Abnormal repair method and prediction analysis method for power grid load data

Abstract

The present application relates to an abnormal repair method and a prediction analysis method for power grid load data, wherein the abnormal repair method comprises: determining target node data in the node data to be repaired of the power grid load data sequence, defining the target node data as the i-th node data of the k-th day; determining the estimated value of the target node data by linear interpolation; determining the estimated value of the i-th node data of the k-1-th day by linear interpolation, and determining the difference between the actual value and the estimated value of the i-th node data of the k-1-th day as the first window error; determining the estimated value of the i-th node data of the k-2-th day by linear interpolation, and determining the difference between the actual value and the estimated value of the i-th node data of the k-2-th day as the second window error; correcting the estimated value of the target node data by the first window error and / or the second window error. The problem of low repair accuracy of power grid load data by traditional data repair methods is solved.

Description

Technical Field

[0001] The present application relates to the field of power data processing, and in particular to an abnormality repair method and a prediction and analysis method for power grid load data. Background Art

[0002] Due to the instability of the current power company's power consumption information collection system in the process of data communication transmission using power carrier or wireless communication, the performance and quality of the collection terminal still fluctuate, resulting in some power measurement data cannot be collected in time, cannot be collected normally or cannot be collected correctly, resulting in low data integrity and poor data quality. The lack and abnormality of these data result in the collected power grid load data cannot be directly used for the overall analysis and application of power business, which seriously affects the practical application process of the system. Therefore, it is necessary to repair the missing nodes or abnormal nodes in the power grid load data sequence.

[0003] Traditional data repair methods are based on statistical information, such as using high-order probability distribution fitting for data cleaning. This data repair method makes less use of historical data information, and thus has low repair accuracy, especially for power grid load data sequences with historical regularities.

[0004] There is currently no effective solution to the problem that traditional data repair methods have low repair accuracy for power grid load data. Summary of the invention

[0005] The present invention provides an abnormal repair method and a prediction analysis method for power grid load data to solve the problem that the traditional data repair method has a low repair accuracy rate for power grid load data.

[0006] In a first aspect, the present invention provides a method for repairing abnormality of power grid load data, comprising:

[0007] Determine target node data in the node data to be repaired in the power grid load data sequence, and define the target node data as the i-th node data on the k-th day;

[0008] Based on the in-th node data and the i+n-th node data of the k-th day, determining the estimated value of the target node data by linear interpolation;

[0009] Based on the in-th node data and the i+n-th node data on the k-1-th day, determine the estimated value of the i-th node data on the k-1-th day by linear interpolation, and determine the difference between the actual value and the estimated value of the i-th node data on the k-1-th day as the first window error;

[0010] Based on the in-th node data and the i+n-th node data on the k-2-th day, an estimated value of the i-th node data on the k-2-th day is determined by linear interpolation, and the difference between the actual value and the estimated value of the i-th node data on the k-2-th day is determined as the second window error;

[0011] The estimated value of the target node data is corrected using the first in-window error and / or the second in-window error.

[0012] In a second aspect, the present invention provides a method for predicting and analyzing power grid load data, comprising:

[0013] Obtaining original power grid load data sequence;

[0014] Repairing the original power grid load data sequence by the abnormal repair method for power grid load data described in the first aspect to obtain a target power grid load data sequence;

[0015] Predict future changes in the target power grid load data series.

[0016] Compared with the related art, the present invention can obtain the estimated value of the target node data more accurately through the optimized interpolation method, and the target node data can be repaired by the estimated value to solve the problems of missing or abnormal values ​​of the target node data. Therefore, the problem of low repair accuracy of power grid load data by traditional data repair methods is solved.

[0017] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of a method for repairing abnormality of power grid load data provided in an embodiment of the present invention;

[0019] Figure 2 A comparison chart of actual power load and predicted power load in the first example;

[0020] Figure 3 A comparison chart of actual power load and predicted power load in the second example;

[0021] Figure 4 A comparison chart of actual power load and predicted power load in the third example;

[0022] Figure 5 This is a comparison chart of the actual power load and the predicted power load in the fourth example. DETAILED DESCRIPTION

[0023] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0024] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0025] In order to solve the problem of low accuracy of traditional data repair methods for power grid load data, the present invention has found through research that the changes in load data show obvious daily and weekly dependence, and the unbiased estimate of the power load at a certain point in time is often similar to the load value at the corresponding point in time of the adjacent day or adjacent week. When the data contains available historical data, the missing values ​​can be estimated using the linear interpolation method to complete the missing value repair work. Based on the above technical concept, the present invention provides a method for repairing abnormal power grid load data.

[0026] In an embodiment of the present invention, a method for repairing abnormality of power grid load data is provided. Figure 1 is a flow chart of a method for repairing abnormality of power grid load data provided in an embodiment of the present invention. Figure 1 As shown, the process includes step S110, step S120, step S130, step S140 and step S150.

[0027] Step S110, determining target node data in the node data to be repaired in the power grid load data sequence, and defining the target node data as the i-th node data on the k-th day.

[0028] The target node data is the node data that currently needs to be repaired. The node data that needs to be repaired mainly includes missing node data and abnormal node data. In either case, the new value of the target node data needs to be determined through subsequent steps to supplement and correct the target node data. For ease of understanding, in the power grid load data sequence, the order of node data can be represented by serial numbers k and i.

[0029] Step S120, determining an estimated value of the target node data by linear interpolation based on the in-th node data and the i+n-th node data of the k-th day.

[0030] Linear interpolation is a traditional method of repairing sequence data, which mainly uses the node data on both sides of the repaired node data (usually adjacent nodes) to estimate the repaired node data. For example, when n is 1 (the adjacent node data of the target node data is normal), the calculation formula for the estimated value of the target node data is:

[0031]

[0032] Simplifying, we get:

[0033]

[0034] in, represents the estimated value of the i-th node data on the k-th day, Φ k (i-1) represents the actual value of the i-1th node data on the kth day, Φ k (i+1) represents the actual value of the i+1th node data on the kth day.

[0035] From the above examples, we can see that the general calculation formula for the estimated value of the i-th node data is:

[0036]

[0037] Among them, Φ k (in) represents the actual value of the in-th node data on the k-th day, Φ k (in) represents the actual value of the i+nth node data on the kth day.

[0038] It should be noted that when the target node data is independent abnormal node data, the estimated value can be calculated based on the adjacent node data of the target node data through the above linear interpolation example process. In some uncommon cases, continuous abnormal node data may appear. For example, the i-th and i+1-th node data on the k-th day are both abnormal. When estimating the k-th node data, its adjacent node data cannot be used. At this time, the next adjacent node data can be used for linear interpolation estimation, that is, the i-2-th and i+2-th node data are used to estimate the i-th node data.

[0039] From the above description, it can be seen that when the target node data is isolated node data to be repaired, n is 1; when at least one adjacent node data of the target node data is node data to be repaired, n is greater than or equal to 1 to ensure that the inth node data and the i+nth node data on the kth day are not node data to be repaired.

[0040] Furthermore, the estimated value calculated only through the above steps ignores the error within the window, that is, the time domain of the i-th node data and its adjacent node data is different, which will cause unsatisfactory repair results. Assume that the time domain of the load data is Ω - , the time domain of the forecast data is Ω + , for a certain window width of load data P(t)(t∈Ω - ), if based on the time domain in Ω + The time domain obtained by load data prediction is in Ω - The load data prediction value is Q(t), then the error within the window is: ε(t)=P(t)-Q(t). Then, in the subsequent steps, the present invention eliminates the error within the window of the estimated value of the i-th node data.

[0041] Specifically, when the load patterns of different days are the same, the errors in the windows corresponding to the same time point (time period) on different days are similarly set. The present invention has found that power load data often has a certain periodicity. According to the research, short-term load repair with a window distance of seven days will achieve good results. Therefore, in order to optimize the estimated value and improve the accuracy of missing repair, it is considered to use the load values ​​at the corresponding time points of the day before, the two days before, and the seven days before the missing day to reduce the error in the window.

[0042] Step S130, based on the in-th node data and the i+n-th node data on the k-1-th day, determine the estimated value of the i-th node data on the k-1-th day by linear interpolation, and determine the difference between the actual value and the estimated value of the i-th node data on the k-1-th day as the first window error.

[0043] The calculation formula for the error in the first window is:

[0044]

[0045] Among them, ε k-1 (i) represents the error in the first window, Φ k-1 (i) represents the actual value of the i-th node data on the k-1th day, Φ k-1 (i-1) represents the actual value of the i-1th node data on the k-1th day, Φ k-1 (i+1) represents the actual value of the i+1th node data on the k-1th day.

[0046] Step S140, based on the in-th node data and the i+n-th node data on the k-2-th day, determine the estimated value of the i-th node data on the k-2-th day by linear interpolation, and determine the difference between the actual value and the estimated value of the i-th node data on the k-2-th day as the error within the second window.

[0047] The calculation formula for the error in the second window is:

[0048]

[0049] Among them, ε k-2 (i) represents the error in the second window, Φ k-2 (i) represents the actual value of the i-th node data on the k-2th day, Φ k-2 (i-1) represents the actual value of the i-1th node data on the k-2th day, Φ k-2 (i+1) represents the actual value of the i+1th node data on the k-2th day.

[0050] Step S150: Correcting the estimated value of the target node data using the first window error and / or the second window error.

[0051] By adding the estimated value of the target node data to the first window error or the second window error, the window error of the estimated value of the target node data can be greatly reduced, making the estimated value of the target node data more accurate. Correspondingly, the mean of the first window error or the second window error can also be added to the estimated value of the target node data, making the estimated value correction more stable.

[0052] Further, the estimated value of the target node data can be corrected by considering the node data of the k-7th day. Accordingly, in some embodiments, step S150 specifically includes: correcting the estimated value of the target node data by using the first window error, the second window error and the i-th node data of the k-7th day. The correction formula of the estimated value of the target node data is:

[0053]

[0054] Simplifying, we get:

[0055]

[0056] in, represents the estimated value of the corrected target node data, represents the estimated value of the target node data before correction, ε k-1 (i) represents the error in the first window, ε k-2 (i) represents the error in the second window, Φ k-7 (i) represents the actual value of the i-th node data on the k-7th day.

[0057] In the above correction formula, different parts have the same correction proportion on the target node data, which is one third.

[0058] Through the above optimized interpolation method, the estimated value of the target node data can be obtained more accurately, and the target node data can be repaired through the estimated value to solve the problem of missing or abnormal values ​​of the target node data. Therefore, the problem of low accuracy of traditional data repair methods for power grid load data repair is solved.

[0059] In an embodiment of the present invention, a method for predicting and analyzing power grid load data is also provided. The method flow includes step S210, step S220 and step S230.

[0060] Step S210, obtaining an original power grid load data sequence.

[0061] Step S220: repairing the original power grid load data sequence by the abnormal repairing method for power grid load data provided by the present invention to obtain a target power grid load data sequence.

[0062] Step S230, predicting future changes of the target power grid load data sequence.

[0063] Based on the above-mentioned abnormal repair method for power grid load data, the present invention further proposes a prediction and analysis method for power grid load data. First, the abnormal repair method for power grid load data provided by the present invention is used to repair the original power grid load data sequence, and the repaired target power grid load data sequence is predicted. Since the abnormal repair method more accurately repairs the abnormal data in the power grid load data sequence, the prediction is more accurate.

[0064] Preferably, the future changes of the target power grid load data sequence can be predicted by the ARIMA model. The ARIMA model is a time series analysis method that is currently widely used. In some embodiments, the time series analysis method of the ARIMA (p, q, d) model structure is used to predict and analyze the power load data. Through observations at a series of time points, the data obtained belongs to a time series. In order to better understand the random mechanism of the observation sequence, this embodiment uses the ARIMA model to model the target power grid load data sequence, and uses the model to predict or forecast the possible values ​​of the sequence in the future. The ARIMA model can be divided into an autoregressive model and a sliding average model according to the stationary characteristics of the time series. The original time series is stabilized by the difference method, and the type of model, the order of the model and the unknown parameters are judged by the characteristics of the autocorrelation and partial autocorrelation functions of the stationary series. After that, the effectiveness of the model is tested, and finally the future time series is analyzed and predicted.

[0065] Specifically, the step of predicting the future changes of the target power grid load data series through the ARIMA model includes step S310, step S320, step S330 and step S330.

[0066] Step S310, determining whether the target power grid load data sequence is stable, if so, executing step S220, if not, stabilizing the target power grid load data sequence and then executing step S220.

[0067] Step S320, determining the change trend of the autocorrelation value and the partial autocorrelation value of the target power grid load data sequence and determining the model order range according to the change trend.

[0068] Step S330, determining the model order within the model order range by using the AIC criterion, and estimating model parameters based on the model order.

[0069] Step S340, obtaining the ARIMA model based on the model order and the model parameters, and predicting future changes of the target power grid load data series through the ARIMA model.

[0070] The above process is explained below through a specific example.

[0071] First, for the ARIMA model, the ARIMA model is generally recorded as ARIMA (p, q, d), which represents the non-stationary random sequence variable Y t After d times of difference processing, we get the stationary sequence X t .

[0072] After that, the stationary sequence X tThe model is fitted as ARIMA(p,q), and the expression of the model is as follows:

[0073] X t =φ1X t-1 +…+φ p X t-p +ε t -θ1ε t-1 -…-θ q ε t-q ,t∈Z

[0074] The first half of the formula is the autoregressive process, p is the autoregressive order, φ1,…,φ p is the autoregressive coefficient; the second half is the sliding average process, q is the sliding average order, θ1,…,θ q is the sliding average coefficient; X t is a stationary data series, {ε t} is an unobserved white noise sequence in the form of W~N(0,σ 2 ).

[0075] In particular, when q = 0, the model becomes an autoregressive model AR(p), expressed as follows:

[0076] X t =φ1X t-1 +…+φ p X t-p +ε t ,t∈Z

[0077] When p = 0, the model becomes a moving average model MA(q), and the expression is as follows:

[0078] X t =ε t -θ1ε t-1 -…-θ q ε t-q ,t∈Z

[0079] The modeling process is as follows.

[0080] The first step is to determine the stationarity of the sequence and to stabilize the sequence (for step S310). When making statistical inferences about the structure of a random process based on observation records, it is usually necessary to make some simplified and reasonable assumptions, among which the most important assumption is stationarity. The basic idea of ​​stationarity is that the statistical laws that determine the characteristics of the process do not change with time. In a sense, the process is at a statistical equilibrium point. In particular, if the following two conditions are met for all time lags k and time t: first, the mean function is always an integer at all times; second, γ t,t-k =γ 0,k Among them, γ i,jis the autocovariance function.

[0081] The difference method is usually used to stabilize non-stationary series. The d-order difference can be recursively defined as:

[0082]

[0083] Among them, Y t is a non-stationary data series, is the first-order difference, is the second-order difference, and so on.

[0084] Starting from the first order difference, the sequence after each difference is tested for stationarity. If it fails, the difference is continued until it passes. The number of differences when passing is the parameter d in the ARIMA model.

[0085] Next is the identification of the ARIMA model (corresponding to step S320). This can be determined by calculating the trend of the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the stabilized sequence. k The calculation formula is as follows:

[0086]

[0087] Among them, X t is a stationary data sequence, and N is the length of the sequence.

[0088] The partial autocorrelation function (PACF) φ of a stationary data series k,j The calculation formula is as follows:

[0089]

[0090] Analyze the trends of the calculated ACF and PACF, and determine the type of ARMA(p,q) according to the table below.

[0091] Table 1 ARMA (p, q) identification principle table

[0092]

[0093] Then, the model order is determined and the parameters are estimated (corresponding to step S330). The approximate range of the optimal model order can be obtained from the trends of ACF and PACF. Here, the optimal model order is selected by checking the AIC criterion.

[0094] In general, AIC can be expressed as: AIC = 2k-2ln(L), where k is the number of parameters and L is the likelihood function. AIC encourages good data fitting but tries to avoid overfitting. Therefore, the order with the smallest AIC is selected as the optimal model order.

[0095] The methods of parameter estimation usually include moment estimation method, least squares estimation method, maximum likelihood estimation method, etc.

[0096] Finally, the model verification prediction is performed (corresponding to step S340). The white noise test is used to test whether there is a correlation between any sequence values ​​in the time series. It is also called a pure random test. White noise sequences cannot be used for analysis and prediction because there is no correlation between their sequence values, that is, historical values ​​have no effect on current values ​​and future values. However, in practice, the sample autocorrelation function of the pure time series of power data cannot be absolutely zero. Therefore, it can be considered that when the value of the autocorrelation function is near zero, the time series data can be treated as a purely random sequence.

[0097] The validity of the ARIMA model generated by fitting is tested, and usually its residual sequence is tested for white noise. If it is not white noise, the model needs to be re-identified and re-fitted before being tested.

[0098] The residual of the data is defined as e t =X t -μ t , where X t is the actual value of the data, μ t is the predicted value, from which the residual sequence {e t}, by judging {e t} is white noise to determine whether the model passes the validity test.

[0099] According to the above prediction model, the minimum mean square error prediction method is used to predict X t Make a prediction and consider the d differences made previously to restore the original data Y t prediction results.

[0100] Assume that Y1,Y2,…,Y t For historical data, predict the value Y for the next l period t+l , time t is called the prediction starting point, l is the prediction lead time, and Y t (l) represents the predicted value, and the formula is Y t (l) = E(Y t+l |Y1,Y2,…,Y t ).

[0101] As follows, the technical effects of the present invention are demonstrated through a number of verification examples.

[0102] Figure 2 This is a comparison chart of the actual power load and the predicted power load in the first example. Figure 2 In the first example (for the power load data of user A, the ARIMA model is directly used for prediction, and the data repair method provided by the present invention is not used for repair in the early stage), the prediction accuracy of the power load is 93.8%, and the curve similarity between the actual power load and the predicted power load is 98.17%.

[0103] Figure 3 This is a comparison chart of the actual power load and the predicted power load in the second example. Figure 3 In the second example (for the power load data of user B, the ARIMA model is directly used for prediction, and the data repair method provided by the present invention is not used for repair in the early stage), the prediction accuracy of the power load is 89.2%, and the curve similarity between the actual power load and the predicted power load is 94.44%.

[0104] The present invention also analyzes random load data, uses two models to repair and predict the load data, and compares the difference with the actual power load, calculates the forecast accuracy and curve similarity (the forecast accuracy and curve similarity have been intelligently realized). For ease of comparison, the actual power load and the model prediction results are displayed in the same figure, which is more intuitive. The following figures are the effect diagrams of the linear method and the ARIMA model fitting compared with the actual power load. The red line in the figure is the actual power load data, and the blue line is the load data obtained by repairing and predicting using the corresponding model.

[0105] Figure 4 This is a comparison chart of the actual power load and the predicted power load in the third example. Figure 4 In the third example (for the power load data of user A, the data repair method provided by the present invention is first used for repair, and then the ARIMA model is used for prediction), the prediction accuracy of the power load is 95.65%, and the curve similarity between the actual power load and the predicted power load is 99.65%.

[0106] Figure 5 This is a comparison chart of the actual power load and the predicted power load in the fourth example. Figure 5 In the fourth example (for the power load data of user A, the data repair method provided by the present invention is first used for repair, and then the ARIMA model is used for prediction), the prediction accuracy of the power load is 97.06%, and the curve similarity between the actual power load and the predicted power load is 99.86%.

[0107] It can be seen from the above verification results that the prediction and analysis method for power grid load data provided by the present invention has high accuracy.

[0108] In the embodiments of the present invention, there is also provided a device for repairing abnormalities of power grid load data and a device for predicting and analyzing power grid load data, which are used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. The terms "module", "unit", "subunit", etc. used below may be a combination of software and / or hardware that implements predetermined functions. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0109] The abnormal repair device of power grid load data includes:

[0110] The node determination module is used to determine the target node data in the node data to be repaired in the power grid load data sequence, and define the target node data as the i-th node data on the k-th day.

[0111] The first estimation module is used to determine the estimated value of the target node data through linear interpolation based on the in-th node data and the i+n-th node data on the k-th day.

[0112] The second estimation module is used to determine the estimated value of the i-th node data on the k-1th day through linear interpolation based on the in-th node data and the i+n-th node data on the k-1th day, and determine the difference between the actual value and the estimated value of the i-th node data on the k-1th day as the first window error.

[0113] The third estimation module is used to determine the estimated value of the i-th node data on the k-2th day through linear interpolation based on the in-th node data and the i+n-th node data on the k-2th day, and determine the difference between the actual value and the estimated value of the i-th node data on the k-2th day as the error within the second window.

[0114] A data correction module is used to correct the estimated value of the target node data through the first window error and / or the second window error.

[0115] The forecasting and analyzing device for power grid load data includes:

[0116] The data acquisition module is used to acquire the original power grid load data sequence.

[0117] The data repair module is used to repair the original power grid load data sequence by the abnormal repair method of power grid load data provided by the present invention to obtain a target power grid load data sequence.

[0118] The data prediction module is used to predict the future changes of the target power grid load data sequence.

[0119] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0120] In an embodiment of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the abnormal repair method for power grid load data or the predictive analysis method for power grid load data provided by the present invention.

[0121] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.

[0122] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.

Claims

1. A method for repairing abnormality of power grid load data, characterized in that: include: Determine target node data in the node data to be repaired in the power grid load data sequence, and define the target node data as the i-th node data on the k-th day; Based on the in-th node data and the i+n-th node data of the k-th day, determining the estimated value of the target node data by linear interpolation; Based on the in-th node data and the i+n-th node data on the k-1-th day, determine the estimated value of the i-th node data on the k-1-th day by linear interpolation, and determine the difference between the actual value and the estimated value of the i-th node data on the k-1-th day as the first window error; Based on the in-th node data and the i+n-th node data on the k-2-th day, an estimated value of the i-th node data on the k-2-th day is determined by linear interpolation, and the difference between the actual value and the estimated value of the i-th node data on the k-2-th day is determined as the second window error; The estimated value of the target node data is corrected using the first in-window error and / or the second in-window error.

2. The method for repairing abnormality of power grid load data according to claim 1, characterized in that: When the target node data is isolated node data to be repaired, n is 1; And / or, when at least one adjacent node data of the target node data is node data to be repaired, the n is greater than or equal to 1 to ensure that the inth node data and the i+nth node data of the kth day are not node data to be repaired.

3. The method for repairing abnormality of power grid load data according to claim 1, characterized in that: The correcting the first estimated value of the target node data by using the first in-window error and / or the second in-window error comprises: The estimated value of the target node data is corrected by using the first in-window error, the second in-window error and the i-th node data on the k-7th day.

4. The method for repairing abnormality of power grid load data according to claim 3, characterized in that: The correction formula for the estimated value of the target node data is: in, represents the estimated value of the corrected target node data, represents the estimated value of the target node data before correction, ε k-1 (i) represents the error in the first window, ε k-2 (i) represents the error in the second window, Φ k-7 (i) represents the actual value of the i-th node data on the k-7th day.

5. A method for predicting and analyzing power grid load data, characterized in that: include: Obtaining original power grid load data sequence; Repairing the original power grid load data sequence by the abnormal repair method for power grid load data according to any one of claims 1 to 4 to obtain a target power grid load data sequence; Predict future changes in the target power grid load data series.

6. The method for predicting and analyzing power grid load data according to claim 5, characterized in that: Predicting future changes in the target power grid load data sequence includes: The future changes of the target power grid load data series are predicted by the ARIMA model.

7. The method for predicting and analyzing power grid load data according to claim 6, characterized in that: Predicting the future changes of the target power grid load data series through the ARIMA model includes: Step S310, determining whether the target power grid load data sequence is stable, if yes, executing step S220, if no, stabilizing the target power grid load data sequence and then executing step S220; Step S320, determining the change trend of the autocorrelation value and the partial autocorrelation value of the target power grid load data sequence and determining the model order range according to the change trend; Step S330, determining the model order within the model order range by using the AIC criterion, and estimating model parameters based on the model order; Step S340, obtaining the ARIMA model based on the model order and the model parameters, and predicting future changes of the target power grid load data series through the ARIMA model.

8. A device for repairing abnormality of power grid load data, characterized in that: include: A node determination module, used to determine target node data in the node data to be repaired in the power grid load data sequence, and define the target node data as the i-th node data on the k-th day; A first estimation module, used to determine an estimated value of the target node data by linear interpolation based on the in-th node data and the i+n-th node data on the k-th day; A second estimation module is used to determine an estimated value of the i-th node data on the k-1-th day through linear interpolation based on the in-th node data and the i+n-th node data on the k-1-th day, and determine the difference between the actual value and the estimated value of the i-th node data on the k-1-th day as the first window error; A third estimation module is used to determine an estimated value of the i-th node data on the k-2th day by linear interpolation based on the in-th node data and the i+n-th node data on the k-2th day, and determine the difference between the actual value and the estimated value of the i-th node data on the k-2th day as the second window error; A data correction module is used to correct the estimated value of the target node data through the first window error and / or the second window error.

9. A prediction and analysis device for power grid load data, characterized in that: include: A data acquisition module, used to acquire the original power grid load data sequence; A data repair module, configured to repair the original power grid load data sequence by using the abnormal repair method for power grid load data according to any one of claims 1 to 4 to obtain a target power grid load data sequence; The data prediction module is used to predict the future changes of the target power grid load data sequence.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the abnormal repair method for power grid load data described in any one of claims 1 to 4 or the predictive analysis method for power grid load data described in any one of claims 5 to 7.

Images