Short-term power load prediction interval prediction method based on pattern recognition

Abstract

The invention provides a short-term power load prediction interval prediction method based on pattern recognition. The method comprises the steps that historical data including electricity consumption, dates, time and working modes are acquired, a time sequence is processed, and clusters are acquired according to date types and abnormal value detection; searching mode features, average values and standard deviations of the typical data, clustering similar modes, and defining final consumption mode features; and defining a probability and a Z value associated with the prediction interval, and calculating the prediction interval PI after synthesizing the data. The method is based on mode recognition and reasoning statistics, is different from a traditional method for predicting a point value by reducing errors, can predict 96 active power intervals every day and one interval every 15 minutes, and can predict the probability related to an actual value in each interval at the same time. The method can be used for any type of users, trend and seasonal components in the data are eliminated, and the data result better conforms to normal distribution.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power load prediction, in particular, and especially relates to a short-term power load prediction interval prediction method based on pattern recognition. BACKGROUND

[0002] With the increasing popularity of renewable energy and the increase of new major loads such as electric vehicles, the power system is moving towards a more sustainable system. In addition, with the increase of population, the power load is also growing. This leads to additional system complexity and uncertainty, therefore, demand prediction becomes an important tool for power management and is crucial for corresponding decision-making.

[0003] The existing prediction methods at the present stage have the following shortcomings:

[0004] (1) Some methods have limitations in feature extraction and selection, and need to consider various features required for load prediction more comprehensively and design more effective feature extraction methods.

[0005] (2) Some methods cannot provide accurate enough prediction intervals when performing interval prediction, and the confidence of the prediction results is not high enough, which needs to be improved to improve the accuracy and reliability of interval prediction. SUMMARY

[0006] According to the above technical problems, a short-term power load prediction interval prediction method based on pattern recognition is provided. The method is based on pattern recognition and inferential statistics, which is different from the traditional method of predicting point values by reducing errors. The method can predict 96 active power intervals every day, one interval every 15 minutes, and can also predict the probability related to the actual value in each interval. The present application can be used for any type of user, eliminating the trend and seasonal components in the data, making the data results more consistent with the normal distribution.

[0007] The technical means adopted by the present application are as follows:

[0008] A short-term power load prediction interval prediction method based on pattern recognition, comprising:

[0009] S1, obtaining historical data including power consumption, date and time, and working mode, and processing time series, and obtaining clustering according to date type and abnormal value detection;

[0010] S2, finding the pattern characteristics, mean value and standard deviation of typical data, and clustering similar patterns to define the final consumption pattern characteristics;

[0011] S3, defining the probability and Z value associated with the prediction interval, and calculating the prediction interval PI by comprehensively processing the data.

[0012] Further, step S1 specifically comprises:

[0013] S11, obtain daily load curve data, and process the obtained daily load curve data to make the processed data conform to normal probability distribution, and separate out abnormal conditions;

[0014] S12, according to the result obtained in step S11, organize the daily load curve data into a classification matrix with a size of MxA, wherein M represents the number of days in the selected period, and A represents the number of features considered each day;

[0015] S13, by applying the energy consumption massive analysis method, perform data conversion on the time series to eliminate the trend and seasonal components in the data;

[0016] S14, decompose and standardize the classification matrix to make the mean value 0 and the standard deviation 1, standardize each column of the classification matrix, calculate the Z value corresponding to each row r and each column c, and the calculation formula is as follows:

[0017]

[0018] wherein x rc represents the value of variable X in the decomposition matrix of the rth row and the cth column; v c represents the mean value; σ c represents the standard deviation of variable X in the cth column; the maximum value of each column Z is stored in vector Z max ; and the minimum value of each column Z is stored in vector Z min .

[0019] S15, calculate the probability of the Z value corresponding to column c within the confidence interval, and the calculation formula is as follows:

[0020] P(Z α1,c <Z c <Z α2,c )=1-α 1,c -α 2,c =1-(α 1,c +α 2,c )=1-α

[0021] wherein Z α1,c and Z α2,c are the lower limit and the upper limit of the confidence interval of each column c respectively; and α 1,c and α 2,c represent the left tail area and the right tail area of the standard normal distribution respectively.

[0022] Further, step S11 specifically comprises:

[0023] S111, analyze period, collect the number of days of power consumption data, each day consists of 96 quarter measurements, thus forming daily load curve data;

[0024] S112, select a date in a week, define the type of analysis day and whether it is a weekday.

[0025] Further, step S2 specifically includes:

[0026] S21, obtain random consumption patterns for each day in each week through a pattern matrix, and each random consumption pattern is represented by four vectors, each column of each pattern matrix corresponds to a random consumption pattern, and the average value of Z min And the average value of Z max The standard deviation value;

[0027] S22, group similar patterns, and use Euclidean distance as the grouping measure, which is expressed as follows:

[0028]

[0029] Wherein, C and B represent vectors with the same number of elements; c i And b i Indicate the normalized pattern feature value.

[0030] Further, step S3 specifically includes:

[0031] S31, adjust the average value;

[0032] S32, according to the adjusted average value and standard deviation, calculate the prediction interval PI, and the calculation formula is as follows:

[0033]

[0034] Wherein, μ represents the median; y α2 Indicates the Z value left under the normal curve by α / 2; X0 indicates the random variable to be predicted; N indicates the number of data;

[0035] S33, according to the calculated prediction interval PI, the number of DLP in the final pattern matrix is represented, and the upper and lower bounds of the prediction interval PI are represented by the following formulas respectively:

[0036]

[0037]

[0038] Further, step S31 specifically includes:

[0039] S311, subtract a basic load value, which is 1% of all observation values, from the obtained average value;

[0040] S312, multiplying the value obtained in step S311 by the value of the seasonal index corresponding to the last week;

[0041] S313, adding the basic load value to the value obtained from step S312;

[0042] S314, defining the probability (1-alpha) and associating it with the prediction interval PI;

[0043] S315, setting the probability condition for the value satisfying the standard normal distribution as follows:

[0044]

[0045] Compared with the prior art, the present application has the following advantages:

[0046] 1. The short-term power load prediction interval prediction method based on pattern recognition provided by the present application can be used for any type of user, and is different from the prediction point value of the traditional method, while reducing the prediction error.

[0047] 2. The short-term power load prediction interval prediction method based on pattern recognition provided by the present application, through robust statistical analysis, considers the seasonality of the data and limits the uncertainty of the prediction with lower computational complexity.

[0048] 3. The short-term power load prediction interval prediction method based on pattern recognition provided by the present application can automatically adjust the model parameters according to the changes of historical load data, adapt to the load changes of different time periods and seasons, and has certain self-adaptability.

[0049] 4. The short-term power load prediction interval prediction method based on pattern recognition provided by the present application can capture the rules and periodicity in the load data through pattern analysis and learning of historical load data, thereby improving the accuracy of load prediction.

[0050] Based on the above reasons, the present application can be widely popularized in the field of power load prediction. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1 The method flowchart of the present application.

[0053] Figure 2This is a schematic diagram of a data acquisition and processing method provided in an embodiment of the present invention. Detailed Implementation

[0054] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0055] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0056] like Figure 1 As shown, this invention provides a short-term power load forecasting interval prediction method based on pattern recognition, comprising:

[0057] S1. Obtain historical data including electricity consumption, date and time, and working mode, process the time series, and obtain clusters based on date type and outlier detection;

[0058] S2. Identify the pattern characteristics, mean, and standard deviation of typical data, and cluster similar patterns to define the final consumption pattern characteristics;

[0059] S3. Define the probability and Z value associated with the prediction interval, and calculate the prediction interval PI by integrating the data.

[0060] In a specific implementation, as a preferred embodiment of the present invention, step S1 specifically includes:

[0061] S11. Obtain daily load curve (DLP) data and process the obtained daily load curve (DLP) data to make the processed data conform to a normal probability distribution and separate out abnormal situations; such as Figure 2 The diagram shows a data acquisition and processing method. In this embodiment, step S11 specifically includes:

[0062] S111, Analysis cycle: The number of days for which electricity consumption data is collected. Each day consists of 96 quarter-hour measurements, thus forming the daily load curve (DLP) data.

[0063] S112. Select a date in the week, define the type of analysis day and whether it is a workday.

[0064] S12. Based on the results obtained in step S11, the Daily Load Curve (DLP) data is organized into a classification matrix of size M×A, where M represents the number of days in the selected period, and A represents the number of features considered each day. In this embodiment, the features used to represent daily electricity consumption are 96 data points of average electricity consumption measurements per quarter and four variables used for data segmentation: date and time, number of days per week, working hours, and temperature. Therefore, the total number of features is 100, i.e., A = 100.

[0065] S13. Due to changes in factors such as workflow, ambient temperature, and holidays, electricity demand is considered a time series with seasonal variations. This type of data needs to be transformed or processed. Therefore, in this embodiment, the data of the time series is transformed by applying the massive energy consumption analysis (SAEC) method to eliminate the trend and seasonal components in the data, making the data instances more consistent with the normal probability distribution and easier to interpret the results.

[0066] S14. Decompose and standardize the classification matrix to make the mean 0 and the standard deviation 1. Standardize each column of the classification matrix and calculate the Z value corresponding to each row r and each column c. The calculation formula is as follows:

[0067]

[0068] Where, x rc This represents the value of variable X in the decomposition matrix at row r and column c; v c σ represents the average value; c This represents the standard deviation of variable X in column c; the maximum value of each column Z is stored in the vector Z. max In the middle; the minimum value of each column Z is stored in the vector Z. min middle;

[0069] S15. Calculate the probability that the Z value corresponding to column c is within the confidence interval. The calculation formula is as follows:

[0070] P(Z α1,c <Z c <Z α2,c )=1-α 1,c -α 2,c =1-(α) 1,c +α 2,c )=1-α

[0071] Among them, Z α1,c and Z α2,c These are the lower and upper limits of the confidence interval for each column c, respectively; α 1,c and α 2,c These represent the left and right tail regions of the standard normal distribution, respectively.

[0072] In a specific implementation, as a preferred embodiment of the present invention, step S2 specifically includes:

[0073] S21. Obtain the random consumption pattern for each day of each week using the pattern matrix. Represent the random consumption pattern with four vectors, and define the Z-axis for each column of the pattern matrix. min and Z max The values ​​of the mean and standard deviation;

[0074] S22. Grouping similar patterns together yields better results with more data. Euclidean distance is used as the grouping metric, and the equation is as follows:

[0075]

[0076] Where C and B both represent vectors with the same number of elements; c i and b i All of these represent values ​​of the pattern features that have undergone normalization.

[0077] In a specific implementation, as a preferred embodiment of the present invention, step S3 specifically includes:

[0078] S31. Before calculating the prediction interval PI, if there is no data transformation, the average value of the final model will not change. If the massive energy consumption analysis (SAEC) method is used for data transformation, the average value needs to be adjusted. The specific adjustment process is as follows:

[0079] S311. Subtract a basic load value from the obtained average value, which is 1% of all observations;

[0080] S312. Multiply the value obtained in step S311 by the value of the seasonal index corresponding to the most recent week;

[0081] S313. Add the basic load value to the value obtained from step S312;

[0082] S314. Define the probability (1-α) and associate it with the prediction interval PI;

[0083] S315. Set the probability conditions for values ​​that satisfy the standard normal distribution as follows:

[0084]

[0085] S32. Calculate the prediction interval PI based on the adjusted mean and standard deviation. The calculation formula is as follows:

[0086]

[0087] Where μ represents the median; y α2 This represents the Z value left under the normal curve at α / 2; X0 represents the random variable to be predicted; N represents the number of data points.

[0088] S33. Based on the calculated prediction interval PI, the number of DLPs in the final model matrix is ​​represented. The upper and lower bounds of the prediction interval PI are expressed by the following formulas:

[0089]

[0090]

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A short-term power load forecasting interval prediction method based on pattern recognition, characterized in that, include: S1. Obtain historical data including electricity consumption, date and time, and working mode, process the time series, and obtain clusters based on date type and outlier detection; S2. Identify the pattern characteristics, mean, and standard deviation of typical data, and cluster similar patterns to define the final consumption pattern characteristics; S3. Define the probability and Z value associated with the prediction interval, and calculate the prediction interval PI by integrating the data.

2. The short-term power load forecasting interval prediction method based on pattern recognition according to claim 1, characterized in that, Step S1 specifically includes: S11. Obtain daily load curve data and process the obtained daily load curve data to make the processed data conform to a normal probability distribution and separate out abnormal situations. S12. Based on the results obtained in step S11, organize the daily load curve data into a classification matrix of size M×A, where M represents the number of days in the selected period and A represents the number of features considered each day. S13. By applying massive energy consumption analysis methods, the time series data is transformed to eliminate trend and seasonal components in the data; S14. Decompose and standardize the classification matrix to make the mean 0 and the standard deviation 1. Standardize each column of the classification matrix and calculate the Z value corresponding to each row r and each column c. The calculation formula is as follows: Z rm =(x rc -v c ) / σ c Where, x rc This represents the value of variable X in the decomposition matrix at row r and column c; v c σ represents the average value; c This represents the standard deviation of variable X in column c; the maximum value of each column Z is stored in the vector Z. max In the middle; the minimum value of each column Z is stored in the vector Z. min middle; S15. Calculate the probability that the Z value corresponding to column c is within the confidence interval. The calculation formula is as follows: P(Z α1,c <Z c <Z α2,c )=1-(α 1,c +α 2,c )=1-α Among them, Z α1,c and Z α2,c These are the lower and upper limits of the confidence interval for each column c, respectively; α 1,c and α 2,c These represent the left and right tail regions of the standard normal distribution, respectively.

3. The short-term power load forecasting interval prediction method based on pattern recognition according to claim 2, characterized in that, Step S11 specifically includes: S111, Analysis cycle: The number of days for collecting electricity consumption data. Each day consists of 96 quarter-hour measurements, thus forming daily load curve data. S112. Select a date in the week, define the type of analysis day and whether it is a workday.

4. The short-term power load forecasting interval prediction method based on pattern recognition according to claim 1, characterized in that, Step S2 specifically includes: S21. Obtain the random consumption pattern for each day of each week using the pattern matrix. Represent the random consumption pattern with four vectors, and define the Z-axis for each column of the pattern matrix. min and Z max The values ​​of the mean and standard deviation; S22. Group similar patterns together, using Euclidean distance as the grouping metric. The equation is as follows: Where C and B both represent vectors with the same number of elements; c i and b i All of these represent values ​​of the pattern features that have undergone normalization.

5. The short-term power load forecasting interval prediction method based on pattern recognition according to claim 1, characterized in that, Step S3 specifically includes: S31. Adjust the average value; S32. Calculate the prediction interval PI based on the adjusted mean and standard deviation. The calculation formula is as follows: Where μ represents the median; y α / 2 This represents the Z value left under the normal curve at α / 2; X0 represents the random variable to be predicted; N represents the number of data points. S33. Based on the calculated prediction interval PI, the number of DLPs in the final model matrix is ​​represented. The upper and lower bounds of the prediction interval PI are expressed by the following formulas:

6. The short-term power load forecasting interval prediction method based on pattern recognition according to claim 5, characterized in that, Step S31 specifically includes: S311. Subtract a basic load value from the obtained average value, which is 1% of all observations; S312. Multiply the value obtained in step S311 by the value of the seasonal index corresponding to the most recent week; S313. Add the basic load value to the value obtained from step S312; S314. Define the probability (1-α) and associate it with the prediction interval PI; S315. Set the probability conditions for values ​​that satisfy the standard normal distribution as follows:

Images