CN109613402B - Detection method of high-resistance grounding fault in distribution network based on wavelet transform and neural network - Google Patents

Abstract

The invention relates to a high-resistance earth fault detection method for a power distribution network based on wavelet transformation and a neural network. The invention uses the evolved neural network to improve the traditional detection method. The evolved neural network is an intelligent system based on a dynamic connection structure, and the topological structure of the system can be adjusted through incremental learning so as to incorporate new information. The invention utilizes discrete wavelet transform to process fault signals, and inputs the fault signals into an evolved neural network so as to detect the high-resistance grounding fault of the power distribution network.

Description

Detection method of high-resistance grounding fault in distribution network based on wavelet transform and neural network

technical field

The invention relates to grounding fault detection of distribution network, in particular to a high-resistance grounding fault detection method of distribution network based on wavelet transform and neural network.

Background technique

High resistance ground fault (HIF) refers to the ground fault that occurs when the power line passes through conductive media such as roads, soil, branches or cement buildings. It may occur at various voltage levels and affect the normal operation of the distribution network. Due to the high impedance characteristics of non-metallic conductive media, when a high-resistance grounding fault occurs in the distribution network, the fault current is very small, and often accompanied by arcs, and ordinary zero-sequence current protection is difficult to detect. Among the high-resistance grounding faults of low-current grounding systems, especially resonant grounding systems, arc-flash grounding faults account for a large part. Due to the free air, the grounding impedance of the arc grounding varies greatly, causing the existing protection to be repeatedly activated and restored, which may cause the protection of adjacent lines and equipment to overstep the trip, and also cause overvoltage in the whole system, thereby damaging the electrical equipment. To expand the accident, reduce the power supply reliability of the power grid. The long-term operation of the line with faults may cause the temperature of the fault point to be too high, thereby causing a fire and causing permanent damage to the electrical equipment, and the step voltage around the grounding point can reach several thousand volts, which seriously threatens the stable operation of the power system and personal safety. Effective detection of high-impedance grounding faults can provide criteria for subsequent line selection and location, so high-impedance grounding fault detection in distribution networks is very important.

SUMMARY OF THE INVENTION

In view of this, the purpose of the present invention is to provide a high-resistance grounding fault detection method of distribution network based on wavelet transform and neural network, which can effectively detect the high-resistance grounding fault of power distribution network.

To achieve the above object, the present invention adopts the following technical solutions:

A method for detecting high-resistance grounding faults in distribution networks based on wavelet transform and neural network, comprising the following steps:

Step S1: collecting the current signal of the distribution network line to be detected;

Step S2: adopt db4 wavelet as mother wavelet, carry out discrete wavelet transform to line current signal, and obtain reconstructed waveform;

Step S3: set the neural network parameters of evolution;

Step S4: take the reconstructed waveform as the input of the evolved neural network, calculate the Manhattan distance between the input and the kth neuron of the evolution layer, the activation degree of the kth neuron and the error between the output and the expected value;

Step S5: if the activation degree of the kth neuron is less than the preset threshold or the error between the output and the expected value exceeds the preset threshold, a new neuron is placed in (k+1) positions, otherwise the weights are updated;

和

Step S6: Calculate the Manhattan distance between the oth neuron and the pth neuron

and

和

均小于阈值D_thr，则将这两个神经元进行聚合；Step S7: If

and

If both are less than the threshold D _thr , the two neurons are aggregated;

Step S8: Repeat steps S4-S7 until all input quantities are processed, obtain the output result of the evolved neural network, and judge whether there is a high-resistance grounding fault in the distribution network to be detected according to the output.

Further, the sampling frequency of the discrete wavelet transform is 10 kHz, and the sampling time is 0.5 s.

Further, the evolved neural network parameters include a neuron activation level threshold A _thr , an error threshold E _thr , a Manhattan distance threshold D _thr of two neurons, and learning rates α ₁ and α ₂ .

Further, the step S4 parameter calculation is specifically:

The Manhattan distance between the jth input I _j and the kth neuron in the evolution layer is

The activation degree of the kth neuron A _k =1-D _jk , and Wi _ik is the weight;

的误差

Output quantity O ₁ and expected value

error

Further, the step S5 is specifically: if A _k is smaller than A _thr or E ₁ exceeds E _thr , place a new neuron at (k+1) positions, otherwise update the weight _Wik(t+1) = _Wik(t) + α ₁ (I _i(t+1) - _Wik(t) ) and W _k1(t+1) =W _k1(t) +α ₂ (A _k E ₁ ).

和

具体为：Further, the said

and

Specifically:

Among them, W _io , W _ip , W _o1 and W _p1 are weights.

Compared with the prior art, the present invention has the following beneficial effects:

1. The evolved neural network of the present invention has no fixed topology, and can evolve into a new structure online after data processing.

2. The evolved neural network of the present invention can learn rapidly through incremental training, has good generalization ability and the ability to reorganize the network model into a constantly changing environment, wherein the structure and parameters are adapted simultaneously.

3. The evolved neural network of the present invention avoids the occurrence of catastrophic forgetting by adding and deleting neurons online and adjusting the connection weight of neurons, and increases the detection reliability and efficiency of high-resistance grounding faults in the distribution network.

Description of drawings

Fig. 1 is the flow chart of the method of the present invention;

FIG. 2 is a schematic structural diagram of an evolved neural network in an embodiment of the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments.

Referring to FIG. 1, the present invention provides a method for detecting high-resistance grounding faults in distribution networks based on wavelet transform and neural network, comprising the following steps:

Step S1: collecting the current signal of the distribution network line to be detected;

Step S2: adopt db4 wavelet as mother wavelet, carry out discrete wavelet transform to line current signal, and obtain reconstructed waveform;

Step S3: set the neural network parameters of evolution;

Step S4: take the reconstructed waveform as the input of the evolved neural network, calculate the Manhattan distance between the input and the kth neuron of the evolution layer, the activation degree of the kth neuron and the error between the output and the expected value;

Step S5: if the activation degree of the kth neuron is less than the preset threshold or the error between the output and the expected value exceeds the preset threshold, a new neuron is placed in (k+1) positions, otherwise the weights are updated;

和

Step S6: Calculate the Manhattan distance between the oth neuron and the pth neuron

and

和

均小于阈值D_thr，则将这两个神经元进行聚合；Step S7: If

and

If both are less than the threshold D _thr , the two neurons are aggregated;

Step S8: Repeat steps S4-S7 until all input quantities are processed, obtain the output result of the evolved neural network, and judge whether there is a high-resistance grounding fault in the distribution network to be detected according to the output.

In this embodiment, the sampling frequency of the discrete wavelet transform is 10 kHz, and the sampling time is 0.5 s.

In this embodiment, the evolved neural network parameters include a neuron activation level threshold A _thr , an error threshold E _thr , a Manhattan distance threshold D _thr for two neurons, and learning rates α ₁ and α ₂ .

Further, the step S4 parameter calculation is specifically:

The Manhattan distance between the jth input I _j and the kth neuron in the evolution layer is

The activation degree of the kth neuron A _k =1-D _jk , and Wi _ik is the weight;

的误差

Output quantity O ₁ and expected value

error

In this embodiment, the step S5 is specifically: if A _k is smaller than A _thr or E ₁ exceeds E _thr , place a new neuron at (k+1) positions, otherwise update the weight _{Wik(t +1)} = _Wik(t) +α ₁ (I _i(t+1) - _Wik(t) ) and W _k1(t+1) =W _k1(t) +α ₂ (A _k E ₁ ) .

和

具体为：In this embodiment, the

and

Specifically:

Among them, W _io , W _ip , W _o1 and W _p1 are weights.

The above descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

Claims (6)

1. a high-resistance grounding fault detection method for distribution network based on wavelet transform and neural network, is characterized in that, comprises the following steps: Step S1: collecting the current signal of the distribution network line to be detected; Step S2: adopt db4 wavelet as mother wavelet, carry out discrete wavelet transform to line current signal, and obtain reconstructed waveform; Step S3: set the neural network parameters of evolution; Step S4: take the reconstructed waveform as the input of the evolved neural network, calculate the Manhattan distance between the input and the kth neuron of the evolution layer, the activation degree of the kth neuron and the error between the output and the expected value; Step S5: if the activation degree of the kth neuron is less than the preset threshold or the error between the output and the expected value exceeds the preset threshold, a new neuron is placed in (k+1) positions, otherwise the weights are updated; Step S6: Calculate the Manhattan distance between the oth neuron and the pth neuron

and

Step S7: If

and

If both are less than the threshold D _thr , the two neurons are aggregated; Step S8: Repeat steps S4-S7 until all input quantities are processed, obtain the output result of the evolved neural network, and judge whether there is a high-resistance grounding fault in the distribution network to be detected according to the output. 2 . The high-resistance grounding fault detection method for distribution network based on wavelet transform and neural network according to claim 1 , wherein the discrete wavelet transform has a sampling frequency of 10 kHz and a sampling time of 0.5 s. 3 . 3. The method for detecting high-resistance grounding faults in distribution networks based on wavelet transform and neural network according to claim 1, wherein the evolved neural network parameters include neuron activation degree threshold A _thr , error threshold E _thr , the Manhattan distance threshold D _thr for the two neurons, the learning rates α ₁ and α ₂ . 4. The high-resistance grounding fault detection method for power distribution network based on wavelet transform and neural network according to claim 3, characterized in that: the step S4 parameter calculation is specifically: The Manhattan distance between the jth input I _j and the kth neuron in the evolution layer is

The activation degree of the kth neuron A _k =1-D _jk , Wi _ik is the weight; Output quantity O ₁ and expected value

error

5. The method for detecting a high-resistance grounding fault in a distribution network based on wavelet transform and neural network according to claim 4, wherein the step S5 is specifically: if A _k is less than A _thr or E ₁ exceeds E _thr , Place a new neuron at (k+1) positions, otherwise update the weight _Wik(t+1) = _Wik(t) +α ₁ (I _i(t+1) - _Wik(t) ) and W _k1(t+1) =W _k1(t) +α ₂ (A _k E ₁ ). 6. The method for detecting high-resistance grounding faults in distribution network based on wavelet transform and neural network according to claim 4, characterized in that: the

and

Specifically:

Among them, W _io , W _ip , W _o1 and W _p1 are weights.

Images