A method for detecting power theft in high-loss lines based on vector autoregression model
By using a vector autoregression model, the dynamic interaction between line loss and electricity consumption is utilized to identify suspected electricity thieves, solving the problem of difficulty in identifying electricity thieves on high-loss lines and achieving efficient electricity theft detection and investigation.
Patent Information
- Application Number
- CN202110117959.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-01-28
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2041-01-28
AI Technical Summary
Existing technologies are insufficient for efficiently identifying electricity theft users, especially on high-loss lines. Traditional methods lack effective data mining and analysis tools, resulting in low inspection efficiency and economic losses.
Using a vector autoregression model-based approach, suspected electricity theft users are identified through boundary cointegration tests and vector autoregression model analysis. By utilizing the dynamic interaction between line loss and electricity consumption, the impact of each user on line loss is quantified, and users with significant fluctuation contributions are identified.
It improved the accuracy of identifying electricity theft users and the efficiency of investigation, reduced the economic losses of power supply companies, and reduced unnecessary investigation costs.
Smart Images

Figure CN112785456B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of power grid line loss analysis, and relates to a method for establishing a vector autoregression model based on line line loss anomaly to analyze, detect and locate abnormal electricity users, so as to locate and identify suspected electricity theft users (abnormal electricity users). Background Art
[0002] Under the market economy, some unscrupulous operators use various means to steal electricity, directly causing loss of revenue for power supply companies. Traditionally, power consumption anomaly detection mainly relies on manual investigation, which often consumes a lot of manpower and material resources but has little effect due to lack of data and lack of directionality in anomaly detection. At present, my country's power grid companies have basically achieved complete collection of power consumption information, and can timely and accurately grasp the power consumption data and customer information of power users, providing effective technical support for identifying power theft users through power consumption data mining and analysis.
[0003] Based on their experience, production technicians of power supply companies have summarized indicators with definite physical meanings, such as zero-sequence current, power reverse and meter voltage loss of low-voltage users, which can accurately identify abnormal electricity usage behavior. However, such methods are only applicable to specific types of electricity theft and cannot detect other forms of electricity theft, such as electricity circumventing the meter.
[0004] Electric power workers have conducted a lot of research on data-driven power consumption anomaly detection from two aspects: unsupervised cluster analysis and supervised classification analysis. These studies generally design feature indicators based on the common characteristics of power theft by users, such as a downward trend in power consumption, abnormal daily load curves, and low reported capacity utilization, and then specifically select classification or clustering algorithms to identify power consumption anomalies. These studies often use abnormal changes in power consumption as the core element to design feature indicators, which is prone to false alarms for the following reasons: There are significant differences in the power consumption behavior characteristics of users in different industries. The power demand of a considerable number of industries directly depends on the order demand, and large or trending fluctuations in user power consumption are normal; users in industries with relatively stable power demand may experience low power anomalies due to external interference such as environmental protection inspections and safety inspections.
[0005] In the line loss management of power supply enterprises, the line loss management of distribution lines has always been the focus and difficulty of line loss management. Surveys show that in the distribution link, line losses of 10kV and below account for 65% to 70% of the total line loss. Since power theft is an important reason for the high line loss rate, in actual work, marketing personnel often choose the distribution lines with the highest line loss rate for power inspection. For example, lines with a line loss rate greater than 5% are generally considered to have power theft. Although the integrated line loss management system records detailed power consumption data of all substations under the distribution line, due to the lack of effective data mining and analysis methods, marketing personnel can only check power theft users one by one based on experience after selecting high-loss lines. It is urgent to study the applicable high-loss line power theft user positioning and identification algorithm to improve the efficiency of power inspection work. Summary of the invention
[0006] Although there are many ways to steal electricity, most electricity thieves often use the voltage division method or the current division method to achieve equal-proportion electricity theft in order to avoid inspection. The change of the line loss of the divided line is basically proportional to the electricity consumption of the electricity thief, and the load curve of the electricity thief will significantly show the synchronous change with the line loss. It can be seen that there is a strong correlation between the line loss and the load of the electricity thief. Using this correlation characteristic, the long-term dynamic interaction relationship between the line loss and the load of the electricity thief can be mined and analyzed, and the electricity thief who causes abnormal fluctuations in line loss can be identified.
[0007] Therefore, the purpose of the present invention is to provide a high-loss line electricity theft detection method based on a vector autoregression model to locate and identify suspected electricity theft users in view of the above-mentioned deficiencies in the prior art.
[0008] To achieve the above purpose, the technical solution adopted by the present invention is: a high-loss line power theft detection method based on a vector autoregression model, the method steps are as follows:
[0009] Step 1: Determine the high-loss line for electricity theft, obtain the unit time power loss data of the high-loss line for electricity theft and the unit time power consumption data of each user under the high-loss line in the same time period, and establish the unit time power loss time series S = {S1, S2, ..., S t} and the unit time electricity consumption time series Y of each subordinate user i = {Y i1 , Y i2 , …Y it}, where i is a subordinate user, i=1, 2, ..., n, n is the number of all subordinate users, and t is the length of the time series;
[0010] The above-mentioned determination of high-loss lines for power theft is based on the line loss rate or line loss fluctuation rate of each line, which is a conventional technology in this field. When determining by line loss rate, it is determined by the power supply enterprise based on the actual local situation. Since the theoretical line loss of 10kV distribution lines is mostly between 2% and 3%, it is generally believed that lines with line losses higher than 5% are high-loss lines.
[0011] The above-mentioned time series S of power loss per unit time and time series Y of power consumption per unit time i The same unit time is used. The unit time is a specific time period. When the unit time is a day (i.e. 24 hours), the power consumption data of each user obtained directly includes the daily power consumption data, and the daily power consumption time series of each subordinate user can be directly established; when the unit time is 15 minutes, 30 minutes, 60 minutes or 12 hours and other time periods, the time period power consumption data can be calculated based on the power consumption data of each user obtained, and then the hourly power consumption time series of each subordinate user can be established.
[0012] Step 2: Construct the time series Y1, Y2, Y3...Y of the unit time power consumption of each user under the high-loss line respectively n The unconstrained error correction model of the unit time loss power time series S is used to test whether there is a long-term cointegration relationship between the loss power of the high-loss line and the power consumption of each subordinate user through the marginal cointegration test method. If there is a long-term cointegration relationship, go to step 3;
[0013] The above-mentioned process of establishing the unconstrained error correction model is a conventional technique in the art.
[0014] The marginal cointegration test method mentioned above is a conventional technique in this field. The constraint condition of the commonly used cointegration test methods (Engle-Granger cointegration test and Johansen cointegration test) is that the time series of the two variables have the same integral order. When the line loss power and the user's power consumption are of different orders, it will not be applicable. Since there may be dozens of users under a line or substation, it is a common phenomenon that the integral orders of each user and the line loss power are different. The marginal cointegration test can test the cointegration relationship between multiple variables at the same time, and compared with the commonly used cointegration test method, it does not require the variables to be of the same order, breaking through the limitation of the integral order. Therefore, the present application uses the marginal cointegration test method to test the relationship between the series, which can more stably identify the cointegration relationship between the loss power and the power consumption of the subordinate users.
[0015] Step 3: Construct the time series Y1, Y2, Y3...Y of the unit time power consumption of each user under the high-loss line nThe vector autoregression model of the unit time loss power time series S (to facilitate the analysis of the dynamic interactive relationship between the power loss of high-loss lines and the power consumption of subordinate users in step 4, namely, impulse response function analysis and dynamic variance decomposition) is used to determine the optimal lag period of the model to make the vector autoregression model stable;
[0016] The optimal lag period of the above model is determined based on the AIC, SC and HQ information criteria in the vector autoregression model, which is a conventional technique in the art.
[0017] The stability test process of the above-mentioned vector autoregression model is a conventional technique in the art.
[0018] Step 4. Based on the established vector autoregression model, the impulse response function is used to analyze the users who have significant long-term and short-term impact on the power loss, and then the dynamic variance decomposition of the power loss and the subordinate users is performed to calculate the contribution of each user to the change in power loss. The users who have a significant impact on the power loss of high-loss lines and have the largest contribution to fluctuations are identified as suspected users of power theft, and on-site inspections are carried out.
[0019] The impulse response function analysis, dynamic variance decomposition and volatility contribution calculation process in the above-mentioned vector autoregression model are all conventional techniques in the art.
[0020] The long-term significant impact of each user on the amount of electricity lost is determined based on the relationship between the t-test statistic of each user variable in the long-term cointegration equation and the 5% significance level: if the t-test statistic is greater than 5%, it indicates that the user has no long-term significant impact on the amount of electricity lost; if the t-test statistic is less than 5%, it indicates that the user has a long-term significant impact on the amount of electricity lost. The positive or negative value of the estimated coefficient of each user variable also represents whether the long-term significant impact is promoting or weakening.
[0021] The short-term significant impact of each user on the amount of electricity lost is to see whether the lagged variables of each user in the unconstrained error correction model equation have a lagged effect on the loss variable, that is, a short-term significant impact. The judgment is made based on the relationship between the t-test statistic of each user's lagged variable in the unconstrained error correction model equation and the 5% significance level: if the t-test statistic of the user's lagged variable is greater than 5%, it indicates that the user has no short-term significant impact on the amount of electricity lost; if the t-test statistic of the user's lagged variable is less than 5%, it indicates that the user has a lagged effect on the amount of electricity lost, that is, a short-term significant impact. The positive or negative value of the coefficient estimate of each user's lagged variable also represents the positive or negative of the lag effect.
[0022] The long-term cointegration equation and the unconstrained error correction model equation are both derived from the marginal cointegration test.
[0023] When abnormal fluctuations occur in the line loss of distribution lines, the present invention quantitatively analyzes the influence of each user of the line loss line on the line loss electricity through marginal cointegration test and impulse response function analysis and dynamic variance decomposition in the vector autoregression model. Compared with traditional association analysis and Granger attribution analysis, it can give the dynamic influence path and fluctuation contribution of each user of the line loss line on the line loss electricity, quantify the influence of each user on the line loss electricity, and there is no limit on stability. Compared with the marketing department's door-to-door inspection one by one, the inspection cost is reduced, and the unnecessary economic losses of the power supply company are reduced. This method can detect electricity theft users at the main station for electricity consumption information collection, and can also detect electricity theft in the form of edge computing on the intelligent gateway or collection terminal in the substation. When implemented in the substation in the form of edge computing, it can avoid the problem of negative line loss affecting the accuracy of electricity theft detection when the electricity consumption data of individual users fails to upload. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 The figure is a flow chart of the method of the present invention.
[0025] Figure 2 It is a unit root test diagram of the vector autoregression model of the embodiment of the present invention.
[0026] Figure 3 It is a pulse response curve diagram of line loss power and power consumption of each dedicated transformer user according to an embodiment of the present invention.
[0027] In the figure, the horizontal axis represents the impact duration, and the vertical axis represents the positive and negative reflections. DETAILED DESCRIPTION
[0028] See also Figure 1 The present invention is a method for detecting electricity theft in high-loss lines based on a vector autoregression model. The specific steps of the method are as follows:
[0029] Step 1: Determine the high-loss line for electricity theft, obtain the unit time power loss data of the high-loss line for electricity theft and the unit time power consumption data of each user under the high-loss line in the same time period, and establish the unit time power loss time series S = {S1, S2, ..., S t} and the unit time electricity consumption time series Y of each subordinate user i = {Y i1 , Y i2 , …Y it}, where i is a subordinate user, i=1, 2, ..., n, n is the number of all subordinate users, and t is the length of the time series;
[0030] The unit time mentioned above may be 15 minutes, 30 minutes, 60 minutes, 12 hours or 24 hours, etc.
[0031] Step 2: Construct the time series Y1, Y2, Y3...Y of the unit time power consumption of each user under the high-loss line respectively n The unconstrained error correction model of the unit time loss power time series S is used to test whether there is a long-term cointegration relationship between the loss power of the high-loss line and the power consumption of each subordinate user through the marginal cointegration test method. If there is a long-term cointegration relationship, go to step 3;
[0032] The unconstrained error correction model mentioned above is expressed as follows:
[0033]
[0034] Among them, ΔS t and ΔY t are the first-order differences of the branch line loss and the user's power, i0, i1, i2, ..., i m Respectively represent the line loss power and the lag period of m users, c, γ, α k and β k represents the parameter to be estimated, μ t is the residual term.
[0035] The marginal cointegration test is performed by testing the lagged variable S in the unconstrained error correction model. t , Y t The joint Wald test of the coefficients is used to achieve this, and the output result is the F statistic. If the value of the F statistic is greater than the upper critical value, there is no need to perform a variable stationarity test to determine that there is a cointegration relationship between the variables; if the value of the F statistic is less than the lower critical value, the null hypothesis is directly determined to be true, that is, there is no cointegration relationship between the variables; if the value of the F statistic is between the upper and lower critical values, the conclusion cannot be drawn directly, and a unit root test of the variable is required.
[0036] Step 3: Construct the time series Y1, Y2, Y3...Y of the unit time power consumption of each user under the high-loss line n The vector autoregression model with the unit time loss power time series S is used to analyze the dynamic interaction between the variables (the power loss of high-loss lines and the power consumption of subordinate users). The optimal lag period of the model is determined based on the AIC, SC and HQ information criteria; after determining the optimal lag period of the model, the stability of the vector autoregression model is tested.
[0037] The construction of the vector autoregression model mentioned above is an existing conventional technology. Specifically, by constructing a vector autoregression model to analyze the power consumption Y of n users i The dynamic interactive relationship between (i=1, 2, ..., n) and the power loss S, the vector autoregression (p) model containing n+1 variables can be expressed as:
[0038] Zt =c+A1Z t-1 +A2Z t-2 ...+A p Z t-p +ε t
[0039] Where t = 1, 2...T, Z t is a k-dimensional endogenous variable vector consisting of n user power and 1 line loss power, p is the lag order, T is the number of samples, A1,...,A p is an n×n dimensional coefficient matrix, ε t is a k-dimensional random perturbation vector.
[0040] The selection of the optimal lag period of the model and the test of model stability are also conventional techniques in this field.
[0041] 1) Selection of the optimal lag period of the model
[0042] Information criteria AIC and SC are used to measure the quality of regression models. The smaller the value, the more accurate and concise the constructed model is. AIC and SC are defined as:
[0043]
[0044] in is the residual sum of squares of the model, k is the number of parameters in the model, and n is the sample size.
[0045] 2) Model stability test
[0046] The condition for the vector autoregression (p) model to be stable is that all roots of the following characteristic equation fall within the unit circle. Then the vector autoregression model is stable, and its overall fit is good and its stability is high.
[0047] |I n λ p -A1λ p-1 -A2λ p-2 -LA p |=0
[0048] Among them I n is the n×n dimensional identity matrix, A1,…,A p is an n×n dimensional coefficient matrix, and λ is the root of the characteristic equation.
[0049] Step 4. Based on the established vector autoregression model, the impulse response function is used to analyze the users who have significant long-term and short-term impact on the power loss, and then the dynamic variance decomposition of the power loss and the subordinate users is performed to calculate the contribution of each user to the change in power loss. The users who have a significant impact on the power loss of high-loss lines and have the largest contribution to fluctuations are identified as suspected power theft users.
[0050] The impulse response function is derived by the following process:
[0051] Any vector autoregression model can be expressed as an infinite order VMA(∞) process:
[0052] Z t+s =ε t+s +C1ε t+s-1 +C2ε t+s-2 +L
[0053] C s =δZ t+s / δε t t=1,2,LT
[0054] C s The element in the ith row and jth column of is equal to the impact on the value of the ith variable in period t+s when the disturbance term of the jth user variable in period t increases by one unit shock while the disturbances in other periods are constant.
[0055] Put C s The element in the i-th row and j-th column is considered as a function of the lag period s:
[0056] δZ i,t+s / δε jt ,s=1,2,3,L
[0057] The above formula is the impulse response function, which describes the Z in period t when other variables and early variables remain unchanged. i,t+s Z jt The impulse response function value indicates the magnitude of the long-term effect of the user's power consumption on the power loss, and the positive or negative value of the function indicates the positive or negative effect of the user's power consumption on the power loss. If the impulse response function value of a user is negative in the short term and a large positive value in the long term, it means that the user has a lag effect on the power loss in the short term, but has a positive and significant effect on the power loss in the long term.
[0058] The dynamic variance decomposition and volatility contribution calculation process are as follows:
[0059] Variance decomposition is to further evaluate the importance of different structural shocks by analyzing the contribution of each structural shock to the change of endogenous variables. i The expression of the corresponding variance in the first s period is as follows:
[0060]
[0061] Where: jj =E(ε t ε t ')
[0062] Z i The variance of can be decomposed into k unrelated effects. To determine the relative effect of each disturbance term on Z i The contribution of the variance is defined as follows:
[0063]
[0064] The relative variance contribution (RVC) is the variance contribution of the jth variable to Z based on the shock. i The relative contribution of the variance of the jth variable to the ith variable is used to analyze the impact of the jth variable on the ith variable. j→i When (s) is large, it means that the jth variable has a greater impact on the ith variable, that is, the contribution is large; on the contrary, if RVC j→i When (s) is small, it can be considered that the j-th variable has little influence on the ith variable, that is, its contribution is small.
[0065] Example 1
[0066] Taking the line loss data of a high-loss line and the power consumption data of its subordinate dedicated transformer users as an example, the daily line loss data of the line and the daily power consumption of its six subordinate dedicated transformer users from April 1 to July 9 were collected, and the sequence S and the daily power consumption sequences Y1, Y2, Y3…Y6 of the six subordinate dedicated transformer users were established respectively.
[0067] 1) Marginal cointegration test
[0068] After the marginal cointegration test of the line loss electricity and the electricity consumption series of the subordinate special transformer users, the marginal test F statistic of the corresponding estimated equation is obtained. At this time, the marginal test F statistics of the three cases without time trend term, with constrained time trend term, and with unconstrained time trend term are: F1 = 7.0819, F2 = 10.053, F3 = 11.455, all of which reject the null hypothesis that there is no cointegration relationship between the variables at the 5% significance level, indicating that there is a long-term cointegration relationship between the line loss electricity and the electricity consumption of the six subordinate special transformer users. According to the AIC and SC information criteria and the serial correlation LM test statistic, the estimated equation is simplified to eliminate the insignificant order, and the final model is set as ARDL (1, 0, 0, 1, 0, 0, 4). Based on this, the long-term cointegration relationship equation and the corresponding error correction model are shown in Equation 1 and Equation 2.
[0069] S=0.0301Y1+9.6807Y2-0.8857Y3+23.1207Y4+1.8884Y5-6.4857Y6(1)
[0070] The error correction model of the power loss of the distribution line and the six subordinate dedicated transformer users is:
[0071] ΔS=-2961.5383+10.0099T-0.4326ECM(-1)-0.6178S(-1)+0.0186Y1
[0072] +5.9817Y2-0.5472Y3(-1)+14.2863Y4+1.1669Y5-4.0075Y6(-1)
[0073] +0.1088ΔY3+3.6154ΔY6(-1)+6.0528ΔY6(-2)+3.6032ΔY6(-3)(2)
[0074] Where ECM(-1) = S-0.0301Y1(-1)-9.6807Y2(-1)+0.8857Y3(-1)
[0075] -23.1207Y4(-1)-1.8884Y5(-1)+6.4857Y6(-1)
[0076] From the estimation results of the long-term cointegration equation, the t-test statistics of each variable show that the coefficient estimates of Y5 and Y6 are very significant, and have a significant impact on line loss electricity; the coefficient estimates of Y1, Y2, Y3, and Y4 are not significant, and have no significant impact on line loss electricity. The direction of the coefficient estimates of each variable shows that Y5 has a significant promoting effect on line loss electricity, and Y6 has a significant weakening effect on line loss electricity.
[0077] From the estimation results of the error correction model, except for the Y1, Y2, Y4, and Y5 sequences, the other sequences all have significant hysteresis effects on line loss. Among them, Y3 has a weak negative hysteresis effect; Y6 has both positive and negative hysteresis effects, and the effect is relatively large.
[0078] 2) Construction and evaluation of vector autoregression model
[0079] Five evaluation statistical criteria are obtained by setting different lag periods. The test value table is shown in Table 1, where the shaded part indicates significance at the 5% significance level.
[0080] Table 1 Test of the hysteresis value of the model
[0081] Lag LogL LR FPE AIC SC HQ 1 -3496.579 NA 9.47e+25 79.675 81.046* 80.228* 2 -3432.570 107.881 6.87e+25 79.338 82.078 80.443 3 -3370.824 94.354* 5.40e+25* 79.052* 83.163* 80.709* 4 -3339.314 43.193 8.76e+25 79.445 84.926 81.654
[0082] From the test results in Table 1, we can see that when the lag period p is 3, all statistics are significant at the 5% significance level, that is, the five evaluation statistical criteria LR, FPE, AIC, SC, and HQ all choose the optimal lag period p of 3. Therefore, it is reasonable to choose to establish a vector autoregression model with a lag period of 3. The model constructed in this example has 7 variables, and the optimal lag order is 3, so there should be 21 roots of the characteristic equation located within the unit circle. Figure 2 It can be seen that the reciprocals of all roots are within the unit circle, indicating that the vector autoregression model with a lag period of 3 has a good overall fit and high stability.
[0083] 3) Impulse response analysis and variance decomposition
[0084] The impulse response function is used to analyze Y3, Y5, and Y6, which have significant long-term and short-term impacts on line loss. The analysis results are as follows Figure 3 As shown, the three curves of different colors represent the pulse function change curves of the three dedicated transformer users, that is, they respectively represent the response of the line loss electricity after the positive unit impact of the electricity consumption of the three dedicated transformer users occurs.
[0085] Dynamic variance decomposition is performed on line loss electricity and subordinate dedicated transformer users. The results are shown in Table 2. In the first few periods, the variance of line loss electricity prediction is mainly explained by its own changes, which is consistent with the characteristics of strong volatility and unpredictability. In the long run, about 77% of the variance changes in line loss electricity can be explained by the six subordinate dedicated transformer users. Specifically, among the six factors that can be explained, Y5 has the highest degree of explanation, about 30%; followed by Y3 and Y6, the two have roughly the same degree of explanation, about 17% and 14% respectively; then Y2, about 8%; and finally Y1 and Y4, both only about 4%. In summary, among the six factors that drive the change in line loss electricity, Y5 has the greatest impact, followed by Y3 and Y6, and the impact of other factors is extremely weak.
[0086] From the above analysis, it can be seen that in the early stage of the power transmission process, the technical loss of the line plays a leading role. With the influence of the subordinate special transformer users, the variance contribution of the line loss electricity mainly comes from Y3, Y5, and Y6. Among them, the impact of Y5 on the line loss electricity is significant. Combined with the above-mentioned impulse response analysis, the Y5 user is determined to be the user with the greatest suspicion of electricity theft under this line. After on-site inspection, it is confirmed that the Y5 user has committed illegal electricity theft.
[0087] Table 2 Variance decomposition table of line loss and power consumption
[0088]
Claims
1. A method for detecting electricity theft in high-loss lines based on a vector autoregression model, characterized in that: The steps of this method are as follows: Step 1. Determine the high-loss line for electricity theft, obtain the unit time power loss data of the high-loss line for electricity theft and the unit time power consumption data of each user under the high-loss line in the same time period, and respectively establish a unit time power loss time series S = {S1, S2, ..., St} and a unit time power consumption time series Yi = {Yi1, Yi2, Yit} of each user under the line, where i is the user under the line, i = 1, 2, ..., n, n is the number of all users under the line, and t is the length of the time series; Step 2. Construct the unconstrained error correction model of the unit time power consumption time series Y1, Y2, Y3...Yn of each user under the high-loss line and the unit time loss power time series S respectively, and use the marginal cointegration test method to test whether there is a long-term cointegration relationship between the loss power of the high-loss line and the power consumption of each user under it. If there is a long-term cointegration relationship, proceed to step 3; Step 3. Construct a vector autoregression model of the unit time power consumption time series Y1, Y2, Y3...Yn of each user under the high-loss line and the unit time power loss time series S, and determine the optimal lag period of the model to make the vector autoregression model stable; Step 4. Based on the established vector autoregression model, the impulse response function is used to analyze the users who have a significant long-term and short-term impact on the power loss, and then the dynamic variance decomposition of the power loss and the subordinate users is performed to calculate the contribution of each user to the change of power loss. The users who have a significant impact on the power loss of high-loss lines and have the largest contribution to fluctuations are identified as suspected power theft users; in: The analysis of users with long-term significant impact on the lost electricity in step 4 is judged based on the relationship between the t-test statistic of each user variable in the long-term cointegration equation and the 5% significance level: if the t-test statistic is greater than 5%, it indicates that the user has no long-term significant impact on the lost electricity; if the t-test statistic is less than 5%, it indicates that the user has a long-term significant impact on the lost electricity; The analysis of users with short-term significant impact on the lost electricity in step 4 is judged based on the relationship between the t-test statistic of the lagged variables of each user in the unconstrained error correction model and the 5% significance level: if the t-test statistic of the lagged variable of the user is greater than 5%, it indicates that the user has no short-term significant impact on the lost electricity; if the t-test statistic of the lagged variable of the user is less than 5%, it indicates that the user has a lag effect on the lost electricity, that is, a short-term significant impact.
2. According to the method for detecting electricity theft in high-loss lines based on a vector autoregression model in claim 1, it is characterized in that: The determination of the high-loss line for power theft in step 1 is based on the line loss rate or line loss fluctuation rate of the line.
3. The high-loss line electricity theft detection method based on vector autoregression model according to claim 1 is characterized in that: The optimal number of lags for the model in step 3 is determined based on AIC, SC and HQ information criteria.
