CN113919610A

CN113919610A - ARIMA model construction method and evaluation method for low-voltage transformer area line loss prediction

Info

Publication number: CN113919610A
Application number: CN202010644244.9A
Authority: CN
Inventors: 胡志勇; 黄健钊; 杨秋月; 姬剑; 张延钊; 刘跃宗; 李小双; 王芳; 崔文婷; 胡青; 李维康
Original assignee: Tianjin Chuneng Power Technology Co ltd; Nanzhao Power Supply Co Of State Grid Henan Electric Power Co
Current assignee: Tianjin Chuneng Power Technology Co ltd; Nanzhao Power Supply Co Of State Grid Henan Electric Power Co
Priority date: 2020-07-07
Filing date: 2020-07-07
Publication date: 2022-01-11

Abstract

The invention discloses an ARIMA model construction method and an ARIMA model evaluation method for low-voltage transformer area line loss prediction, which belong to the technical field of power grid management and are characterized by comprising the following steps: s1, preprocessing at least one collected measurement data in the power system or a group of time sequence data in a database; s2, eliminating periodic influence on the preprocessed time sequence data by adopting a periodic difference operator, carrying out ADF unit root stationarity inspection on the processed data, and eliminating the growth trend by adopting difference operation if the inspection is not passed; s3, determining the order of the ARIMA model through the Chichi information criterion, and constructing the ARIMA model and an error prediction model thereof; and S4, self-correcting the ARIMA model according to the network loss data residual error predicted by the prediction model, and outputting the model after the residual error precision is met. The invention can save a large amount of manpower and material resources and is beneficial to optimizing the network structure of the power distribution network.

Description

ARIMA model construction method and evaluation method for low-voltage transformer area line loss prediction

Technical Field

The invention belongs to the technical field of power grid management, and particularly relates to an ARIMA model construction method and an ARIMA model evaluation method for line loss prediction of a low-voltage transformer area.

Background

In recent years, with the rapid development of industrial construction, the demand for energy from human beings has been increasing. The electric energy is used as an important energy source for supporting the daily life and the economic development of the nation, and plays a vital role in the national development. However, the electric energy has inevitable losses in the production, transportation and distribution links, which not only results in resource waste, but also causes huge economic losses. Line loss, i.e. line loss, is short for power grid power loss, and is loss generated in the processes of transmission, configuration and the like of power. The line loss can reflect the rationality of the power grid structure and operation, and the technology and operation management level of the power enterprise, so that the line loss is not only an economic and technical index for the energy supply enterprise, but also an important index for the national assessment of the power enterprise.

The low-voltage transformer area is a link at the tail end of the power system, the voltage level is low, the topological structure is incomplete, and the basic set-benefit construction and the power detection equipment of each transformer area are relatively lagged behind. If the theoretical line loss rate of the transformer area is calculated by adopting the equivalent resistance method, the trend iteration method and other traditional methods, the result may have larger difference with the actual line loss value. Therefore, a rapid and accurate low-voltage transformer area line loss prediction model is established, a large amount of manpower and material resources can be saved, the optimization of the network structure of the power distribution network is facilitated, and the method has important significance for power supply enterprises to formulate reasonable loss reduction measures.

In conclusion, the ARIMA model construction method and the evaluation method for predicting the line loss of the low-voltage transformer area have great practical significance.

Disclosure of Invention

The invention aims to provide the ARIMA model construction method and the evaluation method for the line loss prediction of the low-voltage transformer area, which can save a large amount of manpower and material resources and are beneficial to optimizing the network structure of a power distribution network.

The invention aims to provide a method for constructing an ARIMA model for predicting line loss of a low-voltage transformer area, which comprises the following steps:

s1, preprocessing at least one collected measurement data in the power system or a group of time sequence data in a database;

s2, eliminating periodic influence on the preprocessed time sequence data by adopting a periodic difference operator, carrying out ADF unit root stationarity inspection on the processed data, and eliminating the growth trend by adopting difference operation if the inspection is not passed;

s3, determining the order of the ARIMA model through the Chichi information criterion, and constructing the ARIMA model and an error prediction model thereof;

and S4, self-correcting the ARIMA model according to the network loss data residual error predicted by the prediction model, and outputting the model after the residual error precision is met.

Further, the S1 specifically includes: detecting the grid loss data through related equipment or extracting related time sequence data from a database, and forming time sequence data P according to the time sequence_t；

Eliminating the periodic variation of data;

the periodic influence is eliminated by a periodic difference operator conversion (10);

(1-B^l)P_t

checking the stationarity of the time sequence data;

detection of sequence data P Using ADF Unit root method_tThe smoothness of the structure is improved.

Furthermore, the ADF unit root method is adopted to process the time sequence data P_tEviewsk is used for realizing unit root detection, and if the detection value s is less than 1% confidence interval, the time sequence data P_tStrictly stable; if the assumed test value s is less than 5% confidence interval, the time series data P_tThe stability is achieved; otherwise time sequence data P_tIt is not smooth.

Further, in S2: the elimination of the data growth trend is specifically:

for time series data P_tUsing a formula

Carrying out i +1 order difference operation, i being the number of times of difference operation, obtaining a new time sequence P_tThe smoothing detection is performed in S3.

Further, in S3: the determination of the ARIMA model order specifically comprises the following steps:

parameters for p and q were judged using the akage information criterion, and AIC values were calculated by the following formula:

AIC＝2k-2ln(L)

wherein k is the number of model parameters and L is a likelihood function;

selecting the minimum term of the AIC value as a model order; obtaining an ARIMA model;

substituting p and q into the formula

Obtaining ARIMA model M_E；

The error prediction model is specifically constructed as follows:

using a predictive model M_EPredicting network loss at time t

Obtaining an error sequence E (t) between the true value and the predicted value, and establishing a prediction model error prediction model of the E (t)

Further, in S4: the ARIMA model adaptability test and correction specifically comprise the following steps:

using predictive models

Predicting error at time t

And corrected using the following formula:

calculating residual error

Check whether E is_i(t) is less than or equal to xi, if residual error E_i(t) if the accuracy is satisfied, outputting, otherwise correcting and re-correcting by the following formula

The second purpose of the invention is to provide an ARIMA model construction system for low-voltage transformer area line loss prediction, which comprises:

the preprocessing module is used for preprocessing at least one acquired measurement data in the power system or a group of time sequence data in the database;

the inspection module is used for eliminating periodic influence on the preprocessed time sequence data by adopting a periodic difference operator, carrying out ADF unit root stability inspection on the processed data, and eliminating the growth trend by adopting difference operation if the inspection is not passed;

the building module is used for determining the order of the ARIMA model through the Chichi information criterion and building the ARIMA model and an error prediction model thereof;

and the self-correction module is used for self-correcting the ARIMA model according to the network loss data residual error predicted by the prediction model, and outputting the model after the residual error precision is met.

The third objective of the present invention is to provide an evaluation method based on the ARIMA model construction system for low-voltage transformer area line loss prediction, including:

s101, inputting a sample data set of line loss of a transformer area, determining the number k of clusters by using an elbow rule, and selecting k samples to be respectively used as cluster centers of initial division;

s102, dividing other objects in the sample data set into a set where a nearest central point is located, continuously moving a clustering center of each allocated cluster class by calculating the average value of all data in the cluster class, and dividing and clustering again until the sum of squares of errors in the clusters is minimum and no change exists;

s103, calculating similarity measurement among objects in the contour coefficient evaluation data set, wherein the closer the value is to 1, the more compact the cluster is, and the better the clustering is;

and S104, determining an evaluation standard of the line loss of the low-voltage distribution network by adopting a success degree analysis method and combining a platform area line loss data clustering analysis conclusion.

The fourth purpose of the invention is to provide an information data processing terminal for realizing the ARIMA model construction method for predicting the line loss of the low-voltage transformer area.

A fifth object of the present invention is to provide a computer-readable storage medium, which includes instructions that, when executed on a computer, cause the computer to execute the ARIMA model construction method for low-voltage transformer area line loss prediction described above.

The invention has the advantages and positive effects that:

by adopting the technical scheme, the method and the device construct a rapid and accurate low-voltage transformer area line loss prediction model, can save a large amount of manpower and material resources, are beneficial to optimizing the network structure of the power distribution network, and have important significance for power supply enterprises to formulate reasonable loss reduction measures.

Drawings

FIG. 1 is a flow chart of a preferred embodiment of the present invention;

FIG. 2 is a timing diagram of line loss rate of primary differential station area in the preferred embodiment of the present invention;

FIG. 3 is a Q-Q diagram of an ARIMA model in a preferred embodiment of the present invention;

FIG. 4 is a diagram of the ARIMA model prediction results in a preferred embodiment of the present invention;

FIG. 5 is a flow chart of a clustering algorithm in a preferred embodiment of the present invention;

FIG. 6 is a line loss rate distribution box chart of a certain county area according to a preferred embodiment of the present invention;

FIG. 7 is a graph of the elbow rule for county line loss rate data in accordance with the preferred embodiment of the present invention;

FIG. 8 is a diagram illustrating data distribution in a preferred embodiment of the present invention.

Detailed Description

In order to further understand the contents, features and effects of the present invention, the following embodiments are illustrated and described in detail with reference to the accompanying drawings:

please refer to fig. 1 to 8; in order to solve the defects of the prior art, the corrected power distribution system network loss non-stationary random ARIMA prediction model is obtained in the first stage, the power distribution system network loss prediction in a future period can be realized, and the line loss evaluation system based on the K-Means clustering algorithm is provided in the second stage, so that the network loss degree can be effectively evaluated to take processing measures.

The second aspect of the disclosure provides a line loss evaluation system based on a K-Means clustering algorithm.

A line loss evaluation system based on a K-Means clustering algorithm comprises the following steps:

inputting a sample set of line loss samples of the transformer area, determining the number k of clusters by using an elbow rule, and selecting k samples to be respectively used as cluster centers of initial division;

dividing other objects in the sample data set into a set where the nearest central point is located, continuously moving a clustering center of each allocated cluster class by calculating the average value of all data in the cluster class, and dividing and clustering again until the error sum of squares in the clusters is minimum and no change exists;

calculating similarity measurement among objects in the contour coefficient evaluation data set, wherein the closer the value is to 1, the more compact the cluster is, and the better the cluster is;

and determining the evaluation standard of the line loss of the low-voltage distribution network, namely the evaluation grade and the evaluation score corresponding to the line loss of the five types of distribution areas, by adopting a success degree analysis method and combining the clustering analysis conclusion of the distribution area line loss data.

TABLE 1 time series data Table

The technical scheme of the invention is as follows:

theory of concrete technical method

The time sequence method can accurately describe the random process of equidistant random time sequences, and is a non-stable random process because the network loss change of the power distribution network is related to the power consumption and the scheduling mode. Constructing a general ARIMA model of the power distribution network loss considering a non-stationary random process:

P_t＝f(t)+g(t)+X(t) (1)

wherein, P_tFor time series of loss of active network, f (t) represents P_tA non-periodic growth trend component of; g (t) represents P_tA periodically varying component of; x (t) is P_tA stationary random process component of (a);

the non-stationary random process (1) is separated from the periodic trend and the non-periodic growth trend, and then becomes a stationary random process, and further is modeled and predicted by ARMA.

The first order difference operation is defined as:

the d-order difference operation is defined as:

to the original sequence P_tA new difference sequence can be obtained by carrying out first-order difference calculation

Examination of

And if not, continuing to use the difference operation until a stable sequence is obtained. The first order difference may eliminate linear (first order) growth trends and the second order difference may eliminate second order trends.

Defining the delay operator B:

BP_t＝P_t-1,B^kP_t＝P_t-k (4)

a typical loss ARIMA model can be described as follows:

wherein s is P_tD is an increasing order number,

Φ(B^s)＝1-ΦB^s-Φ₂B^2s-...-Φ_QB^Ps；Θ(B^s)＝1-ΘB^s-Θ₂B^2s-...-Θ_QB^Qs. Here,. phi.,. phi._iAnd Θ_iAre both constant, P and Q are order, A_tIs also an ARIMA model.

And can be converted into a smooth ARMA model:

equation (6) is referred to as the ARIMA model of order (p, d, q). Epsilon_tWhite noise of mean 0An acoustic smoothing sequence.

The combination of formulas (5) and (6) can result in:

equation (7) is a cumulative autoregressive moving average model of the loss of the network, and has an order of (P, D, Q) × (P, D, Q). Through a series of difference and periodic difference operations, a non-stationary network loss random process P_tCan be converted into a stable random process epsilon_t。

Assuming loss time series P_tSatisfying the model (7), a (p ', q') order ARMA sequence

Satisfies the following conditions:

then X_tThe first prediction of (1) is:

due to the fact that

Can obtain the predicted value

1.2 the technological disclosure process

As shown in fig. 1, an embodiment 1 of the present disclosure provides a method for constructing an ARIMA prediction model for self-correcting a power distribution system network loss, including the following steps:

preprocessing at least one measurement data of the power system acquired by the measurement device or a group of time sequence data in a database;

for the preprocessed time sequence data, a periodic difference operator is adopted to eliminate periodic influence, the processed data is subjected to ADF unit root stability inspection, and if the inspection is not passed, a difference operation is adopted to eliminate the growth trend;

determining the order of the ARIMA model through an Akaichi information criterion (AIC criterion), and constructing the ARIMA model and an error prediction model thereof;

and self-correcting the ARIMA model according to the network loss data residual predicted by the prediction model, and outputting the model after the residual precision is met.

The method specifically comprises the following steps:

extraction and construction of time series data

The ARIMA algorithm requires a time series of equal time intervals as input data. Detecting the grid loss data through related equipment or extracting related time sequence data from a database, and forming time sequence data P according to the time sequence_t。

Eliminating data periodicity variations

Periodic effects are removed by periodic difference operator conversion (10)

(1-B^l)P_t (10)

Temporal data stationarity checking

The ARIMA algorithm requires that the input time series data be smooth, i.e., the mean and variance of the data do not change with time. The method adopts ADF unit root method to detect sequence data P_tThe smoothness of the structure is improved.

Applying ADF unit root method to time sequence data P_tEviewsk is used for realizing unit root detection, and if the detection value s is less than 1% confidence interval, the time sequence data P is shown_tStrictly and stably entering the step 5; if the test value s is less than 5% confidence interval, the time-series data P is indicated_tAnd (5) stabilizing, and entering the step 5; otherwise time sequence data P_tIt is not smooth.

Eliminating data growth trends

For time series data P_tAdopting a formula (3) to perform i +1 order difference operation, wherein i is the number of times of the difference operation, and obtaining a new time sequence P_tAnd carrying out stability detection in the step 3.

ARIMA model order determination

Parameters of p and q are judged using the akage pool information criterion (AIC criterion), and AIC values are calculated by equation (12).

AIC＝2k-2ln(L) (12)

Where k is the number of model parameters and L is the likelihood function.

And selecting the minimum term of the AIC value as the order of the model, wherein the estimated probability distribution is the truest and the model is the closest to the true data.

Obtaining an ARIMA model

Substituting p and q into formula (7) to obtain ARIMA model M_E。

Error prediction model construction

Using a predictive model M_EPredicting network loss at time t

Obtaining an error sequence E (t) between the true value and the predicted value, and establishing a prediction model of E (t) -an error prediction model

ARIMA model adaptability test and correction

Using predictive models

Predicting error at time t

And corrected by equation (13)

Calculating residual error

Check whether E is_i(t) is less than or equal to xi, if residual error E_i(t) if the accuracy is satisfied, outputting, otherwise, correcting by the formula (14) and re-executing the step 8.

1.3 example analysis

And (3) taking the grid loss data of the low-voltage power distribution system in a certain park area as an example, and explaining the establishing, checking and predicting processes of the ARIMA model.

(1) Timing data construction

And extracting line loss data of the transformer area in an original database, and arranging the line loss data according to a time sequence to generate time sequence data of a user. The specific data structure is shown in fig. 2:

(2) time series data stationarity verification

The ARIMA algorithm requires that the input time series data be smooth, i.e., the mean and variance of the data do not change with time. The unit root test has the characteristics that the result is relatively complex, but the test result is more accurate.

ADF inspection assumes that the time series has a unit root. The result consists of a hypothesis test value, a p-value and a threshold of three confidence intervals (1%, 5%, 10%), wherein the hypothesis test value can be used for comparison with the threshold of the three confidence intervals, and if the hypothesis test value is less than the threshold of the 1% confidence interval, the original hypothesis can be strictly rejected; assuming that the test value is less than the 5% confidence interval threshold, the original hypothesis may be rejected, and so on. The p-value is used for comparison with a given level of significance a, and if the p-value is less than the given level of significance, it indicates that the original hypothesis can be rejected. In practical statistical tests, the significance level is mostly 0.05.

The square root test was performed on the region data, and the results are shown in table 2 below:

TABLE 2 line loss unit root test results of distribution room

In table 2, the hypothesis test value is-1.3345, which is greater than the threshold of the 10% confidence interval, while the p value is 0.61, which is much greater than the given significance level of 0.05, so the hypothesis test result is that the hypothesis that the time series has the unit root cannot be rejected, and the time series has the unit root, which is not the stationary sequence.

(3) Time series smoothing

Since the data has obvious linear increasing trend and contains fixed period and is not white noise, the data needs to be subjected to difference processing and then stability test until the sequence is stable. The number of the differences is the order of d in the model ARIMA (p, d, q).

The first difference is performed on the line loss rate time sequence curve of the tower area, and the image is shown in fig. 2:

as can be seen from fig. 2, the timing diagram after one difference has no trend of linear increase, and the amplitudes of the curves are similar. And (3) carrying out unit root check on the time sequence after difference:

TABLE 3 first-order differential tower round platform area line loss time sequence data unit root test result

In the above table, the p value is 0.0012, which is less than the set significance level α (0.05), and the hypothesis test value is-4.04, which is less than the threshold value of 1% confidence interval (-4.01), so that the original hypothesis that the time series has a unit root can be strictly rejected, the time series after the first difference has no square root, and the sequence is stable.

(4) ARIMA model order determination

Selecting the first two thirds of the original data (the first 15 data) as a training set, and fitting the model; the latter third (the last 4 data) was used as a test set for evaluation of the model.

After the first order difference, the time series is transformed into a stationary sequence, so in the ARIMA (p, d, q) model, the value of d should be 1. Parameters of p and q are judged using the akage information criterion (AIC criterion). The AIC criterion is an information criterion widely applied in statistical model selection, is mainly used for solving the problem of model selection, and achieves a certain balance between the complexity of a model and the number of parameters. When selecting the optimal model from a set of alternative models using the AIC criterion, the model with the smallest AIC value should be selected. The smaller the AIC value, the closer the estimated probability distribution is to the true distribution, and the closer the model is to the true data.

The partial ARIMA model fitted to the tower area and its AIC values are shown in table 4:

TABLE 4 partial model information and its AIC value

In table 4, the model with the smallest AIC value is given the values p 2, d 1, and q 2, and thus the best ARIMA model is ARIMA (2,1, 2).

(5) Model suitability test

After the identification and parameter determination of the model is completed, the model needs to be diagnosed and checked in order to find out whether the used model is suitable, and if not, the established model should be modified. In the ARIMA model, the residual of the model is assumed to be a random white noise sequence that follows a normal distribution. And if the white noise of the model residual error in the test result conforms to normal distribution, indicating that the model is a non-random white noise sequence.

This step can use a Q-Q map to check the distribution of residuals. In a Q-Q plot, the more the data tends to a normal distribution, the closer the points in the data are to the straight line in the plot. The Q-Q plot of the upper section model residual is shown in FIG. 3:

in fig. 3, the data points are all located near the straight line, so the residuals of the model can be considered to fit the normal distribution, and the ARIMA (2,1,2) model is valid.

(6) Model error checking

Before prediction is carried out by using the ARIMA (2,1,2) model constructed in the previous section, the data of the test set is compared with the result of model prediction to evaluate the error of the model. Namely, the line loss data of the tower park district in 5 months is predicted by using a model for fitting the line loss data of the district in 1 month to 4 months. The comparison of the predicted results of the model with the actual line loss values is shown in table 5:

TABLE 5 comparison of predicted and actual values of line loss rate of tower area

In evaluating the accuracy of the model, a Root Mean Square Error (RMSE) indicator may be employed. The closer the index is to 0, the higher the fit of the model to the real data. The root mean square error value of the data in the table is 0.1519, and the fitting error is about 0.15%, which shows that the model has good effect.

(7) Line loss prediction

The subsequent line loss rates of 29 days at 5 months, 5 days at 6 months and 12 days at 6 months were predicted by using the ARIMA (2,1,2) model, and the results are shown in table 6:

TABLE 6 predicted values of line loss rate of transformer area

The fixed-order ARIMA model may receive a start time and an end time, predict values of intermediate timing data, and may also specify a step size, from which the model will predict a sequence of timing sequences for the specified step size, starting at the last time point of the training data. The comparison of the time charts generated by the ARIMA (2,1,2) model (16 days 1 to 12 days 6) with the station area line loss rate time chart (16 days 1 to 22 days 5) is shown in fig. 5:

in fig. 4, the solid line is an ARIMA (2,1,2) model timing curve, and the dotted line is a tower area line loss rate timing curve. It can be seen that the trends of the two curves are identical, and the numerical error does not exceed 0.5%.

The method specifically comprises the following steps:

2.1K-means Cluster analysis theory

Clustering analysis is one of the main techniques for data mining application, and can be used as a data analysis method alone or as a preprocessing step of other data mining techniques. Clustering is the process of partitioning a physical or abstract collection into classes composed of similar objects, such that objects in the same class have a high degree of similarity, while objects between different classes differ significantly. The idea of the clustering method is to cluster each element to its nearest center (mean) class, so that the object similarity in the same cluster is high, and the object similarity in different clusters is low.

The basic strategy of the K-means clustering algorithm is as follows: knowing a data set containing n sample data and a given clustering number k, firstly randomly selecting k samples as initially-divided cluster centers respectively, then calculating the distance from the non-divided sample data to each cluster center point by adopting an iterative method according to a similarity measurement function, dividing the sample data into the cluster class where the cluster center closest to the sample data is located, and for each distributed cluster class, continuously moving the cluster center by calculating the average value of all data in the cluster class, and dividing the cluster again until the error square sum in the class is minimum and no change exists. The algorithm has the characteristics that whether each sample data is correctly divided into clusters or not is judged in each iteration process, and if the sample data is incorrect, readjustment is carried out. And after all the data are adjusted, modifying the cluster center, and performing the next iterative computation. If each data sample is assigned to the correct family in an iterative process, the cluster center is not adjusted. And (4) the clustering center is stable and does not change any more, the target function is marked to be converged, the algorithm is ended, and finally the clustering result is evaluated.

To facilitate the description of the clustering step of the K-Means algorithm, some formulas and definitions are introduced:

1) data set s ═ x (x) that needs to be clustered₁,x₂,…,x_n) K cluster centers are (C)₁,C₂,…,C_k)；

The most common measure for computing the distance between two sample objects by the K-means algorithm is the euclidean distance, defined as follows:

wherein x is_i＝(x_i1,x_i2,…,x_ip)，x_j＝(x_j1,x_j2,…,x_jp) A data object representing two p-dimensional attributes.

2) The method for calculating the average distance of all sample points is as follows:

n is the total number of sample objects in the dataset, d (x)_i,x_j) Is a sample point x_iAnd x_jThe euclidean distance of (c).

3) The most commonly used objective function is the squared error criterion function, defined as follows:

N_idenotes the ith cluster set, c_iIndicating the center of the ith cluster. E represents the sum of the squares of the euclidean distances of all data sample objects to the cluster center to which they belong.

In summary, the flow chart of the clustering algorithm is shown in fig. 6, that is, the steps of the K-means algorithm are as follows:

inputting: the data set S, assuming n data objects, has K clusters to be partitioned.

And (3) outputting: and (4) according with the clustering result when the target evaluation function is converged.

The first step is as follows: randomly extracting K data objects from the data set S to serve as central points of primary clustering;

the second step is that: calculating the distance from the rest data objects to the clustering center, and then dividing the data objects into the class to which the central point closest to the data objects belongs;

the third step: recalculating the center of each cluster, adjusting the division of all data objects, and comparing whether the division is changed with the division of the last cluster;

the fourth step: and calculating the value of E, if the value of E is converged, finishing the clustering process and outputting a clustering result. Otherwise, returning to the step 2 to continue the iteration until the clustering division is not changed any more or E reaches a convergence condition.

Clustering parameter selection and result evaluation

When the K-means method is used for clustering, a proper clustering number K needs to be selected, so that data objects in the clusters are similar, and the data objects among the clusters are different. The similarity degree of the data in the clusters and the dissimilarity degree of the data among the clusters can be described by using the distortion degree, and the smaller the distortion degree is, the better the clustering effect is.

The present embodiment determines the number of clusters k using the elbow rule. The elbow rule is that a range of k values is selected, then the distortion degrees of clustering results under different k values are drawn into images, and the point where the descending speed of the distortion degrees is changed from fast to gentle is the elbow, namely the optimal clustering number.

In the embodiment, the clustering effect is evaluated by using the contour coefficient, the range of the contour coefficient is (-1, 1), the contour coefficient is-1 when the clustering is incorrect, the contour coefficient of the high-density clustering is 1, and when the contour coefficient is near 0, the overlapping phenomenon exists between the clusters. The contour factor is larger when the clusters are dense and the separation is better.

For a certain point i in the cluster, the contour coefficient calculation formula is

Where av (i) is equal to the average of the distances of the i vector to other points in the cluster to which it belongs, and m (i) is the minimum of the average distances of the i vector to all points in a cluster nearest to it. The "distance" in the above is the dissimilarity, and the larger the value of the "distance" is, the higher the degree of dissimilarity is. Therefore, av (i) also represents the average of the dissimilarity of the i vector to other points in the cluster to which it belongs, and m (i) represents the minimum of the average dissimilarity of the i vector to other clusters.

2.2 example analysis

The algorithm provided for the embodiment can reasonably cluster the line loss data so as to realize the effect of reasonably dividing the line loss grades according to the distribution condition of the line loss rate. This embodiment collects 20-day line loss data of all districts in a certain county. The 20-day data has a condition that one of the power supply and sales of part of the distribution areas or both of the power supply and sales of the distribution areas are null values, so that the line loss rate cannot be calculated and expressed as null values, and meanwhile, the line loss rate is negative, so that the line loss condition cannot be reasonably determined. The details of the line loss data from 1 month to 5 months and 20 days in a certain county are shown in the following table:

TABLE 7 line loss rate data sheet for district of certain county

As can be seen from table 7, 55524 pieces of line loss rate data were recorded for 20 days in a certain county, and the number of records per day varied from 2700 to 2800. 124 lines in which the line loss rate cannot be calculated as a null value; the line loss rate is negative, and 2539 pieces of data in the line loss state cannot be obtained from the line loss rate, and after the two pieces of data are removed from the whole, 52861 pieces of data remain, and the distribution diagram of the line loss rate of the whole is shown in fig. 7.

As shown in fig. 6, the data contained in the bottom rectangular area is data distributed between the lower quartile (3.59%) and the upper quartile (7.2%), and the green line is the line loss rate average (6.32%). The black circle represents an outlier in the data distribution, and in the data distribution of the line loss rate in a certain county, only an outlier larger than the distribution main body exists. In the figure, the criteria for determining outliers are: upper quartile +1.5 x (upper quartile-lower quartile); i.e., the cells with line loss rate greater than 12.615%, are classified as outliers. The number of outliers was 3493, which accounted for 6.61% of the population.

The present embodiment determines the number of clusters k using the elbow rule. An elbow rule image for a county line loss rate dataset is shown in fig. 8. In this figure, the degree of distortion decreases as the number of clusters increases. When the cluster number is 5, the curve starts to turn gently, and therefore, it can be determined that the optimum cluster number k is 5.

In this embodiment, a K-means algorithm with a parameter K of 5 is used to cluster line loss rate data sets of a certain county area, and the clustering effect is evaluated by using a contour coefficient. The contour coefficient of the current clustering result is shown in fig. 8, and the contour coefficient of the clustering result is 0.547, which indicates that the clustering generates obvious cluster division, but some clusters may be loose, and the clustering result is available.

The category intervals of five line loss rates divided by clustering and the number of line loss data contained in each category are shown in the following table:

TABLE 8 clustering result table for line loss rate of district in certain county

As can be seen from table 8, the numbers of pieces of line loss data included in the first category and the second category having smaller line loss rate values are 26588 and 20610, respectively, and the number of pieces of line loss data included in the category having a higher line loss rate section is smaller. In summary, the reason why the line loss rate distribution status of a certain county area is described above is considered to be that 93.39% of data are densely distributed in the interval where the line loss rate is lower than 12.615% and the other 6.61% of data are distributed in the interval where the line loss rate is higher than 12.615 in the data distribution of the original data set, as shown in fig. 8. The squares represent the line loss rate interval: the line loss rate interval of the small square is 0.01-12.615%, and the line loss rate interval of the large square is 12.615-98.58%; the black dots represent line loss data. It can be seen that the distance between the line loss data in the interval 0.01% to 12.615% is greater than the distance between the line loss data in the interval 12.615% to 98.58%. And the K-means carries out clustering according to the Euclidean distance between the data, so that the difference of the number of the line loss data in different classes in a clustering result can be explained.

2.3 line loss success rating evaluation Standard

According to the low-voltage distribution network line loss clustering analysis results, the low-voltage distribution network line loss can be divided into five types. In this embodiment, a success analysis method is adopted to determine an evaluation criterion of the line loss of the low-voltage distribution network, that is, an evaluation grade and an evaluation score corresponding to the line loss of the five types of distribution areas.

The success degree analysis method is a comprehensive analysis method for project evaluation, and is characterized in that a qualitative conclusion is given to the success degree of a project (namely the success or failure degree of the project to achieve an expected target), and the project is divided into five grades according to the qualitative conclusion, wherein the grade standard is as follows: the method is completely successful, and shows that all the targets of the project are completely realized or exceeded, and compared with the cost, the project obtains huge benefits and influences; success, indicating that most of the goals of the project have been achieved, and achieving the expected benefits and impacts of the project relative to cost; thirdly, partial success shows that the project achieves the original partial target, and only obtains certain benefits and influences relative to the cost; unsuccessful, indicating that the project achieves very limited goals, with little benefit and impact on the project with respect to cost. Fail, indicating that the goal of the project cannot be achieved and the project has to be terminated.

Therefore, according to the grade standard of the success degree analysis method and the clustering analysis conclusion of the line loss data of the transformer area, the line loss standard of the transformer area is divided into the following five grades:

1) the line loss is very low. The line loss of the transformer area is at a very low level, and the operation economy of the transformer area is good.

2) The line loss is low. The line loss of the transformer area is at a low level, and the running economy of the transformer area is good.

3) The line loss is high. The line loss of the transformer area is in a higher level, and the operation economy of the transformer area is poorer.

4) The line loss is high. The line loss of the transformer area is at a high level, and the economical efficiency of the transformer area operation is poor.

5) The line loss is high. The line loss of the transformer area is at a high level, and the operation economy of the transformer area is poor.

Finally, based on the above-mentioned stage area line loss grade standard and stage area category, a stage area line loss percentage evaluation score standard and a stage area classification standard are given, as shown in the following table:

TABLE 9 evaluation standard for line loss of low-voltage distribution network

According to the above table, the evaluation standard of the line loss of the low-voltage distribution network can be known, and the evaluation grade, the evaluation score and the category of the line loss of the transformer area can be correspondingly determined according to the current numerical value and the future predicted value of the line loss of the transformer area.

Line loss success level evaluation example

The low-voltage distribution network line loss evaluation method can be applied to the current station area line loss state level evaluation and the station area line loss future prediction level evaluation, and the line loss evaluation process and results are given by taking part of the station areas as examples.

1) Line loss current situation level evaluation of transformer area

Taking 223 districts in 24 days in 4 months and 24 days in 2019 of a certain county as analysis samples, obtaining the daily line loss value of each district, and determining the evaluation grade and category of each district according to the evaluation standard, wherein the determination result is as follows.

TABLE 10 evaluation results of line loss status of some district in a certain county

2) Future prediction level evaluation of line loss of transformer area

Taking a tower park district in a certain county as an example, combining the determined future line loss prediction value of the tower park district, determining the evaluation grade and category of the future line loss prediction level of the tower park district according to the evaluation standard, and determining the result as follows.

TABLE 11 predicted value of line loss rate of tower area

According to the results in the table 11, the predicted line loss values of the tower park district on three dates are within the range of 5.19% -10.08%, the district belongs to a qualified district, and the evaluation score of the line loss is 80-90.

An ARIMA model construction system for low-voltage transformer area line loss prediction is characterized by comprising:

An information data processing terminal for realizing the ARIMA model construction method for predicting the line loss of the low-voltage transformer area.

A computer-readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the method for constructing a low-voltage transformer area line loss prediction ARIMA model as described above.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.

The above description is only for the preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and all simple modifications, equivalent changes and modifications made to the above embodiment according to the technical spirit of the present invention are within the scope of the technical solution of the present invention.

Claims

1. An ARIMA model construction method for low-voltage transformer area line loss prediction is characterized by comprising the following steps:

2. The method for constructing an ARIMA model for low-voltage transformer area line loss prediction according to claim 1, wherein the S1 specifically is: detecting the grid loss data through related equipment or extracting related time sequence data from a database, and forming time sequence data P according to the time sequence_t；

Eliminating the periodic variation of data;

(1-B^l)P_t

checking the stationarity of the time sequence data;

3. The ARIMA model construction method for low-voltage transformer area line loss prediction as claimed in claim 2, wherein the ADF unit root method is adopted to perform the calculation on the time series data P_tEviewsk is used for realizing unit root detection, and if the detection value s is less than 1% confidence interval, the time sequence data P_tStrictly stable; if the test value s is assumed to be less than 5% confidence interval, thenTime series data P_tThe stability is achieved; otherwise time sequence data P_tIt is not smooth.

4. The method for constructing an ARIMA model for low-voltage transformer area line loss prediction according to claim 2, wherein in S2: the elimination of the data growth trend is specifically:

for time series data P_tUsing a formula

5. The method for constructing an ARIMA model for low-voltage transformer area line loss prediction according to claim 2, wherein in S3: the determination of the ARIMA model order specifically comprises the following steps:

AIC＝2k-2ln(L)

wherein k is the number of model parameters and L is a likelihood function;

substituting p and q into the formula

Obtaining ARIMA model M_E；

The error prediction model is specifically constructed as follows:

using a predictive model M_EPredicting network loss at time t

6. The method for constructing an ARIMA model for low-voltage transformer area line loss prediction according to claim 5, wherein in S4: the ARIMA model adaptability test and correction specifically comprise the following steps:

using predictive models

Predicting error at time t

And corrected using the following formula:

calculating residual error

7. An ARIMA model construction system for low-voltage transformer area line loss prediction is characterized by comprising:

8. An evaluation method of the ARIMA model construction system for low-voltage transformer area line loss prediction according to claim 7, comprising:

9. An information data processing terminal for implementing the ARIMA model construction method for low-voltage transformer area line loss prediction of any one of claims 1 to 6.

10. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method of constructing an ARIMA model for low-voltage station area line loss prediction according to any of claims 1 to 6.