CN110941902A - Lightning stroke fault early warning method and system for power transmission line - Google Patents

Lightning stroke fault early warning method and system for power transmission line Download PDF

Info

Publication number
CN110941902A
CN110941902A CN201911172867.4A CN201911172867A CN110941902A CN 110941902 A CN110941902 A CN 110941902A CN 201911172867 A CN201911172867 A CN 201911172867A CN 110941902 A CN110941902 A CN 110941902A
Authority
CN
China
Prior art keywords
neural network
power transmission
lightning stroke
lightning
sample
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.)
Granted
Application number
CN201911172867.4A
Other languages
Chinese (zh)
Other versions
CN110941902B (en
Inventor
李尚轩
郁琛
吴琛
刘旭斐
常康
赵明
黄燕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nari Technology Co Ltd
Yunnan Power Grid Co Ltd
Original Assignee
Nari Technology Co Ltd
Yunnan Power Grid Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nari Technology Co Ltd, Yunnan Power Grid Co Ltd filed Critical Nari Technology Co Ltd
Priority to CN201911172867.4A priority Critical patent/CN110941902B/en
Publication of CN110941902A publication Critical patent/CN110941902A/en
Application granted granted Critical
Publication of CN110941902B publication Critical patent/CN110941902B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06N3/045—Combinations of 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

Abstract

The invention discloses a lightning stroke fault early warning method and a lightning stroke fault early warning system for a power transmission line, wherein lightning stroke information is input into a PSO-LM-BP neural network model to obtain a prediction result, and whether the power transmission line has a lightning stroke fault or not is judged according to the prediction result and a lightning stroke fault threshold value; the PSO-LM-BP neural network model training step comprises the following steps: inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, taking a sample corresponding to the power transmission corridor with the minimum mean square error as a final training sample of the neural network, and obtaining the final BP neural network through the final training sample, the particle swarm optimization and the LM algorithm training again. The method aims at line lightning stroke fault early warning, solves the contradiction between the prediction accuracy and the lightning zone identification accuracy of the traditional lightning zone prediction method, and actually proves that the method has higher accuracy on line lightning stroke fault early warning.

Description

Lightning stroke fault early warning method and system for power transmission line
Technical Field
The invention belongs to the technical field of power system fault early warning, and relates to a lightning stroke fault early warning method and system for a power transmission line.
Background
Lightning stroke faults are the main cause of transmission line faults. How to reduce the lightning stroke failure rate of the line is a direct method to improve the lightning protection level of the power transmission line, but the economy of the power transmission line is reduced, and the lightning stroke failure cannot be avoided. Therefore, the lightning stroke fault needs to be early warned in a short time before the lightning stroke fault occurs, control measures are taken, and outage loss caused by the lightning stroke fault is reduced. However, the traditional thunder area forecasting method has contradiction between the forecasting accuracy and the thunder area identification accuracy, and the accuracy of lightning stroke fault forecasting is difficult to further improve.
Disclosure of Invention
The invention aims to provide a lightning stroke fault early warning method and a lightning stroke fault early warning system for a power transmission line, and solves the problem that the existing lightning stroke fault prediction is inaccurate.
The technical solution for realizing the purpose of the invention is as follows: a lightning stroke fault early warning method for a power transmission line comprises the following steps:
inputting the lightning strike information into a PSO-LM-BP neural network model to obtain a prediction result, and judging whether the power transmission line has lightning strike faults or not according to the prediction result and the lightning strike fault threshold value;
the PSO-LM-BP neural network model training step comprises the following steps:
inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, taking a sample corresponding to the power transmission corridor with the minimum mean square error as a final training sample of the neural network, and obtaining the final BP neural network through the final training sample, the particle swarm optimization and the LM algorithm training again.
Further, the lightning strike information includes:
number of landmine in line corridor X1(ii) a Distance X between lightning strike nearest to line and line2(ii) a Average lightning current intensity X in line corridor3(ii) a Lightning current intensity X of the lightning falling nearest to the line4。
Further, the generation process of the fault sample and the normal sample is as follows:
dividing the thunderstorm process into discrete time periods according to a set time step for each thunderstorm, taking the counted thunderstorm information in one time step and the lightning stroke fault information in the next time step as a sample, and generating a fault sample;
and generating an original normal sample when the line has the lightning information but does not have a fault, and generating the normal sample by adopting a random down-sampling method for the original normal sample.
Further, training through a particle swarm algorithm and an LM algorithm to obtain an initial BP neural network, comprising the following steps:
optimizing the initial weight between the input layer neuron and the hidden layer neuron of the BP neural network by applying a particle swarm optimization algorithm;
and (3) training the BP neural network by applying an LM algorithm.
Further, optimizing the initial weight between the input layer neuron and the hidden layer neuron of the BP neural network by applying a particle swarm optimization algorithm, comprising the following steps:
setting the initial population as a plurality of pq + q-dimensional row vectors, wherein each particle contains all weight information of the network, and expressing the fitness value of each particle by using the mean square error e of the BP neural network:
Figure BDA0002289201590000021
in the formula, p is the number of neurons of the input layer of the BP neural network, q is the number of neurons of the hidden layer of the BP neural network, n is the number of samples, i is the sample serial number, and i is 1,2, n; y isiFor the ith sample xiThe output, y ', of the output layer is obtained by the forward propagation of the layers of the network'iFor the actual output of the sample, the optimal position P is determined by tracking the history of the individual in each iterationg kAnd group historical best position Pq kUpdating the speed and the position of the particle to obtain the optimal fitness e of the particle corresponding to the minimum mean square error in the population;
the update formulas of speed and position are respectively:
Figure BDA0002289201590000022
Figure BDA0002289201590000023
in the formula, Vi kThe moving speed of the ith particle in the kth iteration; pper kFor the historical best position of the individual in the kth iteration, Pgro kFor the historical best position in the kth iteration of the population, c1And c2Taking a non-negative constant as an acceleration factor; r is1And r2Is a random number between 0 and 1, Xi kIs the position in the kth iteration of the ith particle; w is a variable inertial weight that decreases linearly with the number of iterations:
Figure BDA0002289201590000024
in the formula, wmaxAnd wminMaximum and minimum inertial weight, k, respectivelymaxIs the maximum iteration number;
recalculating the fitness value e of the new particle, updating Pper kAnd Pgro kK is k +1, when the number of iterations reaches kmaxAnd stopping iteration, otherwise, updating the speed and the position of the particles again for calculation to obtain the optimal fitness of the particles corresponding to the minimum mean square error of the population, and taking the value of the variable inertial weight corresponding to the particles with the optimal fitness of the population as the initial weight of the BP neural network.
Further, the LM algorithm is applied to train the BP neural network:
let the error indicator function E of the network be:
Figure BDA0002289201590000025
where N is the number of neurons in the output layer, s is 1,2, …, and N is 1, e in this examples(u) is the error of the s-th output neuron, and u is the vector formed by the mean square error threshold and the weight, then:
Figure BDA0002289201590000031
wherein e (u) is a neural network mean square error column vector,
Figure BDA0002289201590000032
for the gradient of the error indicator function E, J (u) is a Jacobian matrix whose expansion is:
Figure BDA0002289201590000033
u1,…,uNvectors consisting of threshold values and weight values of 1 st to N times of neural network iteration respectively;
the weight correction formula of the LM algorithm is as follows:
xv+1＝xv-[JT(x)J(x)+μI]-1J(x)e(x)
in the formula, xvIs a vector formed by the weights of the neural network in the v iteration, wherein I is a unit matrix, and mu is a proportionality coefficient;
calculating the current mean square error evIf e isv<e0Or v>vmax，e0Is the target mean square error, vmaxIf the set iteration threshold is reached, the training is ended, otherwise, the weight is adjusted according to the weight correction formula, and the adjusted error value e is calculatedv+1If e isv>ev+1If not, the adjustment is invalid, the mu is enabled to be mu/β, the fitness value of each particle, namely the corresponding mean square error is recalculated, and the historical optimal position P of each particle is recordedg kAnd group historical best position Pq kAnd finally obtaining the trained BP neural network model.
Further, inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm algorithm and LM algorithm training to obtain each mean square error, wherein the method comprises the following steps:
taking values of the line corridor width according to a set interval, extracting corresponding lightning information characteristics under each value, wherein N is arranged under each width corridorzPredicting the sample by adopting an initial BP neural network to obtain a new prediction error ea，a＝1,2,…,NzThereby obtaining NzError sequence e of individual samples1,e2,…,eNzAnd calculating the mean square error value e thereofm：
Figure BDA0002289201590000034
Further, the method also comprises the following steps: evaluating the prediction result by the evaluation index, the steps comprising:
defining lightning stroke faults as positive classes, and enabling normal operation to be negative classes; when the lightning stroke fault is correctly predicted, the method is called as a true TP class; when the lightning stroke fault is mispredicted, the lightning stroke fault is called false negative FN; when the normal operation is mispredicted, the operation is called false positive class FP; when normal operation is correctly predicted, the normal operation is called true negative type TN; the effect of lightning strike fault prediction can be determined by the true class rate TPR or hit rate, the negative and positive class rate FPR or false alarm rate, and the accuracy ACC:
Figure BDA0002289201590000041
Figure BDA0002289201590000042
Figure BDA0002289201590000043
a transmission line lightning stroke fault early warning system comprises:
the lightning stroke fault prediction module is used for inputting the lightning stroke information into the PSO-LM-BP neural network model to obtain a prediction result, and judging whether the power transmission line has a lightning stroke fault or not according to the prediction result and the lightning stroke fault threshold value;
the PSO-LM-BP neural network model training step comprises the following steps:
inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, taking a sample corresponding to the power transmission corridor with the minimum mean square error as a final training sample of the neural network, and obtaining the final BP neural network through the final training sample, the particle swarm optimization and the LM algorithm training again.
Further, inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm algorithm and LM algorithm training to obtain each mean square error, wherein the method comprises the following steps:
taking values of the line corridor width according to a set interval, extracting corresponding lightning information characteristics under each value, wherein N is arranged under each width corridorzPredicting the sample by adopting an initial BP neural network to obtain a new prediction error ea，a＝1,2,…,NzThereby obtaining NzError sequence e of individual samples1,e2,…,eNzAnd calculating the mean square error value e thereofm：
Figure BDA0002289201590000044
Compared with the prior art, the invention has the beneficial effects that: the method combines the PSO method and the LM method to construct the BP neural network, solves the problem of accuracy of lightning stroke fault prediction of the power transmission line, and compared with the traditional method for evaluating the fault probability of the power transmission line in a lightning occurrence area, the method has the advantages of lower false alarm rate, higher accuracy and higher iteration speed, and can provide decision basis for line lightning defense.
Drawings
Fig. 1 is a flowchart of a lightning stroke fault early warning method for a power transmission line according to an embodiment of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Example 1:
as shown in fig. 1, a power transmission line lightning stroke fault early warning method based on a PSO-LM-BP neural network comprises the following steps:
step 1, screening input characteristics through historical thunderbolt data;
the input features of the screening were:
1) number of landmine in each power transmission line corridor X1；
2) Distance X between lightning strike nearest to each line and line2；
3) Average lightning current intensity X in each line corridor3；
4) Lightning current intensity X of lightning strike nearest to each line4。
Step 2, dividing thunderstorm time step length and forming training and testing samples; the method comprises the following specific steps:
1) for each thunderstorm, dividing the thunderstorm process into discrete time periods by taking 15 minutes as a time step; the counted thunderbolt information (X) in a time step1、X2、X3、X4) Generating a fault sample by taking the lightning stroke fault information in the next time step as a sample, wherein the lightning stroke fault information is the input characteristic of the BP neural network, and the next time step isThe lightning stroke fault information is output characteristics; selecting an original normal sample: line presence lightning information (X)1、X2、X3、X4) But no failure occurred.
2) And processing the original normal samples by adopting a random down-sampling method to generate normal samples, keeping the number of the normal samples balanced with the number of the fault samples, taking 80% of the normal and fault samples as training samples, and taking the rest samples as test samples. (random downsampling, i.e. randomly reducing the number of normal samples to make the samples balanced)
Step 3, optimizing the initial weight between the input layer neuron and the hidden layer neuron of the BP neural network by applying a PSO (particle swarm optimization) algorithm, and specifically comprising the following steps:
setting various parameters of the PSO algorithm, the number of particle populations, acceleration factors c1 and c2 and the maximum iteration number kmaxAnd randomly initializing each particle to represent the weight of the neural network.
Setting an initial population as a plurality of pq + q-dimensional row vectors, wherein p is the number of neurons of an input layer of a BP neural network, q is the number of neurons of a hidden layer of the BP neural network, each particle contains all weight information of the BP neural network, and the mean square error e of the BP neural network is used for expressing the fitness value of each particle, and the formula is as follows:
Figure BDA0002289201590000051
wherein n is the number of samples, i is the sample number, i is 1,2, n; y isiFor the ith sample xiThe output, y, of the output layer is obtained by the forward propagation layer by layer of the networki' recording individual historical optimum position P for actual output of sampleper kAnd group historical best position Pgro kThe speed and the position of the particles are updated, and the optimal fitness e of the particles in the population, namely the fitness e of the particles corresponding to the minimum mean square error is obtained.
The update formulas of speed and position are respectively:
Figure BDA0002289201590000061
Figure BDA0002289201590000062
in the formula, Vi kThe moving speed of the ith particle in the kth iteration; pper kFor the historical best position of the individual in the kth iteration, Pgro kFor the historical best position in the kth iteration of the population, c1And c2Taking a non-negative constant as an acceleration factor; r is1And r2Is a random number between 0 and 1, Xi kIs the position in the kth iteration of the ith particle; w is a variable inertial weight that decreases linearly with the number of iterations:
Figure BDA0002289201590000063
in the formula, wmaxAnd wminMaximum and minimum inertial weights, respectively, preferably 0.9 and 0.4, kmaxFor maximum number of iterations, initial iteration kmaxThe global searching capability is greatly developed, and k is 1,2,3 …, kmaxAnd w is smaller with emphasis on local search.
Recalculating the fitness value e of the new particle, updating Pper kAnd Pgro kK is k +1, when the number of iterations reaches kmaxAnd (4) if not, updating the speed and the position of the particles again to calculate to obtain the optimal fitness of the particles corresponding to the minimum mean square error of the population, and taking the value of the variable inertial weight corresponding to the particles with the optimal fitness of the population as the initial weight of the BP neural network.
Step 4, constructing a BP neural network model and training the BP neural network by applying an LM algorithm to obtain an initial BP neural network, and the specific steps are as follows:
taking the value of the variable inertial weight corresponding to the population optimal fitness particles as the initial weight of the BP neural network, and initializing each parameter of the LM algorithm, if the target mean square error is e0A proportionality coefficient mu, an adjustment factor β,maximum number of iterations v of the LM algorithmmax。
Constructing a BP neural network model, which specifically comprises the following steps:
the input layer has p neurons, wherein p-1 neurons respectively correspond to p-1-dimensional input features (X) of the sample1、X2、X3、X4) The other neuron is designed for the threshold value of the hidden layer activation function, the value is-1, and the input sample is xi＝[xi1,xi2…xi(p-1),-1]T，xijFor j dimension input feature of i input sample, q hidden layer neurons are provided, wherein q-1 neurons are connected with the input layer by weight, and weight between j (1, 2, …, p) input layer neuron and f (1, 2, …, q) hidden layer neuron is wjfIf another neuron is also set for the output layer activation function, whose value is always-1, then the net input net of the f-th hidden layer neuronfComprises the following steps:
Figure BDA0002289201590000071
the hidden layer activation function adopts sigmoid function, and then the output m of the f hidden layer neuronfComprises the following steps:
Figure BDA0002289201590000072
the neural network outputs are:
Figure BDA0002289201590000073
Figure BDA0002289201590000074
where net represents a weighted sum of the output vectors, yiFor the ith sample xiThe output, w, of the output layer is obtained by the forward propagation layer by layer of the networkfRepresenting the weight between the output layer neuron and the f hidden layer neuron; the lightning stroke fault early warning is one to twoThe classification problem is solved, so that the output layer only needs to set one neuron, and the neuron and the hidden layer neuron are weighted by weight wfAnd (4) connecting.
The LM algorithm is a combination of the gradient descent method and the gauss-newton method. The algorithm adopts approximate second derivative information, so that the required iteration time is short, the convergence is very quick, the stability is good, and the local minimum value can be avoided to a great extent.
The method comprises the following steps of training a BP neural network by applying an LM algorithm to obtain the BP neural network, and specifically comprises the following steps:
let the error index function E of the BP neural network be:
Figure BDA0002289201590000075
where N is the number of neurons in the output layer, s is 1,2, …, and N is 1, e in this examples(u) is the error of the s-th output neuron, and u is the vector formed by the mean square error threshold and the weight, then:
Figure BDA0002289201590000076
wherein e (u) is a neural network mean square error column vector,
Figure BDA0002289201590000077
for the gradient of the error indicator function E, J (u) is a Jacobian matrix whose expansion is:
Figure BDA0002289201590000081
u1,…,uNvectors consisting of thresholds and weights of 1 st, 2 … th and N times of neural network iteration respectively;
the weight correction formula of the LM algorithm is as follows:
xv+1＝xv-[JT(x)J(x)+μI]-1J(x)e(x)
in the formula, xvIs a vector formed by the weights of the neural network in the v-th iteration, I is an identity matrix, and mu is a ratioThe coefficients are illustrated.
If the proportionality coefficient mu is 0, the method is equivalent to the Gauss-Newton method; if the scale factor value is very large, the LM algorithm is close to the gradient descent method. Each iteration succeeds one step, μ decreases by some. Therefore, when the target is close to the error target, the method is closer to the Gauss-Newton algorithm, the calculation speed is higher, and the accuracy is higher. Mu is a heuristic parameter, and for a given parameter mu, mu decreases if the calculated threshold change Δ u enables the error function e (u) to decrease. Otherwise, μ increases.
Calculating the current mean square error evIf e isv<e0Or v>vmax，e0To set mean square error threshold, vmaxEnding the training if the set iteration threshold is reached, otherwise substituting the new vector u into the weight correction formula to form a new Jacobian matrix, adjusting the weight according to the weight correction formula, and calculating the adjusted error value ev+1If e isv>ev+1If not, the adjustment is invalid, the mu is enabled to be mu/β, the fitness value of each particle, namely the corresponding mean square error is recalculated, and the historical optimal position P of each particle is recordedg kAnd group historical best position Pq k. And finally obtaining the trained BP neural network model.
And step 5, determining the optimal transmission corridor width and retraining the neural network to improve the prediction precision, wherein the determination method of the line corridor width comprises the following steps:
the invention takes the width of the line corridor from 1km to 5km every 0.5km, extracts the corresponding lightning information characteristic under each value to obtain a plurality of samples under corridors with different widths, and N is arranged under each corridor with different widthszAnd (4) sampling. Under each width corridor NzPredicting the sample by adopting the BP neural network in the step 4 to obtain a new prediction error ea，a＝1,2,…,NzThereby obtaining NzError sequence e of individual samples1,e2,…,eNzAnd calculating the mean square error value e thereofm：
Figure BDA0002289201590000082
eaIs the a-th new prediction error;
n corresponding to the transmission corridor width with the minimum mean square error in corridors with different widthszAnd (4) taking the samples as training samples of the neural network, and training the BP neural network by adopting the training samples according to the method in the step 4 to obtain a PSO-LM-BP neural network model.
And determining the proper width of the power transmission corridor after comparing the results of the training samples of the power transmission corridors with different widths obtained in the neural network, and obtaining the training sample under the width as a final sample.
Step 6, testing the neural network, and determining the optimal lightning stroke fault threshold value; the method comprises the following specific steps:
testing the trained neural network by adopting a test sample, and predicting a sample x0To obtain a prediction output, noted as y0When outputting y0When the lightning stroke fault threshold value is larger than the lightning stroke fault threshold value, classifying the sample into a fault class; when y is0And when the lightning stroke fault threshold value is less than the lightning stroke fault threshold value, classifying the sample into a normal sample. And adjusting the size of the lightning stroke fault threshold value to obtain the lightning stroke fault threshold value which enables the prediction effect to be optimal.
Step 7, inputting the lightning strike information into a PSO-LM-BP neural network model during use to obtain a prediction result, judging whether the power transmission line has a lightning strike fault or not according to the prediction result, and evaluating the prediction result according to an evaluation index; the method comprises the following specific steps:
1) the method comprises the steps that mine falling data information of a single time step of a certain line monitored in real time is used as input data, and a prediction result is obtained;
2) adopting the evaluation indexes of the second classification as evaluation indexes; lightning faults are defined as positive type, and normal operation is negative type. When correctly predicted, lightning strike faults are called True Positive (TP); when a lightning fault is mispredicted, it is called False Negative (FN); when normal operation is mispredicted, it is called False Positive (FP); when normal operation is predicted correctly, it is called True Negative (TN). The effect of lightning strike fault prediction can be expressed as True Positive Rate (TPR) or hit rate, negative positive rate (FPR) or False alarm rate, and Accuracy (Accuracy, ACC):
Figure BDA0002289201590000091
Figure BDA0002289201590000092
Figure BDA0002289201590000093
a line lightning stroke fault pre-system based on a PSO-LM-BP neural network comprises:
the lightning stroke fault prediction module is used for inputting the lightning stroke information into the PSO-LM-BP neural network model to obtain a prediction result, and judging whether the power transmission line has a lightning stroke fault or not according to the prediction result and the lightning stroke fault threshold value;
the PSO-LM-BP neural network model training step comprises the following steps:
inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, taking a sample corresponding to the power transmission corridor with the minimum mean square error as a final training sample of the neural network, and obtaining the final BP neural network through the final training sample, the particle swarm optimization and the LM algorithm training again.
Inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, wherein the method comprises the following steps of:
taking values of the line corridor width according to a set interval, extracting corresponding lightning information characteristics under each value, wherein N is arranged under each width corridorzPredicting the sample by adopting an initial BP neural network to obtain a new prediction error ea，a＝1,2,…,NzThereby obtaining NzError sequence e of individual samples1,e2,…,eNzAnd calculating the mean square error value e thereofm：
Figure BDA0002289201590000101
The PSO algorithm is provided according to the foraging behavior of the bird swarm, the PSO algorithm is easy to realize, the required adjustment parameters are few, the PSO algorithm is a global search algorithm, when the PSO algorithm is applied to weight optimization of a BP neural network, the PSO algorithm can overcome the defect that local extrema are easy to fall into during training, and global and local optimization are considered. The LM algorithm is an improved form of the Gaussian Newton method, because approximate second derivative information is adopted, the convergence speed is very fast, the stability is good, the advantages of the gradient descent method are combined, and the descent is fast in a plurality of steps at the initial stage of iteration. In conclusion, the BP neural network model is constructed by combining the PSO method and the LM method, the problem of accuracy of the lightning stroke fault prediction of the power transmission line is solved, and compared with the traditional method for evaluating the fault probability of the power transmission line in the lightning occurrence area, the method has higher accuracy and higher iteration speed, and can provide decision basis for line lightning defense.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (10)

1. A lightning stroke fault early warning method for a power transmission line is characterized by comprising the following steps:
inputting the lightning strike information into a PSO-LM-BP neural network model to obtain a prediction result, and judging whether the power transmission line has lightning strike faults or not according to the prediction result and the lightning strike fault threshold value;
the PSO-LM-BP neural network model training step comprises the following steps:
inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, taking a sample corresponding to the power transmission corridor with the minimum mean square error as a final training sample of the neural network, and obtaining the final BP neural network through the final training sample, the particle swarm optimization and the LM algorithm training again.
2. The power transmission line lightning stroke fault early warning method according to claim 1, wherein the lightning strike information comprises:
number of landmine in line corridor X1(ii) a Distance X between lightning strike nearest to line and line2(ii) a Average lightning current intensity X in line corridor3(ii) a Lightning current intensity X of the lightning falling nearest to the line4。
3. The lightning stroke fault early warning method for the power transmission line according to claim 1, wherein the generation process of the fault sample and the normal sample is as follows:
dividing the thunderstorm process into discrete time periods according to a set time step for each thunderstorm, taking the counted thunderstorm information in one time step and the lightning stroke fault information in the next time step as a sample, and generating a fault sample;
and generating an original normal sample when the line has the lightning information but does not have a fault, and generating the normal sample by adopting a random down-sampling method for the original normal sample.
4. The power transmission line lightning stroke fault early warning method according to claim 1, wherein the initial BP neural network is obtained through particle swarm algorithm and LM algorithm training, and the method comprises the following steps:
optimizing the initial weight between the input layer neuron and the hidden layer neuron of the BP neural network by applying a particle swarm optimization algorithm;
and (3) training the BP neural network by applying an LM algorithm.
5. The lightning stroke fault early warning method of the power transmission line according to claim 4, characterized in that the initial weight between the neural neurons of the input layer and the hidden layer of the BP neural network is optimized by applying a particle swarm optimization algorithm, comprising the following steps:
setting the initial population as a plurality of pq + q-dimensional row vectors, wherein each particle contains all weight information of the network, and expressing the fitness value of each particle by using the mean square error e of the BP neural network:
Figure FDA0002289201580000011
in the formula, p is the number of neurons of the input layer of the BP neural network, q is the number of neurons of the hidden layer of the BP neural network, n is the number of samples, i is the sample serial number, and i is 1,2, n; y isiFor the ith sample xiThe output, y, of the output layer is obtained by the forward propagation layer by layer of the networki' As the actual output of the sample, the optimal position P is determined by tracking the individual history in each iterationg kAnd group historical best position Pq kUpdating the speed and the position of the particle to obtain the optimal fitness e of the particle corresponding to the minimum mean square error in the population;
the update formulas of speed and position are respectively:
Figure FDA0002289201580000021
Figure FDA0002289201580000022
in the formula, Vi kThe moving speed of the ith particle in the kth iteration; pper kFor the historical best position of the individual in the kth iteration, Pgro kFor the historical best position in the kth iteration of the population, c1And c2Taking a non-negative constant as an acceleration factor; r is1And r2Is a random number between 0 and 1, Xi kIs the position in the kth iteration of the ith particle; w is a variable inertial weight that decreases linearly with the number of iterations:
Figure FDA0002289201580000023
in the formula, wmaxAnd wminMaximum and minimum inertial weight, k, respectivelymaxIs the maximum iteration number;
recalculating the fitness value e of the new particle, updating Pper kAnd Pgro kK is k +1, when the number of iterations reaches kmaxAnd stopping iteration, otherwise, updating the speed and the position of the particles again for calculation to obtain the optimal fitness of the particles corresponding to the minimum mean square error of the population, and taking the value of the variable inertial weight corresponding to the particles with the optimal fitness of the population as the initial weight of the BP neural network.
6. The power transmission line lightning stroke fault early warning method according to claim 1, characterized in that an LM algorithm is applied to train a BP neural network:
let the error indicator function E of the network be:
Figure FDA0002289201580000024
where N is the number of neurons in the output layer, s is 1,2, …, and N is 1, e in this examples(u) is the error of the s-th output neuron, and u is the vector formed by the mean square error threshold and the weight, then:
▽E＝JT(u)e(u)
where E (u) is the mean square error column vector of the neural network, ▽ E is the gradient of the error index function E, J (u) is the Jacobian matrix whose expansion is:
Figure FDA0002289201580000031
u1,…,uNvectors consisting of threshold values and weight values of 1 st to N times of neural network iteration respectively;
the weight correction formula of the LM algorithm is as follows:
xv+1＝xv-[JT(x)J(x)+μI]-1J(x)e(x)
in the formula, xvIs a vector formed by the weights of the neural network in the v iteration, wherein I is an identity matrix and mu is a proportionality coefficient.
Calculating the current mean square error evIf e isv<e0Or v>vmax，e0Is the target mean square error, vmaxIf the set iteration threshold is reached, the training is ended, otherwise, the weight is adjusted according to the weight correction formula, and the adjusted error value e is calculatedv+1If e isv>ev+1If not, the adjustment is invalid, the mu is enabled to be mu/β, the fitness value of each particle is recalculated, and the historical optimal position P of each particle is recordedg kAnd group historical best position Pq kAnd finally obtaining the trained BP neural network model.
7. The lightning stroke fault early warning method of the power transmission line according to claim 1, wherein the faults and normal samples under the power transmission corridors with different widths are input into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, and the method comprises the following steps:
taking values of the line corridor width according to a set interval, extracting corresponding lightning information characteristics under each value, wherein N is arranged under each width corridorzPredicting the sample by adopting an initial BP neural network to obtain a new prediction error ea，a＝1,2,…,NzThereby obtaining NzError sequence e of individual samples1,e2,…,eNzAnd calculating the mean square error value e thereofm：
Figure FDA0002289201580000032
8. The power transmission line lightning stroke fault early warning method according to claim 1, characterized by further comprising: evaluating the prediction result by the evaluation index, the steps comprising:
defining lightning stroke faults as positive classes, and enabling normal operation to be negative classes; when the lightning stroke fault is correctly predicted, the method is called as a true TP class; when the lightning stroke fault is mispredicted, the lightning stroke fault is called false negative FN; when the normal operation is mispredicted, the operation is called false positive class FP; when normal operation is correctly predicted, the normal operation is called true negative type TN; the effect of lightning strike fault prediction can be determined by the true class rate TPR or hit rate, the negative and positive class rate FPR or false alarm rate, and the accuracy ACC:
Figure FDA0002289201580000041
Figure FDA0002289201580000042
Figure FDA0002289201580000043
9. the utility model provides a transmission line thunderbolt trouble early warning system which characterized in that includes:
the lightning stroke fault prediction module is used for inputting the lightning stroke information into the PSO-LM-BP neural network model to obtain a prediction result, and judging whether the power transmission line has a lightning stroke fault or not according to the prediction result and the lightning stroke fault threshold value;
the PSO-LM-BP neural network model training step comprises the following steps:
inputting fault and normal samples under power transmission corridors with different widths into an initial BP neural network obtained through particle swarm optimization and LM algorithm training to obtain each mean square error, taking a sample corresponding to the power transmission corridor with the minimum mean square error as a final training sample of the neural network, and obtaining the final BP neural network through the final training sample, the particle swarm optimization and the LM algorithm training again.
10. The lightning stroke fault early warning system of the power transmission line according to claim 9, wherein the faults and normal samples under the power transmission corridors with different widths are input into the initial BP neural network obtained through the particle swarm algorithm and LM algorithm training to obtain each mean square error, and the steps comprise:
taking values of the line corridor width according to a set interval, extracting corresponding lightning information characteristics under each value, wherein N is arranged under each width corridorzPredicting the sample by adopting an initial BP neural network to obtain a new prediction error ea，a＝1,2,…,NzThereby obtaining NzError sequence e of individual samples1,e2,…,eNzAnd calculating the mean square error value e thereofm：
Figure FDA0002289201580000044
CN201911172867.4A 2019-11-26 2019-11-26 Lightning stroke fault early warning method and system for power transmission line Active CN110941902B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911172867.4A CN110941902B (en) 2019-11-26 2019-11-26 Lightning stroke fault early warning method and system for power transmission line

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911172867.4A CN110941902B (en) 2019-11-26 2019-11-26 Lightning stroke fault early warning method and system for power transmission line

Publications (2)

Publication Number Publication Date
CN110941902A true CN110941902A (en) 2020-03-31
CN110941902B CN110941902B (en) 2023-09-12

Family

ID=69908267

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911172867.4A Active CN110941902B (en) 2019-11-26 2019-11-26 Lightning stroke fault early warning method and system for power transmission line

Country Status (1)

Country Link
CN (1) CN110941902B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112149898A (en) * 2020-09-21 2020-12-29 广东电网有限责任公司清远供电局 Fault rate prediction model training method, fault rate prediction method and related device
CN113139606A (en) * 2021-04-27 2021-07-20 上海联创设计集团股份有限公司 Heating and ventilation fan fault detection method and system based on artificial intelligence algorithm
CN113151989A (en) * 2021-04-19 2021-07-23 山东大学 Cloth processing method, cloth processing system and sewing robot
CN113295935A (en) * 2021-06-01 2021-08-24 兰州资源环境职业技术学院 Lightning stroke risk assessment method based on high-precision lightning positioning technology

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07301656A (en) * 1994-05-02 1995-11-14 Sumitomo Electric Ind Ltd Transmission line failure block detecting method using neural network
CN102680837A (en) * 2012-05-31 2012-09-19 广州供电局有限公司 Method and device for assessing induction lightning trip-out risk of distributing circuit surrounding transmission line
CN106124830A (en) * 2016-07-11 2016-11-16 清华大学 The method obtaining lightning current based on artificial neural network
CN107290623A (en) * 2017-05-08 2017-10-24 广西电网有限责任公司电力科学研究院 A kind of transmission line malfunction reason recognition methods based on Auto-matching
CN108108520A (en) * 2017-11-29 2018-06-01 海南电网有限责任公司电力科学研究院 A kind of transmission line of electricity damage to crops caused by thunder risk forecast model based on Artificial neural network ensemble
CN108108877A (en) * 2017-11-29 2018-06-01 海南电网有限责任公司电力科学研究院 A kind of transmission line of electricity damage to crops caused by thunder methods of risk assessment based on BP neural network
CN108764452A (en) * 2018-06-01 2018-11-06 福州大学 A kind of Chaos particle swarm optimization algorithm based on the mechanism of discussion
CN110110839A (en) * 2019-04-19 2019-08-09 燕山大学 Cold rolling of strip steel lateral thickness difference forecasting procedure based on PSO-LM-BP neural network
CN110189011A (en) * 2019-05-23 2019-08-30 国电南瑞科技股份有限公司 Dense transmission channel assessment of failure method and system under natural calamity
CN110232203A (en) * 2019-04-22 2019-09-13 山东大学 Knowledge distillation optimization RNN has a power failure prediction technique, storage medium and equipment in short term
CN110298444A (en) * 2019-05-22 2019-10-01 东南大学 A method of it improves based on Hall effect magnetic coder output angle precision

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07301656A (en) * 1994-05-02 1995-11-14 Sumitomo Electric Ind Ltd Transmission line failure block detecting method using neural network
CN102680837A (en) * 2012-05-31 2012-09-19 广州供电局有限公司 Method and device for assessing induction lightning trip-out risk of distributing circuit surrounding transmission line
CN106124830A (en) * 2016-07-11 2016-11-16 清华大学 The method obtaining lightning current based on artificial neural network
CN107290623A (en) * 2017-05-08 2017-10-24 广西电网有限责任公司电力科学研究院 A kind of transmission line malfunction reason recognition methods based on Auto-matching
CN108108520A (en) * 2017-11-29 2018-06-01 海南电网有限责任公司电力科学研究院 A kind of transmission line of electricity damage to crops caused by thunder risk forecast model based on Artificial neural network ensemble
CN108108877A (en) * 2017-11-29 2018-06-01 海南电网有限责任公司电力科学研究院 A kind of transmission line of electricity damage to crops caused by thunder methods of risk assessment based on BP neural network
CN108764452A (en) * 2018-06-01 2018-11-06 福州大学 A kind of Chaos particle swarm optimization algorithm based on the mechanism of discussion
CN110110839A (en) * 2019-04-19 2019-08-09 燕山大学 Cold rolling of strip steel lateral thickness difference forecasting procedure based on PSO-LM-BP neural network
CN110232203A (en) * 2019-04-22 2019-09-13 山东大学 Knowledge distillation optimization RNN has a power failure prediction technique, storage medium and equipment in short term
CN110298444A (en) * 2019-05-22 2019-10-01 东南大学 A method of it improves based on Hall effect magnetic coder output angle precision
CN110189011A (en) * 2019-05-23 2019-08-30 国电南瑞科技股份有限公司 Dense transmission channel assessment of failure method and system under natural calamity

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
崔月生: "基于PSO-BP算法的雷电流概率预测研究" *
沈海实 等: "基于神经网络技术的输电线路走廊雷电活动参数预测分析" *
王炳萱 等: "基于LM-PSO算法和BP神经网络的非线性预测控制" *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112149898A (en) * 2020-09-21 2020-12-29 广东电网有限责任公司清远供电局 Fault rate prediction model training method, fault rate prediction method and related device
CN112149898B (en) * 2020-09-21 2023-10-31 广东电网有限责任公司清远供电局 Training of failure rate prediction model, failure rate prediction method and related device
CN113151989A (en) * 2021-04-19 2021-07-23 山东大学 Cloth processing method, cloth processing system and sewing robot
CN113139606A (en) * 2021-04-27 2021-07-20 上海联创设计集团股份有限公司 Heating and ventilation fan fault detection method and system based on artificial intelligence algorithm
CN113295935A (en) * 2021-06-01 2021-08-24 兰州资源环境职业技术学院 Lightning stroke risk assessment method based on high-precision lightning positioning technology

Also Published As

Publication number Publication date
CN110941902B (en) 2023-09-12

Similar Documents

Publication Publication Date Title
CN110941902B (en) Lightning stroke fault early warning method and system for power transmission line
CN110348615B (en) Cable line fault probability prediction method based on ant colony optimization support vector machine
CN109214566A (en) Short-term wind power prediction method based on shot and long term memory network
CN109766583A (en) Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique
CN110794308B (en) Method and device for predicting train battery capacity
CN109143408B (en) Dynamic region combined short-time rainfall forecasting method based on MLP
Suryo et al. Improved time series prediction using LSTM neural network for smart agriculture application
CN112149898A (en) Fault rate prediction model training method, fault rate prediction method and related device
CN114004102A (en) Reservoir real-time flood control multi-objective robust optimization regulation and control method for resisting flood forecast error disturbance
CN104986347A (en) Real-time detection method for civil aircraft airline pilot operation errors
CN110457737A (en) A method of pollution entering the water is quickly positioned based on neural network
CN108647772B (en) Method for removing gross errors of slope monitoring data
CN114648170A (en) Reservoir water level prediction early warning method and system based on hybrid deep learning model
KR101725629B1 (en) System and method for predicting vehicular traffic based on genetic programming using fitness function considering error magnitude
CN107273971B (en) Feed-forward neural network structure self-organization method based on neuron significance
CN105335763A (en) Fabric defect classification method based on improved extreme learning machine
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
CN110163332B (en) Transformer fault diagnosis method
WO2014173271A1 (en) Optimization method and system for the number of monitoring units of digital man-machine interface
Rosli et al. Intelligent prediction system for gas metering system using particle swarm optimization in training neural network
CN112926664B (en) Feature selection and CART forest short-time strong precipitation prediction method based on evolutionary algorithm
CN115936773A (en) Internet financial black product identification method and system
Hoomod et al. Temperature Prediction Using Recurrent Neural Network for Internet of Things Room Controlling Application
KR102289396B1 (en) Application of reinforcement learning for the advancement of forecasting item demand of repair parts of military equipment
CN113554099A (en) Method and device for identifying abnormal commercial tenant

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