CN111488896A - Distribution line time-varying fault probability calculation method based on multi-source data mining - Google Patents

Abstract

The invention discloses a distribution line time-varying fault probability calculation method based on multi-source data mining, which comprises the following steps of: determining historical fault data of the power distribution network, historical meteorological data and the geographical position of a transformer substation to which each feeder line belongs; screening out data of faults caused by severe weather as a training sample set; respectively counting the fault rate and the repair rate of the distribution line in normal weather and severe weather; constructing an SVM meteorological classifier according to the training sample set and an SVM method; modeling by using a Fock-Planck equation; and determining the time-varying fault probability of the distribution line obtained by current meteorological calculation. The method and the system fully consider the influence of severe meteorological factors on the fault of the distribution line based on the reliability statistical data and the historical meteorological data stored in the power distribution network informatization system, realize automatic and accurate calculation and analysis of the fault probability of the distribution line, and can effectively and accurately calculate the time-varying fault probability of the distribution line.

Description

A time-varying fault probability calculation method for distribution lines based on multi-source data mining

technical field

The invention relates to the technical field of power system analysis, in particular to a method for calculating the time-varying fault probability of distribution lines based on multi-source data mining.

Background technique

With the development of the distribution network, the grid structure has become increasingly complex, the penetration rate of renewable energy in the distribution network has continued to increase, and the distribution line failures caused by different types of risk sources have brought many problems to the operation of the distribution network. Challenges, risk assessment has become an indispensable link in the operation of the distribution network.

Accurate calculation of distribution line failure probability is the basis of operation risk assessment. The existing methods for calculating distribution line failure probability are mainly divided into component outage steady-state probability method, random fuzzy variable modeling method, manual scoring method and fuzzy expert system. Law. The component outage steady-state probability method expresses the fault probability of the distribution line through the steady-state probability in the Markov process. This method fails to characterize the time-varying characteristics of the distribution line fault during operation. In the risk assessment of distribution network operation, the time-varying value reflecting different operating conditions should be used to represent the failure probability; the existing random fuzzy variable modeling method, combined with expert experience, is used to calculate the time-varying failure probability of different lines.

However, the use of random fuzzy variable modeling, manual scoring method and fuzzy reasoning method has strong subjectivity, which may bring certain errors to the calculation of distribution line failure probability. In addition, some scholars have proposed time-varying fault probability calculation methods for distribution lines based on different types of risk sources. However, distribution lines will be affected by a variety of risk sources at the same time during operation, such as severe weather such as lightning strikes and strong winds, as well as their own factors such as aging and defects of distribution lines. Considering only a single source of risk in the above studies is not comprehensive.

In view of this, it is urgent to provide a time-varying fault probability calculation method for distribution line operation risk based on multi-source data such as external weather and the state of the line itself, which truly reflects the time-varying fault probability of distribution lines under different operating conditions. Provide reference for dispatching operators to evaluate system operation risks.

SUMMARY OF THE INVENTION

In order to solve the above-mentioned technical problems, the technical solution adopted in the present invention is to provide a method for calculating the time-varying fault probability of distribution lines based on multi-source data mining, which includes the following steps:

S1. According to the collected power grid data, determine the historical fault data of the distribution network, historical meteorological data and the geographic location of the substations to which each feeder belongs;

S2. Screen out the fault data caused by bad weather according to the historical fault data, and find the corresponding meteorological data in the historical meteorological data as a training sample set;

S3. Count the failure rate and repair rate of distribution lines under normal weather and the failure rate and repair rate under bad weather respectively;

S4. According to the training sample set and the SVM method, construct an SVM meteorological classifier;

S5, using the Fock-Planck equation to model the state transition process of the distribution line in step S3;

S6, determine the current weather, and determine the failure rate and repair rate under the current operating state according to step S4;

S7. Substitute the failure rate and repair rate determined in step S6 into the Fock-Planck equation, and calculate the time-varying failure probability of the distribution line;

Among them, the historical meteorological data includes: daily average temperature, daily maximum/low temperature, daily average wind speed, daily cumulative precipitation, daily cumulative lightning strikes; historical fault data screened out due to thunderstorms, strong winds, and bad weather data covered by ice.

In the above method, the step S2 includes:

The distribution line operation training sample set is:

S={(x _i , y _i ), x _i ∈ R ⁿ , y _i ∈ {0, 1}}

In the formula, X _i represents the i-th n-dimensional input vector, including the meteorological data of temperature, wind speed, precipitation, lightning strike, and humidity;

y _i is the category label, which is divided according to whether the current distribution line operating state has a fault; y _i = 0 means that the distribution line has no fault in the current operating state, y _i = 1 means that the number of faults on the line on the day is greater than 0, indicates a failure.

In the above method, the step S3 includes:

By classifying and statistic on the fault-prone data and the non-fault data in step S2, the state transition rate λ ₀₁ in the fault-prone environment can be obtained respectively as shown in the following formula:

In the formula, N _Lji represents the i-th fault that occurs on the j-th line, and T _L represents the duration of this operating condition;

The repair rate μ ₀₁ is as follows:

In the formula, R _Lji represents the i-th repair of the jth line, and _DL represents the duration of this operating condition.

In the above method, the step S4 includes:

The classification plane equation is constructed as follows:

f(x)=＜ω·x＞+b

In the formula, ω=(ω ₁ , ω ₁ ... ω _n ) is the weight vector, x=(x ₁ , x ₂ ... x _m ) ^T ; b is the threshold, and the training error ε is used as the constraint condition, And introduce slack variables, so that the optimization problem of maximizing the separation surface of the two types of samples can be expressed as:

st y _i [＜ω·x＞+b]≥1-ξ _i

0≤α _i ≤C,ξ _i ≥0,i=1,2...,l

In the formula, C>0 is the penalty coefficient, and the larger the value, the greater the penalty for the data points exceeding ε.

In the above method, the state transition process of the distribution line in the step S5 is modeled as follows:

P'(t)=P(t)Q

In the formula, P ₀ is the probability that the feeder is in normal operation, and P ₁ is the probability that the feeder is in a fault state.

The invention proposes a time-varying fault probability calculation method based on multi-source data mining for distribution line operation risk. The influence of electric line faults realizes the automatic and accurate calculation and analysis of distribution line failure probability, and can effectively and accurately calculate the time-varying failure probability of distribution lines. This method avoids the error caused by the dispatcher's judgment based on their own experience, can objectively reflect the failure probability of the distribution line in different meteorological environments, and realize the accurate identification and analysis of the operation risk of the distribution network.

Description of drawings

Fig. 1 is the flow chart provided by the present invention;

FIG. 2 is a schematic flowchart of the time-varying fault probability calculation process of the operation risk of the distribution line provided by the present invention;

FIG. 3 is a schematic diagram of the framework of the fault-repair two-state cycle process of the distribution line provided by the present invention.

Detailed ways

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

As shown in Figure 1-2, the present invention provides a method for calculating the time-varying fault probability of distribution line operation risk based on multi-source data mining, including the following steps:

S1. According to the collected power grid data, determine the historical fault data, historical meteorological data, historical load data of the distribution network and the geographic location of the substation to which each feeder belongs;

Historical meteorological data include: daily average temperature, daily maximum/low temperature, daily average wind speed, daily cumulative precipitation, daily cumulative lightning strikes, daily average solar radiation, daily average pressure, daily specific humidity, daily relative humidity and other historical data. From the historical fault data, the fault data caused by severe weather such as thunderstorms, strong winds, and icing are screened out, and the corresponding meteorological data are found in the historical meteorological data as training samples.

S2. Screen out the data of faults caused by bad weather according to the historical fault data, and find the corresponding meteorological data in the historical meteorological data as a training sample set.

The following formula represents the distribution line training sample set:

S={(x _i ,y _i ), x _i ∈R ⁿ ,y _i ∈{0,1}} (1)

In the formula, x _i represents the i-th n-dimensional input vector, including meteorological data such as temperature, wind speed, precipitation, lightning strike, and humidity. y _i is the category label, which is divided according to whether the current distribution line operating state has a fault; the label y _i = 0 indicates that the distribution line has no fault in the current operating state, and the label y _i = 1 indicates that the line has failed on the day. A number greater than 0 indicates a failure.

S3. Count the failure rate and repair rate of distribution lines in normal weather and the failure rate and repair rate in bad weather respectively. As shown in Figure 3, no matter in normal weather or bad weather, the fault-repair two-state cycle process of the distribution line, where λ ₀₁ is the change of the distribution line from the normal operation state to the The failure rate of the fault state is a value that changes with time and the operating conditions of the distribution line; μ ₀₁ is the repair rate of the distribution line from the fault state to the normal operating state.

By classifying and statistic on the data with faults and the data without faults in step S2, the failure rate _λ01 in the fault-prone environment can be obtained respectively as shown in the following formula:

In the formula, N _Lji represents the i-th fault that occurs on the j-th line, and _TL represents the duration of this operating condition.

The repair rate μ ₀₁ is as follows:

In the formula, R _Lji represents the i-th repair of the jth line, and _DL represents the duration of this operating condition.

S4. According to the training sample set and the SVM (Support Vector Machine) method, construct an SVM meteorological classifier, including:

The classification plane equation is constructed as follows:

f(x)=<ω·x>+b (4)

In the formula, ω=(ω ₁ , ω ₁ ... ω _n ) is the weight vector, x=(x ₁ , x ₂ ... x _m ) ^T ; b is the threshold, and the training error ε is used as the constraint condition, And the slack variable ξ _i is introduced, so that the optimization problem of maximizing the separation surface of the two types of samples can be expressed as:

In the formula, C>0 is the penalty coefficient, and the larger the value, the greater the penalty for the data points exceeding ε.

The above problem is transformed into a quadratic programming problem with linear inequality constraints, and the Lagrange multiplier method is used to solve the following quadratic programming problem with linear inequality constraints:

为拉格朗日乘子。In the formula,

is the Lagrange multiplier.

The above optimization problem can be solved by using the Lagrange multiplier method; by introducing the kernel function k(x, x _i ), the objective function becomes:

where α _i , α _j ≥ 0 (i,j=1,2,...,l) are Lagrange multipliers, x _i is the ith sample, x _j is the jth sample, and i=1,2,...,l, j=1,2,...,l.

Among all the kernel functions, the radial basis function has a good effect on dealing with nonlinear problems, and the parameters to be determined are less than that of the polynomial kernel function, the radial basis function K(x _i ,x) The form is as follows The formula shows:

Among them, the parameters that affect the training effect are mainly C and σ. In the parameter selection of SVM, the grid search method is adopted in this embodiment, and m and n values of C and σ are respectively used for training, and the accurate prediction is selected according to the model training result. A set of parameters with the highest rate is used as the parameters of the kernel function in the SVM model.

S5. Use the Fock-Planck equation to model the state transition process of the distribution line in step S3, as shown in the following formula:

P'(t)=P(t)Q (9)

In the formula, P ₀ is the probability that the feeder is in normal operation, and P ₁ is the probability that the feeder is in a fault state.

S6. Determine the current weather, and determine the failure rate λ ₀₁ and the repair rate μ ₀₁ in the current operating state according to step S4 .

S7. Substitute the values of the failure rate λ ₀₁ and the repair rate μ ₀₁ determined in step S6 into the Fock-Planck equation, and calculate the time-varying failure probability P of the distribution line.

The invention proposes a time-varying fault probability calculation method based on multi-source data mining for distribution line operation risk. The influence of electric line faults realizes the automatic and accurate calculation and analysis of distribution line failure probability, and can effectively and accurately calculate the time-varying failure probability of distribution lines. This method avoids the error caused by the dispatcher's judgment based on their own experience, can objectively reflect the failure probability of the distribution line in different meteorological environments, and realize the accurate identification and analysis of the operation risk of the distribution network.

The present embodiment is described below through a specific case

In this case, a total of 828 statistical data of power accidents from January 2014 to July 2015 were collected for 120 feeders in a southern region. According to the fault characteristic variable selection method proposed in the above embodiment, the fault characteristic variable is selected, and the obtained optimal fault characteristic variable is shown in Table 1, and the optimal fault characteristic subset evaluation value is 0.722.

Table 1. Optimal fault characteristic variables

type Fault Characteristic Variables Feeder fault characteristics Fault occurrence time, substation to which the feeder belongs external influences Daily maximum temperature, daily minimum temperature, daily average humidity, daily average wind speed own influence factors Line length Operational Influencing Factors Line load

By using the time-varying fault probability calculation method proposed in the above embodiment, the time-varying fault probability of all distribution lines under severe weather can be calculated. The time-varying fault probability and load loss risk value of some distribution lines are shown in Table 2 and Table 3.

Table 2. Time-varying fault probability of some lines

Table 3. Loss of load risk value of some lines

The present invention is not limited to the above-mentioned best embodiment, and anyone should know that structural changes made under the inspiration of the present invention, and all technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention.

Claims (5)

1. a time-varying fault probability calculation method for distribution lines based on multi-source data mining, is characterized in that, comprises the following steps: S1. According to the collected power grid data, determine the historical fault data of the distribution network, historical meteorological data and the geographic location of the substation to which each feeder belongs; S2. Screen out the fault data caused by bad weather according to the historical fault data, and find the corresponding meteorological data in the historical meteorological data as a training sample set; S3. Count the failure rate and repair rate of distribution lines under normal weather and the failure rate and repair rate under bad weather respectively; S4. According to the training sample set and the SVM method, construct an SVM meteorological classifier; S5, using the Fock-Planck equation to model the state transition process of the distribution line in step S3; S6, determine the current weather, and determine the failure rate and repair rate under the current operating state according to step S4; S7. Substitute the failure rate and repair rate determined in step S6 into the Fock-Planck equation, and calculate the time-varying failure probability of the distribution line; Among them, the historical meteorological data includes: daily average temperature, daily maximum/low temperature, daily average wind speed, daily cumulative precipitation, daily cumulative lightning strike; historical fault data screened out due to thunderstorms, strong winds, and bad weather data covered by ice. 2. The computing method according to claim 1, wherein the step S2 comprises: The distribution line operation training sample set is: S={(x _i , y _i ), x _i ∈ R ⁿ , y _i ∈ {0, 1}} In the formula, x _i represents the i-th n-dimensional input vector, including the meteorological data of temperature, wind speed, precipitation, lightning strike, and humidity; y _i is the category label, which is divided according to whether the current distribution line operating state has a fault; y _i = 0 means that the distribution line has no fault in the current operating state, y _i = 1 means that the number of faults on the line on the day is greater than 0, indicates a failure. 3. The computing method according to claim 2, wherein the step S3 comprises: By classifying and statistic on the fault-prone data and the non-fault data in step S2, the state transition rate λ ₀₁ in the fault-prone environment can be obtained respectively as shown in the following formula:

In the formula, N _Lji represents the i-th fault that occurs on the j-th line, and T _L represents the duration of this operating condition; The repair rate μ ₀₁ is as follows:

In the formula, R _Lji represents the i-th repair of the jth line, and _DL represents the duration of this operating condition. 4. The computing method according to claim 3, wherein the step S4 comprises: The classification plane equation is constructed as follows: y=<ω·x>+b In the formula, ω=(ω ₁ , ω ₁ ... ω _n ) is the weight vector, x=(x ₁ , x ₂ ... x _m ) ^T ; b is the threshold, and the training error ε is used as the constraint condition, And the slack variable ξ is introduced, so that the optimization problem of maximizing the separation surface of the two types of samples can be expressed as:

sty _i [<ω·x>+b]≥1-ξ _i

0≤α _i ≤C, ξ _i ≥0, i=1, 2...,l In the formula, C>0 is the penalty coefficient, and the larger the value, the greater the penalty for the data points exceeding ε. 5. The calculation method according to claim 1, characterized in that, in the step S5, the state transition process of the distribution line is modeled as follows:

P'(t)=P(t)Q In the formula, P ₀ is the probability that the feeder is in normal operation, and P ₁ is the probability that the feeder is in a fault state.

Images