Power distribution network fault early warning model based on data driving

Abstract

The invention discloses a power distribution network fault early warning model based on data driving. The power distribution network fault early warning model comprises the following construction steps: S1, collecting fault line historical information and power quality monitoring data; s2, analyzing the historical monitoring data, and extracting a monitoring signal; s3, fitting the fault line by using a back propagation neural network, and constructing a fault positioning model; s4, performing optimization judgment on the fault node position based on a genetic algorithm, constructing a fault node judgment model, and obtaining final fault location; s5, inputting current monitoring data, realizing current fault positioning, and outputting a fault positioning result; s6, fault early warning is given out based on the current fault positioning result; according to the invention, the electric energy quality monitoring device installed in the power distribution network is utilized to obtain the monitoring data, the electric energy quality monitoring data and the hidden fault information in the electric energy quality monitoring data are analyzed, and the positioning and early warning of the power distribution network fault can be rapidly realized.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network fault early warning technology, and specifically relates to a data-driven power distribution network fault early warning model. Background Technology

[0002] The distribution network is the intermediate link connecting the power transmission network and users. Its safe and stable operation directly affects the reliability of power supply. Monitoring and analyzing the distribution network, as well as locating and issuing early warnings of faults, can enable staff to quickly troubleshoot and ensure the stable and reliable operation of the distribution network.

[0003] As one of the data sources for power big data, power quality monitoring data is characterized by its large volume, coexistence of structured and semi-structured data, and high measurement and acquisition frequency. Moreover, power quality monitoring devices are widely installed in the power grid, making data acquisition relatively convenient. By analyzing the acquired power quality monitoring data, we can help identify weak links and key nodes in the distribution network and provide fault early warnings. This not only promotes the improvement of power quality but is also very important for the safe and stable operation of the distribution network.

[0004] Therefore, in order to solve the above problems, it is necessary to develop a data-driven early warning model for power distribution networks. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a data-driven distribution network fault early warning model. By using power quality monitoring devices installed in the distribution network to acquire monitoring data, and analyzing the power quality monitoring data and the fault information contained therein, the fault location and early warning of the distribution network can be quickly realized.

[0006] The objective of this invention is achieved as follows: a data-driven distribution network fault early warning model, comprising the following construction steps:

[0007] S1. Collect historical information on faulty lines and form a historical information set of faulty lines; collect power quality monitoring data through multiple power quality monitoring devices installed in the distribution network and form a historical power quality monitoring dataset and current monitoring data.

[0008] S2. Combine historical information of the faulty line with historical power quality monitoring data for analysis, and extract power quality monitoring signals that can characterize the line fault.

[0009] S3. Using the data in the historical information set of the faulted line and the historical monitoring data set of power quality as samples for data training, the power quality monitoring signal as the input signal and the line where the fault is located as the output signal, the backpropagation neural network is used to fit the faulted line and construct the fault location model.

[0010] S4. Based on the genetic algorithm, optimize and identify the location of the fault node, construct the fault node identification model, and obtain the final fault location;

[0011] S5. Input the current monitoring data into the fault location model based on the backpropagation neural network and the fault node location optimization and discrimination model based on the genetic algorithm to realize the current fault location and output the fault location result;

[0012] S6. Issue a fault warning based on the current fault location results.

[0013] Furthermore, in step S3, the power quality monitoring signal is a voltage signal.

[0014] Furthermore, in step S3, the residual voltage amplitude of the voltage sag monitored by each power quality monitoring device at the time of the fault is used as the input quantity to encode the faulty line and use it as the output quantity. A backpropagation neural network that fits the positional relationship between the residual voltage amplitude and the faulty line is established to construct a fault location model.

[0015] Furthermore, in step S3, the minimum mean square error value is used to evaluate the approximation degree during the backpropagation neural network training process.

[0016]

[0017] In the formula, and y ij These are the elements in the expected and actual output matrices of the neural network, respectively. i represents the i-th fault, n represents the number of faults, j represents the j-th line, and q represents the number of lines. The smaller the E value, the better the fit between the power quality monitoring signal and the location of the fault line. A threshold for the E value is set. When the E value is less than the threshold, the training is complete.

[0018] Furthermore, in step S4, the direction of the monitoring signal at the node on the faulty line is encoded. When the direction of the monitoring signal is positive, the encoding is 1, and when it is negative, the encoding is 0. A fitness function is constructed to determine the relationship between the line state and the direction of the monitoring signal at the node. Based on the fitness function, the disturbance energy of the signal at the kth monitoring node is obtained. After obtaining the optimal disturbance position, the fault node is located by decoding the optimal solution of the population.

[0019] Furthermore, in step S4, the direction of the monitoring signal at the node on the faulty line is the direction of the fault current signal.

[0020] Furthermore, the fitness function is expressed as:

[0021]

[0022] In the formula: Q is a large integer, which is taken as 1 here; r k The monitoring status of the k-th monitoring node; n is the number of monitoring nodes; j is the dimension; A k Let be the monitoring signal weight parameters of the k-th monitoring node.

[0023] Furthermore, the disturbance energy of the signal at the kth monitoring node is expressed as:

[0024]

[0025] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0026] (1) By using the power quality monitoring devices already installed in the power distribution network to obtain monitoring data, there is no need to set up additional data acquisition devices. This not only makes it more convenient and faster to obtain monitoring data, but also saves manpower and resources.

[0027] (2) A fault location model is constructed based on the backpropagation neural network and the genetic algorithm. The backpropagation neural network is used to fit the fault line, which can realize the rapid location of the fault line. The genetic algorithm is used to optimize and judge the location of the fault node. Based on the location of the fault line, the fault node of the fault line is quickly determined, making the location faster and more accurate. Attached Figure Description

[0028] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0029] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0030] like Figure 1 As shown, a data-driven distribution network fault early warning model includes the following construction steps:

[0031] S1. Collect historical information of faulty lines and form a historical information set of faulty lines; collect power quality monitoring data through multiple power quality monitoring devices installed in the distribution network and form a historical power quality monitoring dataset and current monitoring data.

[0032] S2. Combine historical information of the faulty line with historical power quality monitoring data for analysis, and extract power quality monitoring signals that can characterize the line fault.

[0033] S3. Using data from the historical information set of the faulty line and the historical power quality monitoring set as samples for data training, and the power quality monitoring signal as the input signal and the line where the fault is located as the output signal, a backpropagation neural network is used to fit the faulty line and construct a fault location model.

[0034] Specifically, in step S3, the power quality monitoring signal is a voltage signal; in step S3, the residual voltage amplitude of the voltage sag monitored by each power quality monitoring device during the fault is used as the input quantity to encode the faulty line and use it as the output quantity. The input quantity is calculated by each layer to obtain the output quantity, the error with the actual faulty line is calculated and backpropagated, and this process is repeated to continuously update the weights and thresholds of each layer, and a backpropagation neural network that fits the positional relationship between the residual voltage amplitude and the faulty line is established to construct a fault location model.

[0035] Specifically, the input sample matrix X is formed by taking the residual voltage amplitudes recorded by p power quality monitoring devices during n faults:

[0036]

[0037] In the formula, x ij Let Y be the residual voltage amplitude recorded by the j-th power quality monitoring device when the i-th fault occurs. Then, the actual faulty line is taken as the expected output matrix Y of the neural network. * Represented as:

[0038]

[0039] In the formula, For Boolean values, when A value of 1 indicates that the i-th fault is located on line j. A value of 0 indicates that the fault is not on that line; using X and Y... * The Levenberg-Marquard algorithm, which has a relatively fast convergence speed, was used for training.

[0040] Specifically, in step S3, the minimum mean square error value is used to evaluate the approximation degree during the backpropagation neural network training process.

[0041]

[0042] In the formula, and y ij These are the elements in the expected and actual output matrices of the neural network, respectively. i represents the i-th fault, n represents the number of faults, j represents the j-th line, and q represents the number of lines. The smaller the E value, the better the fit between the power quality monitoring signal and the location of the fault line. A threshold for the E value is set. When the E value is less than the threshold, the training is complete.

[0043] S4. Based on the genetic algorithm, optimize and identify the location of the fault node, construct the fault node identification model, and obtain the final fault location;

[0044] Specifically, in step S4, the direction of the monitoring signal at the node on the faulty line is encoded. When the direction of the monitoring signal is positive, the encoding is 1, and when it is negative, the encoding is 0. A fitness function is constructed to determine the relationship between the line state and the direction of the monitoring signal at the node. Based on the fitness function, the disturbance energy of the signal at the kth monitoring node is obtained. After obtaining the optimal disturbance position, the fault node is located by decoding the optimal solution of the population.

[0045] Specifically, in step S4, the direction of the monitoring signal at the node on the faulty line is the direction of the fault current signal.

[0046] Specifically, the fitness function is expressed as: a k (s)=Q-∏ i (1-r k )×n∑ j A k In the formula: Q is a large integer, which is taken as 1 here; r k The monitoring status of the k-th monitoring node; n is the number of monitoring nodes; j is the dimension; A k Let be the monitoring signal weight parameters of the k-th monitoring node; the disturbance energy of the signal at the k-th monitoring node is expressed as:

[0047] S5. Input the current monitoring data into the fault location model based on the backpropagation neural network and the fault node location optimization and discrimination model based on the genetic algorithm to realize the current fault location and output the fault location result;

[0048] S6. Issue a fault warning based on the current fault location results.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A data-driven distribution network fault early warning model, characterized in that: The following are the construction steps: S1. Collect historical information on faulty lines and form a historical information set of faulty lines; collect power quality monitoring data through multiple power quality monitoring devices installed in the distribution network and form a historical power quality monitoring dataset and current monitoring data. S2. Combine historical information of the faulty line with historical power quality monitoring data for analysis, and extract power quality monitoring signals that can characterize the line fault. S3. Using the data in the historical information set of the faulted line and the historical monitoring data set of power quality as samples for data training, the power quality monitoring signal as the input signal and the line where the fault is located as the output signal, the backpropagation neural network is used to fit the faulted line and construct the fault location model. S4. Based on the genetic algorithm, optimize and identify the location of the fault node, construct the fault node identification model, and obtain the final fault location; S5. Input the current monitoring data into the fault location model based on the backpropagation neural network and the fault node location optimization and discrimination model based on the genetic algorithm to realize the current fault location and output the fault location result; S6. Issue a fault warning based on the current fault location results.

2. The data-driven distribution network fault early warning model according to claim 1, characterized in that: In step S3, the power quality monitoring signal is a voltage signal.

3. The data-driven distribution network fault early warning model according to claim 2, characterized in that: In step S3, the residual voltage amplitude of the voltage sag monitored by each power quality monitoring device during the fault is used as the input quantity to encode the faulty line and use it as the output quantity. A backpropagation neural network that fits the positional relationship between the residual voltage amplitude and the faulty line is established to construct a fault location model.

4. The data-driven distribution network fault early warning model according to claim 1, characterized in that: In step S3, the minimum mean square error value is used to evaluate the approximation degree during the backpropagation neural network training process. In the formula, and y ij These are the elements in the expected and actual output matrices of the neural network, respectively. i represents the i-th fault, n represents the number of faults, j represents the j-th line, and q represents the number of lines. The smaller the E value, the better the fit between the power quality monitoring signal and the location of the fault line. A threshold for the E value is set. When the E value is less than the threshold, the training is complete.

5. The data-driven distribution network fault early warning model according to claim 1, characterized in that: In step S4, the direction of the monitoring signal at the node on the faulty line is encoded. When the direction of the monitoring signal is positive, the encoding is 1, and when it is negative, the encoding is 0. A fitness function is constructed to determine the relationship between the line state and the direction of the monitoring signal at the node. Based on the fitness function, the disturbance energy of the signal at the kth monitoring node is obtained. After obtaining the optimal disturbance position, the fault node is located by decoding the optimal solution of the population.

6. The data-driven distribution network fault early warning model according to claim 5, characterized in that: In step S4, the direction of the monitoring signal at the node on the faulty line is the direction of the fault current signal.

7. A data-driven distribution network fault early warning model according to claim 5, characterized in that: The fitness function is expressed as follows: In the formula: Q is a large integer, which is taken as 1 here; r k The monitoring status of the k-th monitoring node; n is the number of monitoring nodes; j is the dimension; A k Let be the monitoring signal weight parameters of the k-th monitoring node.

8. The data-driven distribution network fault early warning model according to claim 5, characterized in that: The disturbance energy of the signal at the kth monitoring node is expressed as:

Images