Electric energy meter measurement missing data complementation method based on univariate dynamic time warping

Abstract

The invention relates to the technical field of electric power, and discloses an electric energy meter measurement missing data complementing method based on univariate dynamic time warping, which comprises the following steps: constructing an initial time sequence INP from normal no-missing electric energy meter data, and determining missing data of a missing sequence; according to the length T of the missing data, combining with the new time sequence INP'to adaptively create a query subsequence, and based on the initial time sequence INP, constructing a reference database; calculating the similarity between the query sub-sequence and each candidate sub-sequence in the reference database by adopting a dynamic time warping algorithm so as to obtain a candidate sub-sequence most similar to the query sub-sequence; and filling missing data with the last T adjacent data of the most similar candidate subsequences. According to the method, the complex dynamic change of the data can be better captured, and a reliable result can still be obtained under the condition that nonlinear fluctuation is possibly caused by factors such as faults and interference in the measurement of the electric energy meter.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping. Background Art

[0002] As the most widely used and basic energy metering device, the completeness and availability of the measured data of the energy meter is of vital importance. However, in actual work, due to the clock error, memory failure, human damage and other reasons of the energy meter, there is often a problem of missing measured data of the energy meter. To address this problem, the existing technology usually adopts data interpolation and fitting to complete the data. In the field of data interpolation, the most commonly used methods are Newton interpolation, multivariate function interpolation, and spline interpolation. In the field of data fitting, the most commonly used methods are polynomial fitting and its improved methods. Although the above methods can currently meet the needs of traditional power marketing metering business, under the situation of new power systems, the frequency of energy meter measurements is getting higher and higher, and the traditional interpolation and fitting methods are less effective in completing high-frequency measurement data. Summary of the invention

[0003] The present invention provides a method for completing missing data of electric energy meter measurement based on univariate dynamic time warping. The method identifies the time series pattern before and after the missing data, finds a historical subsequence similar to the missing data, and infers the missing data based on the subsequence. The historical data is then combined to assist in prediction. At the same time, considering that the electric energy meter data of most users are normal data without missing data, which can be used as a reference sequence pattern, the method also infers the missing data in combination with the normal data.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for completing missing data of electric energy meter measurement based on single variable dynamic time warping, comprising:

[0006] Construct an initial time series INP from normal non-missing electric energy meter data, and at the same time, determine the missing data of the missing sequence R, and place the sequence before the missing data of the missing sequence R after the initial time series INP to form a new time series INP′;

[0007] Adaptively create a query subsequence based on the length T of the missing data combined with the new time series INP′, and build a reference database based on the initial time series INP;

[0008] A dynamic time warping algorithm is used to calculate the similarity between the query subsequence and each candidate subsequence in the reference database, thereby obtaining a first candidate subsequence combination that is most similar to the query subsequence, and at the same time, a second candidate subsequence is obtained based on an autoregressive moving average model and the reference database;

[0009] The missing data is filled based on the first candidate subsequence combination and the second candidate subsequence.

[0010] As an optimization, the initial time series INP is specifically:

[0011] INP=[f1,f2,…,f m ]={x1,x2,…,x 96m};

[0012] Among them, f i is the electric energy meter data sequence, which is a 1×96 time series data, i∈{1,2,…,m}, m represents the total number of normal and non-missing electric energy meters, x n is the electric energy meter data at the nth time point in the initial time series INP, n∈{1,2,…,96m}.

[0013] As an optimization, the deletion sequence is expressed as:

[0014] R={r1,r2,…,r 96};

[0015] Among them, r h is the electric energy meter data of a single time point in the missing sequence, h∈{1,2,…,96};

[0016] The missing data are represented as:

[0017] r [t:t+T-1] ;

[0018] Wherein, t is the starting position of the missing data, and T is the length of the missing data.

[0019] As an optimization, the specific process of adaptively creating a query subsequence according to the length of the missing data is as follows:

[0020] At the end of the normal new time series INP′INP after concatenation, a window subsequence Q={x 96m-10T+t :x 96m+t-1}, let the window subsequence be the query subsequence.

[0021] As an optimization, the first electric energy meter data to the (96m-20T)th electric energy meter data in the initial time series INP are taken as the reference database, that is, the reference database is represented by S DB ={x1:x 96m-20T}.

[0022] As an optimization, a dynamic time warping algorithm is used to calculate the similarity between the query subsequence and each candidate subsequence in the reference database, specifically:

[0023] The query subsequence and the candidate subsequence are aligned by a dynamic time warping algorithm, and nonlinear time deformation is allowed to minimize the distance between the query subsequence and the candidate subsequence, wherein minimizing the distance between the query subsequence and the candidate subsequence is expressed as:

[0024]

[0025] Among them, Q k represents the kth electric energy meter data of the query subsequence Q, Y k (i) is the k-th electric energy meter data of the i-th candidate subsequence Y(i) in the reference database;

[0026] The first candidate subsequence combination is W first candidate subsequences that are most similar to the reference database, and the first candidate subsequence combination is expressed as:

[0027]

[0028] Among them, sort() means sorting from small to large.

[0029] As an optimization, the autoregressive moving average model is specifically expressed as:

[0030]

[0031] in, is the autoregressive coefficient, ε is white noise, p is the order of the autoregressive term, θ is the moving average coefficient, q is the order of the moving average term, and t is the starting position of the missing data. The second candidate subsequence based on the autoregressive moving average model is specifically: the second candidate subsequence is obtained based on the reference database and the autoregressive moving average model

[0032]

[0033] in, is the predicted value at time t.

[0034] As an optimization, the specific expression for filling the missing data based on the combination of the first candidate subsequence and the second candidate subsequence is:

[0035]

[0036] in, j mRepresents the number of the mth first subsequence.

[0037] The present invention also discloses a system for completing missing data of electric energy meter measurement based on single variable dynamic time warping, which is used to implement the aforementioned method for completing missing data of electric energy meter measurement based on single variable dynamic time warping, comprising:

[0038] A sequence construction module is used to construct an initial time series INP from normal non-missing electric energy meter data, and at the same time, determine the missing data of the missing sequence R, and place the sequence before the missing data of the missing sequence R after the initial time series INP to form a new time series INP′;

[0039] A query module, used to adaptively create a query subsequence according to the length T of the missing data, and build a reference database based on the initial time series INP;

[0040] a calculation module, configured to calculate the similarity between the query subsequence and each candidate subsequence in the reference database using a dynamic time warping algorithm, thereby obtaining a first candidate subsequence combination most similar to the query subsequence, and at the same time, a second candidate subsequence based on an autoregressive moving average model;

[0041] The execution module is used to fill the missing data based on the first candidate subsequence combination and the second candidate subsequence.

[0042] The present invention also discloses an electronic device, comprising: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping as described above.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] 1. Time series completion method based on dynamic time warping (DTW) algorithm:

[0045] The present invention proposes a method for completing electric energy meter data based on a dynamic time warping (DTW) algorithm. The method uses a nonlinear alignment technology between time series to accurately complete data by finding similar time series segments when data is missing. The method is particularly suitable for processing high-frequency measurement data or situations where data is missing over a long period of time.

[0046] 2. Adaptive matching and filling mechanism:

[0047] In the completion process, the DTW algorithm is not only based on the value of the time series, but also finds the best match by adaptively matching the time axis deformation between the sequences (which comes with the DTW algorithm). This method enables it to better capture the complex dynamic changes of the data, especially in the case of nonlinear fluctuations in the measurement of electric energy meters due to factors such as faults and interference. The completion method can still obtain reliable results.

[0048] 3. Efficient time series processing capabilities:

[0049] This method can efficiently process a large number of time series in the scenario of high-frequency data measurement of electric energy meters, and has good scalability. It is not limited to electric energy meter data, but is also applicable to other time series data completion problems with periodic or non-periodic characteristics. By calculating the optimal time alignment path, the present invention can reduce errors when completing missing data and improve the overall completion accuracy.

[0050] 4. Robustness to complex data missing situations:

[0051] The DTW algorithm of the present invention is not only suitable for missing single data points, but also can handle the situation of missing continuous data in a long period of time. In particular, when the electric energy meter is disturbed or fails, the data missing segment is long, and the traditional interpolation method cannot effectively solve it. The present invention can complete the long segment of data by finding similar time series segments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0053] Figure 1 It is a flow chart of a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to the present invention;

[0054] Figure 2 It is a schematic diagram of connecting normal and non-missing electric energy meter data in series in the embodiment;

[0055] Figure 3 A schematic diagram of missing electric energy meter data in an embodiment;

[0056] Figure 4 Schematic diagram of DTW calculation results in the embodiment;

[0057] Figure 5 It is the error histogram of the completed electric energy meter data and the real electric energy meter data. DETAILED DESCRIPTION

[0058] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0059] This embodiment 1 provides a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping, such as Figure 1 As shown, including:

[0060] S1. construct an initial time series INP from normal non-missing electric energy meter data, and at the same time, determine the missing data of the missing sequence R, and place the sequence before the missing data of the missing sequence R after the initial time series INP to form a new time series INP′;

[0061] In some embodiments, the initial time series INP is specifically:

[0062] INP=[f1,f2,…,f m ]={x1,x2,…,x 96m};

[0063] Among them, f i is the electric energy meter data sequence, which is a 1×96 time series data, i∈{1,2,…,m}, m represents the total number of normal and non-missing electric energy meters, x n is the electric energy meter data at the nth time point in the initial time series INP, n∈{1,2,…,96m}. One substation collects 96 data points per day.

[0064] The missing sequence is represented by:

[0065] R={r1,r2,…,r 96};

[0066] Among them, r h is the electric energy meter data at the hth time point in the missing sequence, h∈{1,2,…,96}.

[0067] The missing data are represented as:

[0068] r [t:t+T-1] ;

[0069] Wherein, t is the starting position of the missing data, and T is the length of the missing data.

[0070] That is, the normal and non-missing electric energy meter data is summarized as F = {f1, f2, ..., f m}, where m represents the total number of normal and non-missing electric energy meter data; f iis the electric energy meter data sequence, which is a 1×96 time series data, i∈{1,2,…,m}. Then all normal non-missing electric energy meter data are connected in series, that is, a 1×96m time series INP=[f1,f2,…,f m ]={x1,x2,…,x 96m}, where x i is the electric energy meter data at a single time point, i∈{1,2,…,96m}. x represents the electric energy meter data at a single time point.

[0071] The sequence of electric energy meters that needs to be completed is R = {r1, r2, ..., r 96}, some of the data are missing, and the missing data are marked as NA, that is, r [t:t+T-1] =NA, where t is the starting position of the missing data and T is the length of the missing segment.

[0072] Then the data before the deletion is placed after INP to form a new normal sequence INP′={INP,{r1,r2,…,r t-1}}.

[0073] S2. Adaptively create a query subsequence based on the length T of the missing data and the new time series INP′, and build a reference database based on the initial time series INP;

[0074] The missing sequence is equivalent to the missing data that has not been measured, but actually exists. The query subsequence is extracted from the end of the normal time series (here INP′, which contains the data before the missing sequence is missing) to include some information of the missing sequence. There are enough sequence samples in the power grid, so the missing sequence can find an approximate sequence in the normal sequence. The missing sequence is just that some values ​​are not measured. If the missing values ​​are measured, the missing sequence becomes a normal sequence.

[0075] The query subsequence is constructed as follows:

[0076] At the end of the new normal time series INP′ after concatenation, a window subsequence Q with a fixed length of 10T is extracted. 96m-10T+t :x 96m+t-1}, the subsequence is defined as the query subsequence before the missing data and is used for similarity matching with the existing data.

[0077] If the length of the query subsequence is too short, the error will be too large, and if the length is too long, the calculation pressure will be too large. Therefore, the present invention adopts a length one order of magnitude higher than the missing length as the length of the query subsequence.

[0078] The construction of the reference database is as follows:

[0079] The concatenated initial normal time series INP is used to construct the reference database S DB , which is used to perform similarity matching with the query subsequence Q. In the time series INP, the first electric energy meter data to the (96m-20T)th electric energy meter data are taken as the reference database, and the reference database is selected as S DB ={x1:x 96m-20T}.

[0080] The present invention hopes to perform a query in the previous normal sequence, so the reference database is this interval, which can avoid the query subsequence from querying itself.

[0081] S3, using a dynamic time warping algorithm to calculate the similarity between the query subsequence and each candidate subsequence in the reference database, thereby obtaining a first candidate subsequence combination that is most similar to the query subsequence, and at the same time, obtaining a second candidate subsequence based on an autoregressive moving average model and the reference database;

[0082] First, a dynamic time warping algorithm is used to calculate the similarity between the query subsequence and each candidate subsequence in the reference database, thereby obtaining a specific process of the first candidate subsequence combination most similar to the query subsequence.

[0083] In order to evaluate the similarity between the query subsequence Q and each candidate subsequence in the reference database, the dynamic time warping (DTW) algorithm is used. The DTW algorithm aligns two time series by a dynamic programming method, allowing nonlinear time deformation to minimize the distance between them. That is, the query subsequence and the candidate subsequence are aligned by a dynamic time warping algorithm, allowing nonlinear time deformation to minimize the distance between the query subsequence and the candidate subsequence. The DTW distance is defined as follows:

[0084]

[0085] Among them, Q k is the kth data of the query subsequence Q, Y k (i) is the kth data of the i-th candidate subsequence Y(i) in the reference database. The DTW distance calculates the optimal matching path to obtain the minimum distance between the two.

[0086] If the first candidate subsequence is x1~x 10T , then the second candidate subsequence is x2~x 10T+1 .

[0087] KNN-based matching window selection:

[0088] According to the DTW distance calculated in the previous steps, the W value of KNN is set to 5.

[0089] That is, the first candidate subsequence combination is W first candidate subsequences that are most similar to the reference database, and the first candidate subsequence combination is expressed as:

[0090]

[0091] Among them, sort() means sorting from small to large.

[0092] Select the 5 most similar subsequences in the reference database to the query subsequence Q That is, the 5 candidate subsequences with the smallest DTW distance are obtained:

[0093]

[0094] Among them, sort() means sorting from small to large. j1, j2, j3, j4, j5 represent the corresponding 5 subsequences The revelation number.

[0095] Initial completion of missing data:

[0096] The best matching W (in this embodiment, W=5) subsequences found through the above steps Align it with the time series of the missing part, and complete the missing data based on the matching result. The mean of the subsequent T data is used to fill in the missing data, that is: This step can complete the preliminary filling of missing data in the electricity meter.

[0097] In addition to calculating and

[0098] The specific process of obtaining the second candidate subsequence based on the autoregressive moving average model and the reference database is as follows.

[0099] First, design an autoregressive model (AR):

[0100] The autoregressive model is used to represent the relationship between current values ​​and past values. The mathematical expression of the AR model is:

[0101]

[0102] in, is the autoregressive coefficient, ε t is white noise (error term), y t ∈INP, p is the order of the autoregressive term, and here the present invention selects p=96 / 12=8 (ie, using the data of the last two hours to predict the data of the next time), and t is the starting position of the missing data.

[0103] Then, design the moving average model (MA):

[0104] The moving average model represents the relationship between the current value and the past forecast errors. The mathematical expression of the MA model is:

[0105] y t =θ1ε t-1 +θ2ε t-2 +…+θ q ε t-q +ε t

[0106] Among them, θ i is the moving average coefficient, q is the order of the moving average term, and here, the present invention takes q=4 (i.e., assuming that the current value is the weighted sum of the errors in the past hour).

[0107] Again, the autoregressive model and the moving average model are combined to construct the autoregressive moving average model (ARMA):

[0108] Combining the AR and MA parts to form an ARIMA model:

[0109]

[0110] Because of y t ∈INP, so the coefficients of the ARIMA model can be obtained using the least squares fitting method With θ i The goal of fitting is to minimize the variance of each error term.

[0111] Finally, forecasting is done using the autoregressive moving average model:

[0112] The fitted model is used for prediction. The prediction formula is based on the known reference database and model parameters, and usually uses the following recursive formula:

[0113]

[0114] in is the predicted value at time t+h, so the value of the missing part can be obtained by this formula

[0115] S4. Fill in the missing data based on the first candidate subsequence combination and the second candidate subsequence.

[0116] The specific expression for filling the missing data based on the combination of the first candidate subsequence and the second candidate subsequence is:

[0117]

[0118] in, j m Represents the number of the mth first subsequence.

[0119] After a lot of data experiments, when 70% of the prediction results based on DTW are combined with 30% of the prediction results based on ARMA, the overall prediction effect is the best, that is, α = 0.7, β = 0.3. Therefore, the specific completion data is expressed as:

[0120]

[0121] in, j m Represents the number of the mth first subsequence.

[0122] The present invention can achieve the following effects:

[0123] 1. Improve data completion accuracy:

[0124] Compared with traditional linear interpolation, polynomial fitting and other methods, the DTW algorithm can provide higher completion accuracy when processing complex electricity meter data by considering the dynamic changes and nonlinear deformation of time series. Traditional interpolation methods are usually only applicable to small ranges or regular missing, while DTW can maintain high completion accuracy even when large ranges are missing through global optimal matching.

[0125] 2. Adapt to the dynamic changes of nonlinear time series:

[0126] The DTW algorithm used in the present invention allows the time series to be nonlinearly distorted on the time axis, so as to capture the sudden changes, spikes or other abnormal fluctuations that may appear in the electric energy meter measurement data. In contrast, traditional interpolation and fitting methods often have poor effects when facing such dynamic changes and are prone to large errors.

[0127] 3. Suitable for high-frequency data missing completion, with high robustness:

[0128] The measurement frequency of electric energy meters is usually high, and thousands or even more data points may be collected every day. In such high-frequency data, the completion effect of traditional methods usually decreases rapidly as the amount of missing data increases. The DTW algorithm can effectively handle large-scale and long-term missing data in electric energy meter data, and maintain strong robustness, and can complete missing data for multiple days or even weeks.

[0129] 4. High tolerance to data noise and anomalies:

[0130] The present invention can reduce the influence of noise on the completion result by matching the shape (rather than the value) of the time series. In particular, for common interference and noise in electric energy meter data, the DTW algorithm can effectively filter out these abnormal information, thereby improving the reliability of the completed data.

[0131] 5. Strong scalability and versatility:

[0132] This method can also be extended to multi-dimensional time series completion. By introducing additional data dimensions (such as voltage, current, etc.), the accuracy of the completion results can be further improved.

[0133] 6. Potential for real-time applications:

[0134] Since the DTW algorithm can efficiently process large-scale time series data and has the possibility of online application, the algorithm can be integrated into smart meters in the future to achieve real-time data monitoring and missing data completion, providing important technical support for fault detection and recovery in smart grids.

[0135] Although the present invention is aimed at completing single-dimensional electric energy meter data, the method is also applicable to the extended application of multi-dimensional time series. In future application scenarios, various data of the electric energy meter (such as voltage, current, power, etc.) can be used as multi-dimensional input, and more complex data completion tasks can be achieved by expanding the similarity calculation method of the DTW algorithm.

[0136] Taking a 10kV substation as an example, there are 65 electricity meters under the jurisdiction of this substation, of which 64 have completely normal data and 1 needs to complete the data.

[0137] Step 1: Connect the data of each group of 64 normal electric energy meters in series through step 1, and the result is as follows: Figure 2 shown.

[0138] That is, normal data INP = {x1, x2, ..., x 6144}.

[0139] Step 2: Then visualize the missing electricity meter data that needs to be filled, such as Figure 3 shown.

[0140] Step 3: The missing part starts from time 50 and ends at time 60, with a total of 11 missing data, so the length T is set to 11. Select the reference database as S DB ={x1:x 5924}, window subsequence Q = {x 6033 :x 6144}. Then, the DTW distance between all subsequences in the reference database and subsequence Q is calculated by moving the window, and the five subsequences with the minimum DTW distance are found. The DTW calculation results are as follows: Figure 4 shown.

[0141] Step 4: Use the found sequence to replace the missing corresponding part. Then use ARMA for auxiliary correction and finally replace. The replacement result is as follows. This data sequence is the completed electric energy meter data sequence.

[0142] The error histogram of this data compared with the true data is as follows Figure 5 The average error is 0.1413, and the accuracy of minute-level data completion is 99.94%.

[0143] Embodiment 2 further discloses a system for completing missing data of electric energy meter measurement based on single variable dynamic time warping, which is used to implement the aforementioned method for completing missing data of electric energy meter measurement based on single variable dynamic time warping, including:

[0144] A sequence construction module is used to construct an initial time series INP from normal non-missing electric energy meter data, and at the same time, determine the missing data of the missing sequence R, and place the sequence before the missing data of the missing sequence R after the initial time series INP to form a new time series INP′;

[0145] A query module, used to adaptively create a query subsequence according to the length T of the missing data, and build a reference database based on the initial time series INP;

[0146] a calculation module, configured to calculate the similarity between the query subsequence and each candidate subsequence in the reference database using a dynamic time warping algorithm, thereby obtaining a first candidate subsequence combination most similar to the query subsequence, and at the same time, a second candidate subsequence based on an autoregressive moving average model;

[0147] The execution module is used to fill the missing data based on the first candidate subsequence combination and the second candidate subsequence.

[0148] Example 3 also discloses an electronic device, comprising: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping as described above.

[0149] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for completing missing data of electric energy meter measurement based on single variable dynamic time warping, characterized in that: include: Construct an initial time series INP from normal non-missing electric energy meter data, and at the same time, determine the missing data of the missing sequence R, and place the sequence before the missing data of the missing sequence R after the initial time series INP to form a new time series INP′; Adaptively create a query subsequence based on the length T of the missing data combined with the new time series INP′, and build a reference database based on the initial time series INP; A dynamic time warping algorithm is used to calculate the similarity between the query subsequence and each candidate subsequence in the reference database, thereby obtaining a first candidate subsequence combination that is most similar to the query subsequence, and at the same time, a second candidate subsequence is obtained based on an autoregressive moving average model and the reference database; The missing data is filled based on the first candidate subsequence combination and the second candidate subsequence.

2. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 1 is characterized in that: The initial time series INP is specifically: INP=[f1,f2,...,f m ]={x1,x2,…,x 96m }; Among them, f i is the electric energy meter data sequence, which is a 1×96 time series data, i∈{1,2,…,m}, m represents the total number of normal and non-missing electric energy meters, x n is the electric energy meter data at the nth time point in the initial time series INP, n∈{1,2,...,96m}.

3. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 2 is characterized in that: The missing sequence is represented by: R＝{r1,r2,…,r 96}； Among them, r h is the electric energy meter data of a single time point in the missing sequence, h∈{1,2,…,96}; The missing data are represented as: r[ t:t+T-1] ； Wherein, t is the starting position of the missing data, and T is the length of the missing data.

4. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 1, characterized in that: The specific process of adaptively creating a query subsequence according to the length of the missing data is as follows: At the end of the normal new time series INP′ after concatenation, a window subsequence Q={x 96m-10T+t :x 96m+t-1 }, let the window subsequence be the query subsequence.

5. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 1, characterized in that: In the initial time series INP, the first electric energy meter data to the (96m-20T)th electric energy meter data are taken as the reference database, that is, the reference database is represented by S DB ={x1:x 96m-20T }.

6. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 1, characterized in that: The dynamic time warping algorithm is used to calculate the similarity between the query subsequence and each candidate subsequence in the reference database, specifically: The query subsequence and the candidate subsequence are aligned by a dynamic time warping algorithm, and nonlinear time deformation is allowed to minimize the distance between the query subsequence and the candidate subsequence, wherein minimizing the distance between the query subsequence and the candidate subsequence is expressed as: Among them, Q k represents the kth electric energy meter data of the query subsequence Q, Y k (i) is the k-th electric energy meter data of the i-th candidate subsequence Y(i) in the reference database; The first candidate subsequence combination is W first candidate subsequences that are most similar to the reference database, and the first candidate subsequence combination is expressed as: Among them, sort() means sorting from small to large.

7. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 6, characterized in that: The autoregressive moving average model is specifically expressed as: in, is the autoregressive coefficient, p is white noise, p is the order of the autoregressive term, θ is the moving average coefficient, q is the order of the moving average term, and t is the starting position of the missing data; the second candidate subsequence based on the autoregressive moving average model is specifically: the second candidate subsequence is obtained based on the reference database and the autoregressive moving average model in, is the predicted value at time t.

8. The method for completing missing data of electric energy meter measurement based on single variable dynamic time warping according to claim 7, characterized in that: The specific expression for filling the missing data based on the combination of the first candidate subsequence and the second candidate subsequence is: in, j m Represents the number of the mth first subsequence.

9. A system for completing missing data of electric energy meter measurement based on single variable dynamic time warping, used to implement a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping as described in any one of claims 1 to 8, characterized in that: include: A sequence construction module is used to construct an initial time series INP from normal non-missing electric energy meter data, and at the same time, determine the missing data of the missing sequence R, and place the sequence before the missing data of the missing sequence R after the initial time series INP to form a new time series INP′; A query module, used to adaptively create a query subsequence according to the length T of the missing data, and build a reference database based on the initial time series INP; a calculation module, configured to calculate the similarity between the query subsequence and each candidate subsequence in the reference database using a dynamic time warping algorithm, thereby obtaining a first candidate subsequence combination most similar to the query subsequence, and at the same time, a second candidate subsequence based on an autoregressive moving average model; The execution module is used to fill the missing data based on the first candidate subsequence combination and the second candidate subsequence.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes a method for completing missing data of electric energy meter measurement based on single variable dynamic time warping as described in any one of claims 1 to 8.

Images