Electrical load prediction method

Abstract

An electrical load prediction method is used for predicting electrical load data of an (N + 1) th day through an autoregressive moving average model according to electrical load data of an Nth day, and comprises the following steps: S1, preprocessing the electrical load data of the Nth day; s2, analyzing and processing the preprocessed electrical load data of the Nth day; s3, determining the number P of model autoregression items and the number Q of moving average items in the model for the analyzed and processed electrical load data of the Nth day; s4, generating an ARIMA according to P, Q and training data, substituting the analyzed and processed electrical load data of the Nth day into the ARIMA, and predicting the electrical load data of the (N + 1) th day; according to the invention, time sequence prediction is carried out on the electrical load data, and data support is provided for peak load shifting of the power grid and stable operation of the power grid.

Description

Technical Field

[0001] The present invention relates to the field of electricity consumption analysis, and in particular to a method for predicting electricity load. Background Art

[0002] The development and popularization of new energy in recent years has also brought new challenges and opportunities to the power grid and impacted local distribution networks.

[0003] Taking new energy vehicles as an example, first, the clustering of disorderly charging of new energy vehicles leads to load growth, especially charging during peak load periods, which will aggravate the peak-to-valley difference of power grid load and increase the burden on power system operation; second, the geographical clustering of disorderly charging of new energy vehicles causes local overload problems in local distribution networks, especially in residential areas and public fast charging stations; third, new energy vehicles are high-power, non-linear loads, and a large number of disorderly charged new energy vehicles connected to the power grid will affect the power quality of the distribution network. At the same time, as a flexible load resource, new energy vehicles can participate in the demand-side response of the power grid, shaving peaks and filling valleys for the power grid to ensure safe operation of the power grid.

[0004] How to use data analysis methods such as stability analysis and randomness test to process and analyze real-time charging load data has become an urgent problem that needs to be solved. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting power load. The present invention performs time series prediction on power load data, and provides data support for peak shaving and valley filling of power grid to ensure stable operation of power grid.

[0006] To achieve the above-mentioned purpose, the present invention provides a method for predicting electricity load, which predicts the electricity load data of the N+1th day through an autoregressive moving average model based on the electricity load data of the Nth day, and the method comprises: step S1, preprocessing the electricity load data of the Nth day; step S2, analyzing and processing the preprocessed electricity load data of the Nth day; step S3, determining the number of autoregressive terms P in the model and the number of sliding average terms Q in the model for the analyzed and processed electricity load data of the Nth day; step S4, generating an autoregressive moving average model based on the number of autoregressive terms P in the model, the number of sliding average terms Q in the model and training data, substituting the analyzed and processed electricity load data of the Nth day into the autoregressive moving average model, and predicting the electricity load data of the N+1th day.

[0007] Preferably, the preprocessing in step S1 includes: checking the integrity of the electricity load data on the Nth day, checking the continuity of the electricity load data on the Nth day, and checking the accuracy of the electricity load data on the Nth day; if any of the integrity, continuity, and accuracy tests fails, correcting the electricity load data on the Nth day.

[0008] Preferably, the step S2 includes: step S21, performing a stability analysis on the electricity load data of the Nth day after preprocessing, if the stability analysis passes, proceeding to step S22, if the stability analysis fails, proceeding to step S23; step S22, performing a randomness test on the electricity load data of the Nth day after the stability analysis, if the randomness test passes, proceeding to step S3, if the randomness test fails, proceeding to step S23; step S23, performing correction processing on the electricity load data of the Nth day after preprocessing, and returning to step S21.

[0009] Preferably, the stability analysis in step S21 adopts a unit root test.

[0010] Preferably, the unit root test performs stability analysis on the time series data of the preprocessed electricity load data on the Nth day; the analyzed time series data includes mean, variance and autocovariance.

[0011] Preferably, the randomness test in step S22 adopts white noise test.

[0012] Preferably, the correction process in step S23 includes setting reasonable parameters or performing differential processing on the data.

[0013] Preferably, step S3 includes using a unit root test and a partial autocorrelation function to obtain the number P of autoregressive terms in the model and the number Q of sliding average terms in the model.

[0014] Preferably, step S3 includes using an AIC function to obtain the number P of autoregressive terms in the model and the number Q of sliding average terms in the model.

[0015] Preferably, the step S4 further comprises using the mean absolute percentage error to measure the running accuracy of the autoregressive moving average model;

[0016]

[0017] Wherein, n represents the number of times the power load data for the N+1th day is predicted within a single day, Ai represents the power load data for the Nth day input for the ith time, and Fi represents the power load data for the N+1th day predicted by the autoregressive moving average model for the ith time; if the mean absolute percentage error exceeds the preset threshold, intervention measures are taken to ensure the accuracy of the power load time series prediction; if it does not exceed the threshold, the status quo is maintained until the prediction is completed.

[0018] In summary, compared with the prior art, the power load forecasting method provided by the present invention has the following beneficial effects:

[0019] First, the present invention adopts ARIMA, and the result is more accurate;

[0020] Second, the present invention uses a rolling prediction method to perform prediction, and the real-time performance of the prediction data is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a partial schematic diagram of electricity load data in 2023 according to an embodiment of the present invention.

[0022] Figure 2 This is a partial schematic diagram of power load data in 2023 at 0 o'clock in an embodiment of the present invention.

[0023] Figure 3 This is a partial schematic diagram of electricity load data in 2023 at 6 o'clock in an embodiment of the present invention.

[0024] Figure 4 (a) is a data diagram of an embodiment of the present invention using ACF and PACF at time 00:00:00; Figure 4 (b) is a data graph of an embodiment of the present invention using ACF and PACF at time 06:00:00.

[0025] Figure 5 This is a comparison chart of actual and predicted electricity load data in 2023 at 0 o'clock in an embodiment of the present invention.

[0026] Figure 6 This is a comparison chart of actual and predicted electricity load data in 2023 at 6 o'clock in an embodiment of the present invention.

[0027] Figure 7 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the attached embodiment of the present invention Figure 1 ~Attached Figure 7 , the technical solutions, structural features, objectives achieved and effects in the embodiments of the present invention are described in detail.

[0029] It should be noted that the drawings are in a very simplified form and use non-precise proportions. They are only used to conveniently and clearly assist in explaining the embodiments of the present invention, and are not used to limit the conditions for the implementation of the present invention. Therefore, they have no substantive technical significance. Any structural modification, change in proportional relationship or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical content disclosed by the present invention.

[0030] It should be noted that, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only the elements explicitly listed, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0031] The present invention provides a method for predicting power load. The method predicts the power load data of the N+1th day (possible future data) through ARIMA (Autoregressive Integrated Moving Average Model) based on the power load data of the Nth day (existing data), realizes the time series prediction of the power load data, and provides data support for peak shaving and valley filling of the power grid to ensure the stable operation of the power grid. The principle of ARIMA is that ARIMA regards the data sequence formed by the prediction object over time as a random sequence, so as to use the past value to predict the future value.

[0032] like Figure 7 As shown, the method includes:

[0033] Step S1, preprocessing the electricity load data of the Nth day;

[0034] Preprocessing refers to the preliminary processing of the electricity load data on the Nth day, such as ensuring its integrity and continuity.

[0035] Step S2, analyzing and processing the pre-processed electricity load data for the Nth day;

[0036] The analysis and processing mainly refers to the stability analysis and randomness test (described later) of the pre-processed electricity load data on the Nth day, so as to ensure the feasibility of substituting the data into ARIMA for prediction in the subsequent steps.

[0037] Step S3, determining the number of autoregressive terms P in the model and the number of sliding average terms Q in the model for the Nth day's power load data after analysis and processing;

[0038] Because the premise of substituting the power load data into ARIMA is to determine P and Q in the power load data, P and Q have the function of pre-setting the basic parameters of ARIMA such as size and cycle period applicable to the power load data; the present invention proposes two schemes to determine P and Q, which will be described in detail later.

[0039] Step S4, generating ARIMA according to the number of autoregressive terms P in the model, the number of sliding average terms Q in the model and the training data, substituting the analyzed and processed electricity load data of the Nth day into ARIMA, and predicting the electricity load data of the N+1th day;

[0040] The roles of P and Q are as described in step S3. The training data refers to historical data, such as electricity load data in previous years, which is used to construct ARIMA. It can be understood here that the training data is used as the basis for constructing ARIMA, and P and Q are used to further optimize ARIMA to make it suitable for the electricity load data on the Nth day after analysis and processing.

[0041] The specific implementation of the above steps is introduced below.

[0042] In step S1, the electricity load data for the Nth day is existing data, and the preprocessing is to ensure that the electricity load data for the Nth day meets the following three requirements: (1) Data integrity: the current electricity load data is complete in terms of value, and there is no null value data; (2) Data continuity: the current electricity load data is continuous in time, and there is no missing time; (3) Data accuracy: there is no data fluctuation in the current electricity load data; if any of the items has a problem, intervention and correction are required.

[0043] The step S2 specifically includes:

[0044] Step S21, performing stability analysis on the pre-processed power load data of the Nth day, if the stability analysis is passed, proceeding to step S22, if the stability analysis is not passed, proceeding to step S23;

[0045] As mentioned above, because it is necessary to ensure the feasibility of substituting data into ARIMA for prediction, one of the important requirements is that the time series data (which is the relevant data of time flow) of the pre-processed electricity load data on the Nth day is stable. If the time series data is unstable, it will cause deviations in time prediction during ARIMA calculation. Therefore, the present invention uses ADF test (Augmented Dickey-Fuller test is a statistical method for detecting whether time series data is stable, also known as unit root test) to perform stability analysis on the time series data of the pre-processed electricity load data on the Nth day; if the time series data is stable, it is manifested as the mean, variance and autocovariance of the time series data do not change with time.

[0046] Among them, the advantage of stable time series data is that if the time series data is stable, the fitting curve obtained based on the time series data can still move along the existing form "inertia" in the future, that is, it has "consistency", which is convenient for subsequent analysis and processing.

[0047] The principle of the ADF test is to calculate the ADF statistic T, which is a collection of mean, variance and autocovariance. If the T value is less than the statistical threshold, it means that the time series data is stationary. If the T value is greater than the statistical threshold, the time series data is non-stationary. The statistical threshold can be determined by a value set based on historical conditions or normal conditions.

[0048] Step S22, performing a randomness test on the power load data of the Nth day after the stability analysis, if the randomness test passes, proceeding to step S3, if the randomness test fails, proceeding to step S23;

[0049] The power load data after stability analysis does not mean that it is feasible to substitute it into ARIMA for prediction. It needs to be further tested for randomness, that is, white noise test, to satisfy the white noise W < white noise threshold.

[0050] White noise is a random signal in the power load data, which is characterized by the fact that the values ​​at each time point are independent and have the same distribution, and the autocorrelation coefficients of all white noises are zero; before substituting the power load data into ARIMA for prediction, a white noise test is required to confirm that the power load data under the time series data is predictable. Among them, the white noise threshold can be determined by a value set based on historical conditions or normal conditions.

[0051] Step S23, correcting the pre-processed electricity load data for the Nth day, and returning to step S21.

[0052] Among them, the correction processing can adopt various means, such as manually setting reasonable parameters, performing differential processing on data, etc., which are not limited here. It can be achieved through stability analysis and randomness testing. This is the existing technology and will not be elaborated here.

[0053] In S3 described above, two schemes are proposed to determine the number of autoregressive terms P in the model and the number of sliding average terms Q in the model.

[0054] The first method is to obtain P and Q through ACF (Autocorrelation Function) and PACF (Partial Auto-Correlation Function); both ACF and PACF measure the correlation between the value at a certain moment in the power load data and the historical value. Specifically, ACF represents the correlation under the existence of other lagging factors; PACF represents the pure correlation after removing other lagging factors. In practical applications, ACF needs to focus on the point where the autocorrelation coefficient drops significantly to zero, and PACF needs to focus on the point where the partial autocorrelation coefficient drops significantly to zero.

[0055] The second method is to obtain P and Q through the AIC function (Akaike Information Criterion, an indicator used to measure the goodness of fit of a statistical model, proposed by Japanese statistician Hiroji Akaike in 1974).

[0056] In the above S4, in a preferred embodiment, in order to ensure the stability of ARIMA prediction operation, the present invention also uses MAPE (Mean Absolute Percentage Error) to measure the operation accuracy of ARIMA. MAPE is a commonly used indicator to measure the accuracy of ARIMA. It is used to calculate the relative error between the ARIMA prediction value (electricity load data on the N+1th day) and the actual value (electricity load data on the Nth day), and is expressed as a percentage.

[0057]

[0058] Where n represents the number of times the electricity load data for the N+1th day is predicted within a single day, Ai represents the electricity load data for the Nth day input for the ith time (actual value, there are values ​​at each time point in the Nth day, and each input value should be the existing value), and Fi represents the electricity load data for the N+1th day output by the ARIMA forecast for the ith time (predicted value).

[0059] If the MAPE exceeds the preset threshold, manual intervention measures are taken to ensure the accuracy of the power load timing forecast. If it does not exceed the threshold, the status quo is maintained until the forecast is completed. The higher the MAPE, the higher the accuracy. In addition, the threshold can be determined by a value set based on historical conditions or normal conditions.

[0060] The following will be combined Figure 1 to Figure 6 , the specific embodiments of the present invention are described in detail, such as Figure 1 , Figure 2 , Figure 3 As shown, the power load data of the embodiment of the present invention is based on the national standard 32960, that is, the data collected by the technical specification of the electric vehicle remote service and management system, and is sourced from the charging data of new energy vehicles in Jiading District of a certain city. It is updated every 2 minutes, and the time span is from January 1, 2023 to December 31, 2023.

[0061] Taking the prediction of May 18, 2023 data based on the data on May 17, 2023 as an example, the data on May 17, 2023 is preprocessed for data integrity, data continuity, and data accuracy, and then the preprocessed data on May 17, 2023 is obtained, and step S1 is completed at this time.

[0062] When entering step S2, it is necessary to analyze and process the pre-processed data on May 17, 2023; in the original data on May 17, 2023, the maximum load is approximately at 00:00:00 every day, and the minimum load is approximately at 06:00:00, so in the pre-processed data on May 17, 2023, these two moments are selected for data analysis to facilitate analysis and judgment (the same applies to subsequent steps S3 and S4).

[0063] Therefore, in a specific embodiment of the present invention, the ADF test results are as follows:

[0064] At 00:00:00, the T value is -1.288, which is greater than 95% × statistical threshold = -2.873, and also greater than 90% × statistical threshold = -2.573;

[0065] At 06:00:00, the T value = -0.673, which is greater than 95% × statistical threshold = -2.873, and also greater than 90% × statistical threshold = -2.573.

[0066] From the above, it can be seen that the ADF test T values ​​are all greater than the statistical threshold, indicating that the time series data of the preprocessed data on May 17, 2023 is non-stationary and needs to be corrected. This embodiment uses differential processing to make it stable. As for why the values ​​of 95% and 90% of the statistical thresholds are compared, the error fluctuation is taken into account and the range is expanded. However, even so, the T value is still greater than the statistical threshold, so correction is a necessary step.

[0067] After the correction is completed, a randomness test is performed, where the white noise threshold is 0.05, the white noise W = 1.870373e-210 at 00:00:00, and the white noise W = 0.000000e+0 at 06:00:00. The above two time points pass the white noise test, and the "data after analysis and processing on May 17, 2023" is output to complete step S2.

[0068] Here, we proceed to step S3 to determine P and Q, as shown in Figure 4 (a) and Figure 4 As shown in (b), under the first method, the calculation results are as follows: at 00:00:00, the P value is 20 and the Q value is 1; at 06:00:00, the P value is 12 and the Q value is 4.

[0069] Under the second method, the AIC results are as follows: at 00:00:00, the P value is 20 and the Q value is 1; at 06:00:00, the P value is 12 and the Q value is 4.

[0070] After P and Q are determined, the first step is to generate ARIMA based on P, Q and training data. The training data at this time is the electricity load from January 1, 2023 to May 17, 2023. Because it is assumed that the data on May 18, 2023 needs to be predicted, May 18, 2023 is considered unknown, and then ARIMA is generated.

[0071] Substitute the analyzed data of May 17, 2023 into the generated ARIMA, and predict the electricity load data of May 18, 2023, which is a complete process.

[0072] In addition, the present invention Figure 5 , Figure 6 A comparison chart of the predicted data for the whole year of 2023 and the actual data for the whole year of 2023 based on the generated ARIMA (for example, using the electricity load for the whole year of 2022 as training data) is shown. It can be clearly concluded that the accuracy of the prediction result at 00:00:00 is 94.29%; the accuracy of the prediction result at 06:00:00 is 94.14%; therefore, the present invention has considerable technical effects.

[0073] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A method for predicting power load, characterized in that: The method predicts the electricity load data on the N+1th day through an autoregressive moving average model based on the electricity load data on the Nth day, and the method includes: Step S1, preprocessing the electricity load data for the Nth day; Step S2, analyzing and processing the pre-processed electricity load data for the Nth day; Step S3, determining the number of autoregressive terms P in the model and the number of sliding average terms Q in the model for the Nth day's power load data after analysis and processing; Step S4, generating an autoregressive moving average model according to the number of autoregressive terms P in the model, the number of sliding average terms Q in the model and the training data, substituting the analyzed and processed electricity load data of the Nth day into the autoregressive moving average model, and predicting the electricity load data of the N+1th day.

2. A method for predicting power load according to claim 1, characterized in that: The pre-processing in step S1 includes: Verify the integrity, continuity, and accuracy of the electricity load data on the Nth day; If any of the tests of completeness, continuity and accuracy fails, the electricity load data for the Nth day shall be corrected.

3. A method for predicting power load according to claim 2, characterized in that: The step S2 comprises: Step S21, performing stability analysis on the pre-processed power load data of the Nth day, if the stability analysis is passed, proceeding to step S22, if the stability analysis is not passed, proceeding to step S23; Step S22, performing a randomness test on the Nth day power load data after the stability analysis, if the randomness test is passed, proceeding to step S3, if the randomness test is not passed, proceeding to step S23; Step S23, correcting the pre-processed electricity load data for the Nth day, and returning to step S21.

4. A method for predicting power load according to claim 3, characterized in that: The stability analysis in step S21 adopts a unit root test.

5. A method for predicting power load according to claim 4, characterized in that: The unit root test performs stability analysis on the time series data of the preprocessed electricity load data on the Nth day; the analyzed time series data includes mean, variance and autocovariance.

6. A method for predicting power load according to claim 5, characterized in that: The randomness test in step S22 adopts white noise test.

7. A method for predicting power load according to claim 6, characterized in that: The correction process in step S23 includes setting reasonable parameters or performing differential processing on the data.

8. A method for predicting power load according to claim 7, characterized in that: The step S3 includes using a unit root test and a partial autocorrelation function to obtain the number P of autoregressive terms in the model and the number Q of sliding average terms in the model.

9. A method for predicting power load according to claim 8, characterized in that: The step S3 includes using the AIC function to obtain the number P of autoregressive terms in the model and the number Q of sliding average terms in the model.

10. A method for predicting power load according to claim 9, characterized in that: The step S4 also includes The mean absolute percentage error is used to measure the accuracy of the autoregressive moving average model. Wherein, n represents the number of times the power load data for the N+1th day is predicted within a single day, Ai represents the power load data for the Nth day input for the ith time, and Fi represents the power load data for the N+1th day predicted by the autoregressive moving average model for the ith time; if the mean absolute percentage error exceeds the preset threshold, intervention measures are taken to ensure the accuracy of the power load time series prediction; if it does not exceed the threshold, the status quo is maintained until the prediction is completed.

Images