2 … Djci … Djc(n+1)/2]
Hbc＝[Abc1 Abc2 … Abci … Abc(n+1)/2 Dbc1 Dbc2 … Dbci … Dbc(n+1)/2]
Hc＝[Ac1 Ac2 … Aci … Ac(n+1)/2 Dc1 Dc2 … Dci … Dc(n+1)/2]
combining the original data, the final detection vector group is:
Xjc＝[Tjc Hjc]T
Xbc＝[Tbc Hbc]T
Xc＝[ΔTc Hc]T
(5) Setting an attention value and an early warning value corresponding to each temperature index of the joint;
Setting an attention value and an early warning value according to operation experience, and respectively giving an upper threshold value and a lower threshold value:
| Index (I) | Attention value | Early warning value |
| Joint core temperature | 70℃ | 85℃ |
| Joint surface temperature | 45℃ | 50℃ |
| Relative temperature difference | 20℃ | 30℃ |
(6) Detecting temperature abnormality by adopting an isolated forest algorithm, dividing a detection sample into a training set and a testing set, carrying out abnormality marking on the training set, and establishing a detection classifier of joint surface temperature-wire core temperature, surface temperature-relative temperature difference and wire core temperature-relative temperature difference;
The isolated forest algorithm is set as a standard parameter: the sample size is 256; the depth of the tree is 8; the number of isolated forest trees is 100; the maximum feature number is 1. After the setting of the index threshold is completed, different classification models are selected. And carrying out anomaly detection test on the converted complete sample, and the sample obtained by combining the temperature of the joint wire core, the temperature of the joint surface and the internal and external relative temperature difference.
(7) And (3) according to the steps (3), (4), (5) and (6), carrying out temperature anomaly detection on the test set by the established detection model, and realizing joint temperature early warning.
Specific examples are given below:
And taking 30-day data sets of temperature data of different joints from 7 months in 2019 to 15 months in one cable pit of a certain 10kV distribution network as a study sample. The monitoring interval time is 1h, and the period is 24h. Wherein 480 groups of monitoring data from 7 months 15 days to 8 months 3 days zero are used as initial training samples, the training samples are added by subsequent rolling, and the data from 8 months 4 days to 8 months 13 days zero are tested.
FIG. 2 is a sequence of initial temperature of the joint. FIG. 3 shows the joint variation trend obtained by single-step and three-step prediction of the LSTM network, and it can be seen that the EEMD-LSTM single-step and three-step prediction can predict the joint temperature variation trend. FIG. 4 is a ROC curve of different classifiers, wherein the AUC values of the classifiers are all between 0.83 and 0.85, and two groups of indexes of core temperature and relative temperature difference are selected for combination, so that the abnormality detection effect is better; selecting two groups of index combinations of surface temperature and relative temperature difference, and detecting the inferior effect; the AUC average value of the detection of the whole sample is 0.981, and the detection effect is excellent. Fig. 5 shows that the abnormal points far from the normal temperature data are detected at the joint of the tag 13A3 in the cable duct, and the abnormal points are distributed outside the abnormal value contour, and the abnormal points at the tag 13A3 can be detected within the range of-6.05044 to-7.67423.
The foregoing description of the preferred embodiments of the present invention has been presented only in terms of those specific and detailed descriptions, and is not, therefore, to be construed as limiting the scope of the invention. It should be noted that modifications, improvements and substitutions can be made by those skilled in the art without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.
Claims (6)
1. The power cable joint temperature abnormality early warning method combining EEMD-LSTM and an isolated forest algorithm is characterized by comprising the following steps of:
Step 1, monitoring temperature information of a tag connector, including connector surface temperature, core temperature and internal and external relative temperature difference, based on a corresponding online monitoring system aiming at a cable line in a 10kV distribution network cable trench in a certain area to obtain an initial temperature sequence;
Step 2, decomposing the initial temperature sequence by adopting integrated empirical mode decomposition (EEMD), gradually decomposing a plurality of subsequences, and processing a joint initial temperature non-stable sequence;
step 3, establishing an LSTM network unit, optimizing LSTM network unit parameters, predicting each subsequence obtained by decomposition by using the LSTM network unit, reconstructing a subsequence prediction result, and outputting a joint temperature value at a future moment;
Step 4, extracting a temperature index data set from the joint monitoring real-time temperature data and the prediction data, carrying out characteristic transformation on an original data set by utilizing Harr wavelet transformation, and extracting temperature change trend information to obtain a temperature detection sample;
Step 5, setting attention values and early warning values corresponding to the temperature indexes of the joint;
Step 6, detecting temperature abnormality by adopting an isolated forest algorithm, dividing a detection sample into a training set and a testing set, carrying out abnormality marking on the training set, and establishing a detection sub-classifier of joint surface temperature-wire core temperature, surface temperature-relative temperature difference and wire core temperature-relative temperature difference;
and 7, detecting temperature abnormality of the test set by the established detection model according to the steps 3, 4, 5 and 6, and realizing joint temperature early warning.
2. The power cable joint temperature anomaly early warning method combining EEMD-LSTM and isolated forest algorithm according to claim 1, wherein the integrated empirical mode decomposition in step 2 is specifically expressed as:
The process of processing the adaptor temperature sequence using EEMD is specifically shown as follows:
(1) Setting zero-mean Gaussian white noise with the amplitude value of k and the decomposition repetition number of M, wherein the noise is subjected to normal distribution;
(2) Adding the Gaussian white noise n (t) into the cable joint temperature initial sequence x (t) to carry out m-th EMD decomposition;
1) Searching all local minimum points and maximum points in the sequence added with noise, using a 3-time spline interpolation function to fit and obtain an upper envelope sequence u (t) and a lower envelope sequence v (t), and taking the average value of the upper envelope sequence u (t) and the lower envelope sequence v (t) to obtain an average value sequence m (t) = (u (t) +v (t))/2;
2) Subtracting the average value sequence from the sequence added with noise to obtain a middle sequence h (t) =x (t) -m (t);
3) Judging whether h (t) meets an eigenmode function IMF, if so, defining the eigenmode function IMF as IMF1, otherwise, considering the eigenmode function as new x (t), and repeating the steps 1) -2); by such pushing, the mth EMD decomposition is completed;
4) Calculating a residual component r (t) =x (t) -h (t); repeating the steps 1) -3) for the residual components, performing the same stripping, and stopping when the decomposition termination condition is met;
(3) And carrying out aggregate average on IMFs obtained by M times of EMD decomposition to offset the amplitude influence caused by the added white noise.
3. The method for early warning of abnormal temperature of a power cable joint according to claim 1, wherein the establishing of the LSTM network unit in step 3 is expressed as:
Predicting the sub-sequence decomposed in the step 2 by using an LSTM network unit; optimizing LSTM network element parameters; in LSTM network structure parameters, the single-step prediction node of the output layer is 1, and the window length is 24; the number of the three-step prediction nodes is 3, and the window length is 72; the length of the input window is the length of a historical sequence for predicting the joint temperature; the single LSTM layer is used for controlling the calculation cost within a reasonable range, and meanwhile, the network performance requirement is met, and the number of hidden layers is 1; during each training, after determining the lengths of the input layer and the output layer, searching the hidden layer neuron dimension cell_num, the network learning rate factor c and the batch size S batch by using a particle swarm algorithm; and predicting each subsequence obtained by decomposition by using the LSTM network unit, reconstructing a subsequence prediction result, and outputting a joint temperature value at a future moment.
4. The power cable joint temperature anomaly early warning method combining EEMD-LSTM and isolated forest algorithm according to claim 1, wherein the characteristic transformation is carried out on an original data set by using haar wavelet transformation in the step 4, temperature change trend information is extracted, and a temperature detection sample is obtained, and the specific process is as follows:
Extracting data of temperature indexes in the monitoring sample, namely joint wire core temperature T jc, surface temperature T bc and internal and external temperature difference delta T c to obtain respective one-dimensional matrixes;
Tjc＝[Tjc1 Tjc2…Tjci…Tjcn]
Tbc＝[Tbc1 Tbc2…Tbci…Tbcn]
ΔTc＝[Tc1 Tc2…Tci…Tcn]
(1) Calculating an average value A jc、Abc、Ac of adjacent data;
(2) Calculating the difference between the original data and the average value; calculating an average value, wherein part of information is lost, and the size of the matrix is half of the original size; for storing the lost part of detail information, subtracting the calculated average value from the previous data of the adjacent data pair to obtain a difference D jc、Dbc、Dc; when the temperature acquisition period is very short, the difference value can be regarded as the temperature rise rate; storing the calculated mean value and the difference value to obtain a new transformed matrix, wherein the new transformed matrix is used as a characteristic vector of joint temperature and expressed as follows:
Hjc＝[Ajc1 Ajc2…Ajci…Ajc(n+1)/2 Djc1 Djc2…Djci…Djc(n+1)/2]
Hbc＝[Abc1 Abc2…Abci…Abc(n+1)/2 Dbc1 Dbc2…Dbci…Dbc(n+1)/2]
Hc＝[Ac1 Ac2…Aci…Ac(n+1)/2 Dc1 Dc2…Dci…Dc(n+1)/2]
combining the original data, the final detection vector group is:
Xjc＝[Tjc Hjc]T
Xbc＝[Tbc Hbc]T
Xc＝[ΔTc Hc]T。
5. The power cable joint temperature anomaly early warning method combining EEMD-LSTM and isolated forest algorithm according to claim 1, wherein the detection of temperature anomalies by adopting isolated forest algorithm in step 6 is specifically expressed as:
the isolated forest algorithm is set as a standard parameter: the sample size Num sub is 256; the depth HEIGHTLIMIT of the tree is 8; the number Num tree of the isolated forest trees is 100; the maximum feature number is 1; after the setting of the index threshold is completed, selecting different classification models; and carrying out anomaly detection test on the converted complete sample, and the sample obtained by combining the temperature of the joint wire core, the temperature of the joint surface and the internal and external relative temperature difference.
6. The power cable joint temperature anomaly early warning method combining EEMD-LSTM and isolated forest algorithm according to claim 1, wherein the method is characterized in that the step 7 of detecting the temperature anomaly of the test set by the established detection model, realizing joint temperature early warning, and comprises the following steps:
judging the detection performance by an ROC curve, wherein the curve takes the false alarm rate FPR as an abscissa and the detection rate TPR as an ordinate; the calculation formulas of TPR and FPR are as follows:
TPR＝TP/(TP+FN)
FPR＝FP/(FP+TN)
TP indicates true positive, i.e., the actual joint temperature is abnormal and the detection result is abnormal, FN indicates false negative, i.e., the actual joint temperature is abnormal and the detection result is normal, FP indicates false positive, i.e., the actual joint temperature is normal and the detection result is abnormal, TN indicates true negative, i.e., the actual joint temperature is normal and the detection result is normal;
the area under the ROC curve is called AUC, and when different classification models are compared, the AUC of each model is compared and used as an index of the model quality;
and inputting the temperature test set to be detected into the selected detection classifier for detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111473398.7A CN114169237B (en) | 2021-11-30 | 2021-11-30 | Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111473398.7A CN114169237B (en) | 2021-11-30 | 2021-11-30 | Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114169237A CN114169237A (en) | 2022-03-11 |
| CN114169237B true CN114169237B (en) | 2024-05-03 |
Family
ID=80483206
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111473398.7A Active CN114169237B (en) | 2021-11-30 | 2021-11-30 | Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114169237B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115438102B (en) * | 2022-11-09 | 2023-03-24 | 成都新希望金融信息有限公司 | Space-time data anomaly identification method and device and electronic equipment |
| CN115760484A (en) * | 2022-12-07 | 2023-03-07 | 湖北华中电力科技开发有限责任公司 | Method, device and system for improving potential danger identification capability of power distribution station area and storage medium |
| CN116627953B (en) * | 2023-05-24 | 2023-10-27 | 首都师范大学 | Method for repairing loss of groundwater level monitoring data |
| CN117713505B (en) * | 2024-02-06 | 2024-04-05 | 北京嘉捷恒信能源技术股份有限公司 | High-side current detection method and system for switching power supply |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019233189A1 (en) * | 2018-06-04 | 2019-12-12 | 江南大学 | Method for detecting sensor network abnormal data |
| CN111241466A (en) * | 2020-01-15 | 2020-06-05 | 上海海事大学 | Ship flow prediction method based on deep learning |
| CN111738520A (en) * | 2020-06-24 | 2020-10-02 | 中国电子科技集团公司第二十八研究所 | System load prediction method fusing isolated forest and long-short term memory network |
| CN112287602A (en) * | 2020-10-28 | 2021-01-29 | 北京国信会视科技有限公司 | Motor car axle temperature fault early warning method based on machine learning and isolated forest |
-
2021
- 2021-11-30 CN CN202111473398.7A patent/CN114169237B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019233189A1 (en) * | 2018-06-04 | 2019-12-12 | 江南大学 | Method for detecting sensor network abnormal data |
| CN111241466A (en) * | 2020-01-15 | 2020-06-05 | 上海海事大学 | Ship flow prediction method based on deep learning |
| CN111738520A (en) * | 2020-06-24 | 2020-10-02 | 中国电子科技集团公司第二十八研究所 | System load prediction method fusing isolated forest and long-short term memory network |
| CN112287602A (en) * | 2020-10-28 | 2021-01-29 | 北京国信会视科技有限公司 | Motor car axle temperature fault early warning method based on machine learning and isolated forest |
Non-Patent Citations (1)
| Title |
|---|
| 基于EEMD和改进Elman神经网络的电力电缆故障监测及预警算法研究;杨涛;白晗;刘洋;王云龙;张楚曼;;实验室研究与探索;20181015(10);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114169237A (en) | 2022-03-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114169237B (en) | Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm | |
| CN110829417B (en) | Electric power system transient stability prediction method based on LSTM double-structure model | |
| CN111650451A (en) | Converter transformer fault reason identification method and system | |
| CN111931849B (en) | Hydropower unit operation data trend early warning method | |
| CN102393929A (en) | State overhauling control method of distribution network equipment state assessment system | |
| CN113780401A (en) | Composite insulator fault detection method and system based on principal component analysis method | |
| CN112632148B (en) | Intelligent early warning method for power distribution network components under extreme weather condition | |
| CN114254818A (en) | Low-voltage distribution network cascading failure early warning method based on risk assessment model | |
| CN112883634A (en) | DC measurement system state prediction method and system based on multi-dimensional analysis | |
| CN115308465A (en) | Power supply potential safety hazard intelligent diagnosis system and method based on BP-SOM combined neural network | |
| CN109711687B (en) | Insulator state fuzzy evaluation method based on improved entropy method | |
| CN116840765B (en) | Voltage transformer error state evaluation method based on multivariate time sequence analysis | |
| CN117407675A (en) | Lightning arrester leakage current prediction method based on multi-variable reconstruction combined dynamic weight | |
| CN113313371A (en) | Power distribution network risk assessment method and device, computer equipment and storage medium | |
| CN116677570A (en) | Fault early warning method and system based on cabin temperature monitoring of offshore wind turbine | |
| ul Hassan et al. | Online static security assessment for cascading failure using stacked De-noising Auto-encoder | |
| CN112946369B (en) | Cable aging identification method using power line communication device | |
| Sui et al. | Transformer oil temperature prediction based on long and short-term memory networks | |
| CN112001073B (en) | Reliability analysis and research method for traction power supply system | |
| CN113935423A (en) | Power battery fault early warning method and system for coupling fuzzy control rule with Elman neural network | |
| Jing et al. | State evaluation of power cable based on RBF information fusion algorithm using multi-parameters | |
| Zhang et al. | Electric vehicle charging fault diagnosis based on bp neural network | |
| CN115224684A (en) | Intelligent power distribution network risk state identification method and system based on immune hazard theory | |
| CN114899946B (en) | Electric power system distributed fault diagnosis method based on average consistency | |
| CN112327220B (en) | Distribution transformer health on-line diagnosis method and equipment based on multi-source data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |