Flexible load prediction method based on optimized grey correlation model

Abstract

The invention relates to a flexible load prediction method based on an optimized grey correlation model, which belongs to the technical field of optimized operation of a power system and comprises the following steps of: checking and correcting defects of load data; modeling a grey correlation model; correcting a grey correlation model parameter alpha; verifying the model precision according to the deviation; and carrying out multi-load prediction based on associated parameters. According to the method, the model parameter alpha of the grey correlation model is optimized, the grade ratio sigma (0) (k) falls into a feasible region, the outside of the grey correlation model is optimized, the concept of the correlation degree of a prediction curve and an actual curve is put forward, a multi-load classification prediction scheme is put forward, periodic changes and multi-factor influences of different loads are solved, and the prediction precision is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of power system optimization operation, and in particular relates to a flexible load prediction method based on an optimized grey correlation model. Background Art

[0002] In order to adhere to the principle of sustainable development, my country has set up "dual carbon" goals. Accelerating the clean and low-carbon transformation of the power system is an important way to achieve the "dual carbon" goals. On the one hand, it promotes energy conservation and emission reduction of power industry loads, reduces carbon emissions through electric-to-heat heating, and increases the terminal utilization of electric energy; on the other hand, it vigorously develops new energy, increases the installed capacity of new energy grid-connected, and reduces the power generation of traditional thermal power units. However, the uncertainty of new energy output and power load has brought severe challenges to the safe, stable and economic operation of the new power system. Therefore, the study of the balance and stability of the power system based on controllable load regulation has important theoretical significance and application value for improving the flexibility of the power system and the absorption of new energy.

[0003] Load forecasting is generally divided into long-term, medium-term, short-term and ultra-short-term forecasts. Commonly used short-term load forecasting methods, such as regression analysis and time series algorithms, are affected by user behavior and environmental factors and have high uncertainty. The sudden changes in the electric power curve caused by time-of-use electricity prices cannot obtain accurate forecast results by only considering weather factors. Therefore, the accuracy of these short-term forecasting techniques may be greatly reduced.

[0004] At the same time, for demand response prediction methods, the error of optimal operation prediction is usually represented by a random variable in some probability distribution. However, for some special types of loads, the accurate statistical characteristics of such random variables cannot be provided.

[0005] In addition, oversimplification of the uncertainty on the demand side may result in failure to reflect the actual operating status of the distributed heating network. The prediction results will be affected by the sequence characteristics, time-of-use pricing and climate-related effects on the distributed temperature control load. Therefore, for the short-term prediction of adjustable loads, improving the prediction accuracy is still a technical problem that this field has been eager to solve. Summary of the invention

[0006] In view of the deficiencies of the prior art, the present invention provides a flexible load forecasting method based on an optimized grey correlation model, which solves the problem that the load forecasting accuracy in the prior art is not ideal.

[0007] To achieve the above objectives, the present invention provides a flexible load prediction method based on an optimized grey relational model, comprising the following steps:

[0008] Step 1: Check and correct defects in load data;

[0009] Step 2: Grey relational model building;

[0010] Step 3, correction of grey relational model parameter α;

[0011] Step 4: Verify the model accuracy based on the deviation;

[0012] Step 5: Multi-load forecasting based on associated parameters;

[0013] Step 6: Output the results and end.

[0014] Preferably, in step 1, the defect inspection and correction of load data includes the following steps:

[0015] (1) Data loss processing to determine the daily 96-point load,

[0016] |l(i)-l(i-1)|<λ i=2,3,·,, λ is a constant (9)

[0017] If the loads between two adjacent time points all fall within the λ threshold, they will be directly eliminated; if there are data points falling within the λ threshold for more than 2 hours in a row, they will be processed according to the following principles:

[0018] For holidays, find the corresponding holiday load of the previous year as the benchmark load, find the closest ordinary working day load before the holiday, and use the difference as the annual growth rate as shown in formula 10;

[0019] L(y,d)=L(y-1,d)+Δ (10)

[0020] For ordinary days, find N-1 normal load data sources before the abnormal data day, and mark the data source of the day to be corrected as N (N is a constant); if the load data of k normal points other than 96-k points at the time to be supplemented are known, then:

[0021]

[0022] Where P R (i, t) is the load data of the i-th data source at time t among the N-1 data sources; P AVE (t) is the arithmetic mean of the load data of N-1 data sources at time t; is the arithmetic mean of the load data of k normal points on the day to be corrected; is the arithmetic mean value of the load data of N-1 data sources corresponding to the k normal points on the day to be corrected; P R (N,t) is the load data at time t to be supplemented;

[0023] (2) Data jump processing,

[0024]

[0025] If the load value between two points satisfies one of the above equations, it is determined that there is a sudden load change; for the head load, it is corrected by the level ratio of the right neighbor; for the end load, it is corrected by the level ratio of the left neighbor; for the middle section, the weighted average of the front and rear loads is taken, and the expression is:

[0026]

[0027] Preferably, in step 2, the GM (1,1) model building includes the following steps:

[0028] The original sequence x (0) Generate a first-order sequence using 1-AGO accumulation;

[0029] x (1) =[x (1) (1),x (1) (2),…,x (1) (n)] (1)

[0030] in:

[0031]

[0032] x (1) The sequence satisfies the first-order differential equation model;

[0033]

[0034] Using the least squares method, we can solve the parameters a and u and get:

[0035]

[0036] in:

[0037] The desired Substituting back into the differential equation (3), we have

[0038]

[0039] Solving equation (6) yields equation (7):

[0040]

[0041] Formula (7) is called the time response function model of the GM (1,1) model, which is the specific calculation formula of the grey prediction of the GM (1,1) model. The above formula is then reduced by 1-IAGO to obtain the original series x (0) The grey prediction model is

[0042]

[0043] Preferably, in step 3, the α parameter of the GM (1,1) model is modified by the following steps:

[0044] Take α = 0.5, and use the least square method to solve the parameters α and u according to formula (3);

[0045] The parameters obtained will be Substituting into formula (19),

[0046]

[0047] Recalculate Recorded as Will and Compare: Given an arbitrarily small positive integer ε, if Enter formula (1) and perform the calculation again; when When , the iteration ends and turns to formula (3);

[0048] Establish the prediction model of GM(1,1), see formula (7); Perform cumulative reduction and get x (0) The predicted value of (k) is shown in formula (8).

[0049] Preferably, in step 4, verifying the model accuracy according to the deviation includes:

[0050] For a given sequence x (0) , can we establish a GM(1,1) with higher accuracy? (0) The level ratio σ (0) (k) is determined by its size and the interval it belongs to; for any x (0) (k)∈x (0) , k = 1, 2, 3, ..., n, let σ (0) (k) is x (0) The level ratio is:

[0051]

[0052] Level ratio σ (0) (k) must fall within the feasible region ItG = (0.1353, 7.839), and if we want to establish a high-precision GM (1, 1) model, the level ratio σ (0) (k) should fall in the subinterval ItGM = (1-ε, 1+ε) close to 1, where ε is a specified sufficiently small real number. In fact, if there is an original sequence x (0) , and let σ (0) (k) is x (0) The level ratio is

[0053] The criterion for feasibility analysis of M(1,1) modeling is: the modeling sequence x (0) The level ratio σ (0) (k) If:

[0054]

[0055] Then we think x (0) It can be modeled as GM(1,1);

[0056] For the sequence whose level ratio test does not conform to formula (15), data transformation must be performed so that the level ratio of the transformed sequence falls within the acceptable range. The level ratio of the processed sequence σ y (k) should be as close to 1 as possible, that is, the level ratio deviation σ y (k) should be as small as possible.

[0057] Preferably, in step 5, the multi-load forecasting based on associated parameters includes:

[0058] The basic idea of ​​correlation is to judge the similarity between curves. The larger the correlation, the better the corresponding prediction model. Given n prediction value curves, denoted as

[0059] x i ={x i (1),x i (2),…,x i (n)|i=1,2,…,n} (20)

[0060] The actual value curve is x0, then each x i The correlation coefficient with x0 at point k is

[0061]

[0062] In the formula, |x0(k)-x i (k)| is the actual value x0(k) and the predicted value x at the kth point i (k) absolute difference; min i min k |x0(k)-x i (k)| is the minimum difference between the two levels, indicating that in each prediction curve x i Based on finding the minimum difference in (k), we can find all the curves x one by one according to i = 1, 2, ..., m. i (k) minimum difference; max i max k |x0(k)-x i (k)| is the maximum difference of the second level, which has a similar meaning to the minimum difference; ρ is the resolution coefficient, which ranges from 0 to 1 and is generally taken as 0.5;

[0063] By combining the correlation coefficients of each point, the prediction curve x can be obtained. i The correlation degree between (k) and the actual curve x0(k) is.

[0064]

[0065] Preferably, in step 5, the loads are divided into three types: air conditioning Electric Vehicles illumination The correlation coefficients and correlation degrees of these three loads are calculated separately to improve the prediction accuracy.

[0066] The present invention provides a flexible load prediction method based on an optimized grey correlation model, which has the following beneficial effects:

[0067] (1) Optimize the model parameter α of the grey relational model so that the level ratio σ (0) (k) falls within the feasible domain, which can improve the prediction accuracy of the model.

[0068] (2) External optimization of the grey correlation model and the concept of correlation between the predicted curve and the actual curve can improve the prediction accuracy.

[0069] (3) A multi-load classification forecasting scheme is proposed to solve the periodic changes and multi-factor influences of different types of loads and improve the forecasting accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] Example

[0073] The present invention provides a flexible load prediction method based on an optimized grey correlation model, comprising the following steps:

[0074] Step 1: Check and correct defects in load data;

[0075] The load data anomalies mainly include: ① data loss, which appears as a straight line segment on the graph; ② data jump, which appears as peaks and glitches on the graph.

[0076] (1) Data loss processing to determine the daily 96-point load,

[0077] |l(i)-l(i-1)|<λ i=2,3,·,λ is a constant (9)

[0078] If the load l between two adjacent time points all falls within the λ threshold, it will be directly eliminated; if there are data points falling within the λ threshold for more than 2 hours in a row, it will be processed according to the following principles:

[0079] For holidays, find the corresponding holiday load of the previous year as the benchmark load, find the closest ordinary working day load before the holiday, and use the difference as the annual growth rate as shown in formula 10;

[0080] L(y,d)=L(y-1,d)+Δ (10)

[0081] For ordinary days, find N-1 normal load data sources before the abnormal data day, and mark the data source of the day to be corrected as N (N is a constant); if the load data of k normal points other than 96-k points at the time to be supplemented are known, then:

[0082]

[0083] Where P R (i, t) is the load data of the i-th data source at time t among the N-1 data sources; P AVE (t) is the arithmetic mean of the load data of N-1 data sources at time t; is the arithmetic mean of the load data of k normal points on the day to be corrected; is the arithmetic mean value of the load data of N-1 data sources corresponding to the k normal points on the day to be corrected; P R (N,t) is the load data at time t to be supplemented;

[0084] (2) Data jump processing,

[0085]

[0086] If the load value between two points satisfies one of the above equations, it is determined that there is a sudden load change; for the head load, it is corrected by the level ratio of the right neighbor; for the end load, it is corrected by the level ratio of the left neighbor; for the middle section, the weighted average of the front and rear loads is taken, and the expression is:

[0087]

[0088] Step 2: Grey relational model building;

[0089] The original sequence x (0) Generate a first-order sequence using 1-AGO accumulation;

[0090] x (1) =[x(1) (1),x (1) (2),…,x (1) (n)] (1)

[0091] in:

[0092]

[0093] x (1) The sequence satisfies the first-order differential equation model;

[0094]

[0095] Using the least squares method, we can solve the parameters a and u and get:

[0096]

[0097] in:

[0098] The desired Substituting back into the differential equation (3), we have

[0099]

[0100] Solving equation (6) yields equation (7):

[0101]

[0102] Formula (7) is called the time response function model of the GM (1,1) model, which is the specific calculation formula of the grey prediction of the GM (1,1) model. The above formula is then reduced by 1-IAGO to obtain the original series x (0) The grey prediction model is

[0103]

[0104] Step 3, correction of grey relational model parameter α;

[0105] Take α = 0.5, and use the least square method to solve the parameters α and u according to formula (3);

[0106] The parameters obtained will be Substituting into formula (19),

[0107]

[0108] Recalculate Recorded as Will and Compare: Given an arbitrarily small positive integer ε, if Enter formula (1) and perform the calculation again; when When , the iteration ends and turns to formula (3);

[0109] Establish the prediction model of GM(1,1), see formula (7); Perform cumulative reduction and get x (0) The predicted value of (k) is shown in formula (8).

[0110] Step 4: Verify the model accuracy based on the deviation;

[0111] For a given sequence x (0) , can we establish a GM(1,1) with higher accuracy? (0) The level ratio σ (0) (k) is determined by its size and the interval it belongs to. (0) (k)∈x (0) , k = 1, 2, 3, ..., n, let σ (0) (k) is x (0) The level ratio is:

[0112]

[0113] Level ratio σ (0) (k) must fall within the feasible region ItG = (0.1353, 7.839), and if we want to establish a high-precision GM (1, 1) model, the level ratio σ (0) (k) should fall in the subinterval ItGM = (1-ε, 1+ε) close to 1, where ε is a specified sufficiently small real number. In fact, if there is an original sequence x (0) , and let σ (0) (k) is x (0) The criterion for feasibility analysis of M(1,1) modeling is: (0) The level ratio σ (0) (k) If:

[0114]

[0115] Then we think x (0) It can be modeled as GM(1,1).

[0116] For the sequence whose level ratio test does not conform to formula (15), data transformation must be performed so that the level ratio of the transformed sequence falls within the acceptable range. y (k) should be as close to 1 as possible, that is, the level ratio deviation σ y (k) should be as small as possible. The principle of data processing is to minimize σ y (k) Due to

[0117]

[0118] When the data series develops at a fast rate (i.e., the |a| value is large), the prediction accuracy of the GM(1,1) model is poor. The reason is that in the current traditional calculation method of GM(1,1) modeling, the background value is generated by

[0119]

[0120] It can be seen that this calculation method simply takes x (1) (k) and x (1) The average value of (k+1), that is, α=0.5, does not take the size of a into consideration.

[0121] Background value z (1) The exact calculation formula for (k) should be

[0122] z (1) (k+1)=αx (1) (k)+(1-α)x (1) (k+1) (18)

[0123] a and α have the following relationship

[0124]

[0125] Using L'Hôpital's rule, it can be proved that when a→0, the limit value of α is 0.5. The corresponding relationship between a and α can be calculated from the above formula, see Table 1.

[0126] It can be seen that when |a| is small, α is very close to 0.5; when |a| is large, α deviates greatly from 0.5. Therefore, different α can be selected for background value z according to different a values. (1) Calculation of (k).

[0127] Step 5: Multi-load forecasting based on associated parameters;

[0128] The basic idea of ​​correlation is to judge the similarity between curves. The larger the correlation, the better the corresponding prediction model. Given n prediction value curves, denoted as

[0129] x i ={x i (1),x i (2),…,x i (n)|i=1,2,…,n} (20)

[0130] The actual value curve is x0, then each x i The correlation coefficient with x0 at point k is

[0131]

[0132] In the formula, |x0(k)-xi (k)| is the actual value x0(k) and the predicted value x at the kth point i (k) absolute difference; min i min k |x0(k)-x i (k)| is the minimum difference between the two levels, indicating that in each prediction curve x i Based on finding the minimum difference in (k), we can find all the curves x one by one according to i = 1, 2, ..., m. i (k) minimum difference; max i max k |x0(k)-x i (k)| is the maximum difference of the second level, which has a similar meaning to the minimum difference; ρ is the resolution coefficient, which ranges from 0 to 1 and is generally taken as 0.5;

[0133] By combining the correlation coefficients of each point, the prediction curve x can be obtained. i The correlation degree between (k) and the actual curve x0(k) is.

[0134]

[0135] The present invention divides loads into three types: air conditioning Electric Vehicles illumination The correlation coefficients and correlation degrees of these three loads are calculated separately to improve the prediction accuracy.

[0136] Step 6: Output the results and end.

[0137] Table 1 Relationship between a and α

[0138]

[0139] Engineering example verification: The original sequences of three different types of loads were selected from the power load of a 10kV node in a city in Liaoning Province from April to May 2022, and the ordinary GM (1,1) model and the GM (1,1) prediction model optimized by the present invention were constructed. These models were used for prediction respectively, and the error statistics are shown in Table 2.

[0140] Table 2

[0141]

[0142]

[0143] The results show that the average error of the prediction method based on the optimized grey relational model is approximately in the range of 1% to 3%, which is significantly better than the ordinary GM (1,1) prediction model.

[0144] It should be noted that, in this article, relational terms such as first and second 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 those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article, or device. In the absence of further restrictions, the elements defined by the sentence "including a reference structure" do not exclude the existence of other identical elements in the process, method, article, or device including the elements.

[0145] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the invention, the scope of the invention being defined by the appended claims and their equivalents.

Claims

1. A flexible load prediction method based on optimized grey relational model, characterized in that: The following steps are involved: Step 1: Check and correct defects in load data; Step 2: Grey relational model building; Step 3, correction of grey relational model parameter α; Step 4: Verify the model accuracy based on the deviation; Step 5: Multi-load forecasting based on associated parameters; Step 6: Output the results and end.

2. The flexible load prediction method based on the optimized grey relational model according to claim 1 is characterized in that: In step 1, the defect inspection and correction of load data includes the following steps: (1) Data loss processing to determine the daily 96-point load, |l(i)-l(i-1)|<λ i=2,3,,,λ is a constant (9) If the loads between two adjacent time points all fall within the λ threshold, they will be directly eliminated; if there are data points falling within the λ threshold for more than 2 hours in a row, they will be processed according to the following principles: For holidays, find the corresponding holiday load of the previous year as the benchmark load, find the closest ordinary working day load before the holiday, and use the difference as the annual growth rate as shown in formula 10; L(y,d)=L(y-1,d)+Δ (10) For ordinary days, find N-1 normal load data sources before the abnormal data day, and mark the data source of the day to be corrected as N (N is a constant); if the load data of k normal points other than 96-k points at the time to be supplemented are known, then: Where P R (i, t) is the load data of the i-th data source at time t among the N-1 data sources; P AVE (t) is the arithmetic mean of the load data of N-1 data sources at time t; is the arithmetic mean of the load data of k normal points on the day to be corrected; is the arithmetic mean value of the load data of N-1 data sources corresponding to the k normal points on the day to be corrected; P R (N,t) is the load data at time t to be supplemented; (2) Data jump processing, If the load value between two points satisfies one of the above equations, it is determined that there is a sudden load change; for the head load, it is corrected by the level ratio of the right neighbor; for the end load, it is corrected by the level ratio of the left neighbor; for the middle section, the weighted average of the front and rear loads is taken, and the expression is:

3. The flexible load prediction method based on optimized grey relational model according to claim 1 is characterized in that: In the step 2, the grey relational model building includes the following steps: The original sequence x (0) Generate a first-order sequence using 1-AGO accumulation; x (1) =[x (1) (1),x (1) (2),…,x (1) (n)] (1) in: x (1) The sequence satisfies the first-order differential equation model; Using the least squares method, we can solve the parameters a and u and get: in: The desired Substituting back into the differential equation (3), we have Solving equation (6) yields equation (7): Formula (7) is called the time response function model of the GM (1,1) model, which is the specific calculation formula of the grey prediction of the GM (1,1) model. The above formula is then reduced by 1-IAGO to obtain the original series x (0) The grey prediction model is 4. The flexible load prediction method based on the optimized grey relational model according to claim 3 is characterized in that: In step 3, the α parameter of the GM (1,1) model is modified as follows: Take α = 0.5, and use the least square method to solve the parameters α and u according to formula (3); The parameters obtained will be Substituting into formula (19), Recalculate Recorded as Will and Compare: Given an arbitrarily small positive integer ε, if Enter formula (1) and perform the calculation again; when When , the iteration ends and turns to formula (3); Establish the prediction model of GM(1,1), see formula (7); Perform cumulative reduction and get x (0) The predicted value of (k) is shown in formula (8).

5. The flexible load prediction method based on optimized grey relational model according to claim 1 is characterized in that: In step 4, checking the model accuracy according to the deviation includes: For a given sequence x (0) , can we establish a GM(1,1) with higher accuracy? (0) The level ratio σ (0) (k) is determined by its size and the interval it belongs to; for any x (0) (k)∈x (0) , k = 1, 2, 3, ..., n, let σ (0) (k) is x (0) The level ratio is: Level ratio σ (0) (k) must fall within the feasible region ItG = (0.1353, 7.839), and if we want to establish a high-precision GM (1, 1) model, the level ratio σ (0) (k) should fall in the subinterval ItGM = (1-ε, 1+ε) close to 1, where ε is a specified sufficiently small real number. In fact, if there is an original sequence x (0) , and let σ (0) (k) is x (0) The level ratio is The criterion for feasibility analysis of M(1,1) modeling is: the modeling sequence x (0) The level ratio σ (0) (k) If: Then we think x (0) It can be modeled as GM(1,1); For the sequence whose level ratio test does not conform to formula (15), data transformation must be performed so that the level ratio of the transformed sequence falls within the acceptable range. The level ratio of the processed sequence σ y (k) should be as close to 1 as possible, that is, the level ratio deviation σ y (k) should be as small as possible.

6. The flexible load prediction method based on optimized grey relational model according to claim 1 is characterized in that: In step 5, the multi-load forecasting based on the associated parameters includes: The basic idea of ​​correlation is to judge the similarity between curves. The larger the correlation, the better the corresponding prediction model. Given n prediction value curves, denoted as x i ={x i (1),x i (2),…,x i (n)|i=1,2,…,n} (20) The actual value curve is x0, then each x i The correlation coefficient with x0 at point k is In the formula, |x0(k)-x i (k)| is the actual value x0(k) and the predicted value x at the kth point i (k) absolute difference; min i min k |x0(k)-x i (k)| is the minimum difference between the two levels, indicating that in each prediction curve x i Based on finding the minimum difference in (k), we can find all the curves x one by one according to i = 1, 2, ..., m. i (k) minimum difference; max i max k |x0(k)-x i (k)| is the maximum difference of the second level, which has a similar meaning to the minimum difference; ρ is the resolution coefficient, which ranges from 0 to 1 and is generally taken as 0.5; By combining the correlation coefficients of each point, the prediction curve x can be obtained. i The correlation degree between (k) and the actual curve x0(k) is.

7. The flexible load prediction method based on optimized grey relational model according to claim 6 is characterized in that: In step 5, the loads are divided into three types: air conditioning Electric Vehicles illumination The correlation coefficients and correlation degrees of these three loads are calculated separately to improve the prediction accuracy.

Images