An improved photovoltaic decomposition method based on capacity estimation

Abstract

The present invention discloses an improved photovoltaic decomposition method based on capacity estimation, comprising the following steps: generating a candidate capacity sequence; obtaining a capacity curve based on the candidate capacity sequence; performing capacity estimation based on the capacity curve; constructing a probability density function; generating a typical sample; performing maximum likelihood estimation to obtain a preliminary result of photovoltaic decomposition; obtaining a confidence coefficient; and correcting the preliminary result using the confidence coefficient to obtain a final photovoltaic decomposition result. The present invention corrects the original result based on the capacity estimation result, thereby enhancing the robustness of the original photovoltaic decomposition algorithm. The method of the present invention has a high capacity estimation accuracy and maintains good adaptability in the presence of a small number of missing values. At the same time, it improves the photovoltaic decomposition algorithm, increases the calculation speed of the photovoltaic decomposition algorithm, and improves the accuracy of the photovoltaic decomposition algorithm.

Description

Technical Field

[0001] The present invention belongs to the field of photovoltaic decomposition based on post-meter net load data, and in particular relates to an improved photovoltaic decomposition method based on capacity estimation. Background Art

[0002] To achieve the goals of "carbon neutrality and carbon peak," an increasing number of distributed photovoltaic systems are being integrated into the power system, with a significant number installed on building rooftops. These rooftop distributed photovoltaic systems have small and unstable capacity, and their metering data is presented as net load behind the meter. This means that it is impossible to deploy separate meters to measure the photovoltaic output power. However, knowledge of the photovoltaic output power curve is crucial for the operation and planning of distribution systems. Therefore, a decoupling technology is required to decompose the photovoltaic output power curve from the net load curve. This is called photovoltaic decomposition.

[0003] However, existing photovoltaic decomposition methods cannot effectively solve the problem of photovoltaic decomposition due to their own limitations. The main research limitations are as follows:

[0004] Problems with reliance on the physical model and geographic location information of the photovoltaic array. Some work uses the physical model of the photovoltaic array to estimate the photovoltaic output power curve. By analyzing the impact of external meteorological information such as irradiance, temperature, wind direction and air humidity on the photovoltaic output power, a unified physical model of the photovoltaic array is constructed to decompose the photovoltaic output power curve from the net load power. However, it is extremely difficult to accurately obtain the detailed geographic information of the rooftop distributed photovoltaics under each distribution system, because this involves user privacy. There are also problems such as some distributed photovoltaics are not installed under the authorization of the power grid company, which hinder the acquisition of geographic information; on the other hand, considering the different photovoltaic panel models, conversion efficiency, etc. of different users

[0005] Differences, it is difficult for physical models to adaptively adjust to individual differences.

[0006] Some work has used high-resolution color satellite imagery to automatically identify the location and size of small photovoltaic arrays. This work is effective for identifying photovoltaic arrays over large areas. However, it can only estimate the physical size of the photovoltaic array, but not the capacity, because the capacity of a photovoltaic array of the same size may vary depending on the type of photovoltaic panel and its corresponding operating conditions.

[0007] Some data-driven methods use the differences between net load curves on days with similar weather conditions to estimate PV system parameters. This approach partially addresses the reliance on physical models and privacy information, but it fails to account for differences in user load consumption on similar days. Furthermore, this approach requires a longer period of net load data as a basis for clustering similar weather patterns.

[0008] Therefore, in order to solve the above problems, a capacity estimation method and photovoltaic decomposition strategy driven by short-term net load data are urgently needed. Summary of the Invention

[0009] The present invention is proposed to solve the problems existing in the prior art, and its purpose is to provide an improved photovoltaic decomposition method based on capacity estimation.

[0010] The technical solution of the present invention is: an improved photovoltaic decomposition method based on capacity estimation, comprising the following steps:

[0011] A. Generate candidate capacity sequence;

[0012] B. Obtaining a capacity curve based on the candidate capacity sequence;

[0013] C. Estimating capacity based on capacity curves;

[0014] D. Construct probability density function;

[0015] E. Generate typical samples;

[0016] F. Perform maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition;

[0017] G. Get the confidence coefficient;

[0018] H. Use the confidence coefficient to correct the preliminary results and obtain the final photovoltaic decomposition results.

[0019] Furthermore, step A generates a candidate capacity sequence. The specific process is as follows:

[0020] First, take the net load power data as input;

[0021] Then, the net load nighttime power extreme value and the net load daytime power extreme value are counted and recorded on a monthly basis;

[0022] Finally, the candidate capacity sequence is formed by the additive combination of the net load nighttime power extreme value and the net load daytime power extreme value.

[0023] Furthermore, step B obtains a capacity curve based on the candidate capacity sequence. The specific process is as follows:

[0024] First, sort the obtained candidate capacity sequences;

[0025] Then, a capacity curve is established based on the sequence number and capacity.

[0026] Furthermore, step C performs capacity estimation based on the capacity curve. The specific process is as follows:

[0027] First, based on the capacity curve, the changing trend of the capacity curve is obtained;

[0028] Then, the capacity is estimated using the changing trend of the capacity curve.

[0029] Furthermore, step D constructs the probability density function, and the specific process is as follows:

[0030] First, obtain data information of some significant users;

[0031] Then, the monthly nighttime and daytime electricity consumption of the consumer load is extracted based on the data information;

[0032] Finally, a probability density function is constructed based on the monthly nighttime and daytime electricity consumption.

[0033] Furthermore, step E generates typical samples. The specific process is as follows:

[0034] Photovoltaic output power curves are clustered to form typical samples.

[0035] Furthermore, step F performs maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition. The specific process is as follows:

[0036] First, linear combination is solved for typical samples;

[0037] Then, after obtaining the optimal weights, the preliminary results of PV decomposition are obtained.

[0038] Furthermore, step G obtains the confidence coefficient, and the specific process is as follows:

[0039] First, based on the data information of some observable users, the real value of capacity is obtained;

[0040] Then, we obtain the capacity estimate of the considerable users;

[0041] Finally, the confidence coefficient is obtained by comparing the estimated capacity with the true value of the capacity.

[0042] Furthermore, step H uses the confidence coefficient to correct the preliminary results to obtain the final photovoltaic decomposition results. The specific process is as follows:

[0043] First, the capacity values ​​in the preliminary results are obtained;

[0044] Then, the capacity estimate obtained using the capacity estimation method;

[0045] Then, the capacity estimate and the capacity value are processed to obtain a threshold value;

[0046] Then, the capacity estimation is compared with the threshold to determine whether it is necessary;

[0047] Finally, the confidence coefficient is used to correct the part that needs correction.

[0048] The beneficial effects of the present invention are as follows:

[0049] The present invention utilizes the intermittent characteristics of photovoltaic power generation and the stable power characteristics of user consumption load during unused periods to achieve rapid and accurate estimation of photovoltaic capacity under net load.

[0050] The present invention forms typical samples by clustering the photovoltaic output power curves of observable users, constructs a probability density function by utilizing the correlation between nighttime and daytime shown by the consumption load, and obtains the optimal weight by solving the optimization problem, thereby achieving a preliminary estimation of the photovoltaic output power curve.

[0051] This method corrects the original capacity estimation results based on the original data, enhancing the robustness of the original photovoltaic decomposition algorithm. This method achieves high capacity estimation accuracy and maintains good adaptability even with a small number of missing values. It also improves the photovoltaic decomposition algorithm, increasing its computational speed and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the architecture of the present invention;

[0053] Figure 2 It is a capacity estimation curve diagram of the present invention. DETAILED DESCRIPTION

[0054] Hereinafter, the present invention will be described in detail with reference to the accompanying drawings and embodiments:

[0055] like Figures 1 to 2 As shown, an improved photovoltaic decomposition method based on capacity estimation includes the following steps:

[0056] A. Generate candidate capacity sequence;

[0057] B. Obtaining a capacity curve based on the candidate capacity sequence;

[0058] C. Estimating capacity based on capacity curves;

[0059] D. Construct probability density function;

[0060] E. Generate typical samples;

[0061] F. Perform maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition;

[0062] G. Get the confidence coefficient;

[0063] H. Use the confidence coefficient to correct the preliminary results and obtain the final photovoltaic decomposition results.

[0064] Step A generates a candidate capacity sequence. The specific process is as follows:

[0065] First, take the net load power data as input;

[0066] Then, the net load nighttime power extreme value and the net load daytime power extreme value are counted and recorded on a monthly basis;

[0067] Finally, the candidate capacity sequence is formed by the additive combination of the net load nighttime power extreme value and the net load daytime power extreme value.

[0068] Step B obtains the capacity curve based on the candidate capacity sequence. The specific process is as follows:

[0069] First, sort the obtained candidate capacity sequences;

[0070] Then, a capacity curve is established based on the sequence number and capacity.

[0071] Step C estimates the capacity based on the capacity curve. The specific process is as follows:

[0072] First, based on the capacity curve, the changing trend of the capacity curve is obtained;

[0073] Then, the capacity is estimated using the changing trend of the capacity curve.

[0074] Step D constructs the probability density function. The specific process is as follows:

[0075] First, obtain data information of some significant users;

[0076] Then, the monthly nighttime and daytime electricity consumption of the consumer load is extracted based on the data information;

[0077] Finally, a probability density function is constructed based on the monthly nighttime and daytime electricity consumption.

[0078] Step E generates typical samples. The specific process is as follows:

[0079] Photovoltaic output power curves are clustered to form typical samples.

[0080] Step F performs maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition. The specific process is as follows:

[0081] First, linear combination is solved for typical samples;

[0082] Then, after obtaining the optimal weights, the preliminary results of PV decomposition are obtained.

[0083] Step G obtains the confidence coefficient. The specific process is as follows:

[0084] First, based on the data information of some observable users, the real value of capacity is obtained;

[0085] Then, we obtain the capacity estimate of the considerable users;

[0086] Finally, the confidence coefficient is obtained by comparing the estimated capacity with the true value of the capacity.

[0087] Step H uses the confidence coefficient to correct the preliminary results to obtain the final photovoltaic decomposition results. The specific process is as follows:

[0088] First, the capacity values ​​in the preliminary results are obtained;

[0089] Then, the capacity estimate obtained using the capacity estimation method;

[0090] Then, the capacity estimate and the capacity value are processed to obtain a threshold value;

[0091] Then, the capacity estimation is compared with the threshold to determine whether it is necessary;

[0092] Finally, the confidence coefficient is used to correct the part that needs correction.

[0093] Specifically, step A generates a candidate capacity sequence; step B obtains a capacity curve based on the candidate capacity sequence; step C performs capacity estimation based on the capacity curve and integrates a capacity estimation module.

[0094] Specifically, step D constructs a probability density function; step E generates typical samples; and step F performs maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition and integrates a photovoltaic decomposition module.

[0095] Specifically, step G obtains the confidence coefficient; step H uses the confidence coefficient to correct the preliminary result to obtain the final photovoltaic decomposition result and integrate the capacity correction.

[0096] Specifically, the capacity estimation module performs capacity estimation based on a change trend of the capacity estimation curve.

[0097] Specifically, the photovoltaic decomposition module mainly utilizes the source-load characteristics of observable users to perform photovoltaic decomposition on the net load data of unknown users.

[0098] Yet another embodiment

[0099] A. Generate candidate capacity sequence

[0100] For the net load data measured by the IoT meter, its value is equal to the user consumption load minus the photovoltaic output power, as shown in formula (1).

[0101] Considering that the user consumption load is zero, the absolute value of the net load is the photovoltaic output power. The peak value of the absolute value of the net load at this time is equal to the peak value of the photovoltaic output power, that is, the photovoltaic capacity.

[0102] In reality, there is consumer load, which makes the absolute value of the minimum net load smaller than the actual photovoltaic capacity.

[0103] Theoretically, when the net load P reaches its minimum value during the day, the consumer load should be as small as possible and the photovoltaic output power should be as large as possible. Therefore, if we can accurately estimate this smaller consumer load value, the photovoltaic capacity value can be calculated using formula (1).

[0104] P=LG (1)

[0105] In the formula, P represents net load, L represents user consumption load, and G represents photovoltaic output power.

[0106] Specifically, the present invention provides the following specific solution to obtain a candidate capacity sequence:

[0107] First, the net load data of each day in a month is divided into daytime and nighttime.

[0108] Then, the absolute values ​​of the minimum power value at night and the minimum power value during the day are obtained, as shown in formula (2).

[0109]

[0110] Where D represents the number of days, T d 、T n represent the daytime time point of the dth day and the nighttime time point of the nth day, respectively. h (t) represents the net load power value at hour t.

[0111] P day,p (d) refers to all the minimum values ​​of net load during the day, representing the photovoltaic power generation under the conditions of minimum consumer load interference or optimal meteorological conditions.

[0112] P night,p (n) is an estimate of the smaller consumer load value, representing the constant power consumption of electrical appliances when no one is using them at night.

[0113] The additive combination of the two constitutes the candidate sequence set of capacity, as shown in formula (3).

[0114] C candi (i)={P day,p (d)+P night,p (n)|d∈D,n∈D},i=1,..,D 2 (3)

[0115] B. Obtaining a capacity curve based on the candidate capacity sequence

[0116] By using the generated candidate capacity sequence and counting the frequency of statistical data through appropriate methods, a rough estimation of the capacity can be achieved.

[0117] However, this estimation method will theoretically be smaller than the actual capacity value. This is because when only the photovoltaic effect is considered, the photovoltaic capacity value represents the peak value of the photovoltaic output power curve, which will be slightly larger than the point with the highest frequency in the extreme value of the power curve.

[0118] The present invention provides a method to achieve capacity estimation to solve this problem:

[0119] Sort the candidate capacity sequences and draw the corresponding capacity estimation curve, such as Figure 2 As shown by the solid line in the middle, the capacity value is estimated using the changing trend of the curve. The steps are as follows:

[0120] a. Mark the maximum point (i.e. the last point) of the capacity estimation curve as the end point.

[0121] b. Find the first point whose value is greater than the absolute value of the minimum net load value (as shown in formula (4)) as the starting point,

[0122]

[0123] c. The capacity value sequence between the starting point and the ending point is used as the capacity sequence to be determined.

[0124] d. Draw a straight line connecting the starting point and the end point, and find the tangent line of the curve of the capacity sequence segment to be determined. The tangent point is the estimated capacity value.

[0125] C. Estimating capacity based on capacity curves;

[0126] The first three steps above determine the upper and lower limits of the estimated capacity. In the absence of measurement errors, due to the existence of user consumption load, the capacity value must be greater than the absolute value of the minimum net load value P floor , so this is used as the lower limit of the capacity estimate.

[0127] The overall slope of the curve between the starting point and the tangent point is significantly smaller than that between the tangent point and the end point. This is because the "plateau" between the starting point and the tangent point represents the interval with the highest frequency of capacity values, and is also the interval with the largest probability density in the entire capacity sequence to be determined. By selecting the tangent point, the value with the highest probability of occurrence in this interval is given an adaptive increment to achieve capacity estimation.

[0128] D. Constructing a probability density function

[0129] On a monthly basis, there is a high correlation between total daytime and nighttime consumer load. At night, when no PV power is generated, net load data equals consumer load data. The nighttime consumer load and its correlation with daytime consumer load can be used to estimate daytime consumer load. This correlation is based on the correlation between nighttime net load and daytime consumer load.

[0130] This step uses this correlation to construct the joint probability density function of the daytime and nighttime monthly consumption loads.

[0131] First, the hourly consumption load during the day and night is accumulated to obtain the monthly consumption load, as shown in formula (5).

[0132]

[0133] Then, Gaussian mixture modeling (GMM) is used to construct the joint distribution of monthly nighttime and daytime consumption loads of observable users, as shown in the following equation (6).

[0134]

[0135] Where f(·,·) represents the joint probability density function estimated by GMM, Λ={S,θ k ,μ k ,Σ k} are all parameters in GMM, which need to be learned using data from a large number of users.

[0136] The parameter set to be solved is transformed into an optimization problem, as shown in Equation (7), and solved using the expectation-maximization (EM) algorithm.

[0137]

[0138] Among them, N represents the total number of users, L day,m (j), L night,m (j) represents the monthly daytime power sum and nighttime power sum of the j-th user, respectively.

[0139] E. Generate typical samples

[0140] Considering that within a certain geographical range, the weather conditions received by different users are highly similar, the photovoltaic output power curves generated by them due to the photovoltaic power generation mechanism are also highly similar. Therefore, the photovoltaic output power curve of the unknown user can be represented by a linear combination of the photovoltaic output power curves of the observable users.

[0141] The present invention clusters photovoltaic output power curves known to users to form typical samples, so as to improve calculation efficiency.

[0142] First, all known photovoltaic output power curves are normalized, as shown in formula (8).

[0143]

[0144] Among them, G h (t) represents the photovoltaic output power curve, G max and G min Represent the maximum and minimum values ​​of the photovoltaic output power curve respectively.

[0145] Then, all normalized standard photovoltaic output power curves are clustered by mean shift, and the cluster centers of the generated N clusters are selected as typical samples. And calculate its power and

[0146]

[0147] F. Perform maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition

[0148] Using the net load equal to the consumption load minus the photovoltaic power generation, we can derive the monthly cumulative sum equation (9).

[0149]

[0150] Where w=[w1,··w N ] T Represents the unknown weight that needs to be solved.

[0151] Combined with the probability density function generated by the Gaussian mixture model, the constrained optimization problem of formula (10) is constructed as follows:

[0152]

[0153] Among them, the physical meaning of the constraint conditions is that the photovoltaic power generation power is positive and the consumption load is positive.

[0154] The above optimization problem can be solved by numerical calculation method to obtain the optimal weight, and then the weight Provide a typical sample of photovoltaic output power curve Perform linear combination to obtain preliminary estimation results of photovoltaic curve

[0155] G. Get the confidence coefficient

[0156] In reality, not all users strictly conform to the data distribution generated by GMM. For some users, this non-compliance can cause an overall offset in the estimated PV curves, significantly different from the true results. To account for these user characteristics, we perform capacity correction on the initial PV decomposition results.

[0157] First, an unsupervised capacity estimation is performed on the observable users to obtain an estimated value Compare the actual capacity value C and calculate the ratio as the confidence coefficient, as shown in formula (11).

[0158]

[0159] Where N represents the number of visible users.

[0160] H. Use the confidence coefficient to correct the preliminary results and obtain the final photovoltaic decomposition results to estimate the capacity of unknown users. Then the photovoltaic decomposition was performed to obtain preliminary results. Peak The comparison between the ratio and the confidence coefficient is used to determine whether to make a correction. The photovoltaic decomposition results that meet the judgment conditions of formula (12) are corrected by formula (13) to obtain the final photovoltaic decomposition results.

[0161]

[0162] The present invention utilizes the intermittent characteristics of photovoltaic power generation and the stable power characteristics of user consumption load during unused periods to achieve rapid and accurate estimation of photovoltaic capacity under net load.

[0163] The present invention forms typical samples by clustering the photovoltaic output power curves of observable users, constructs a probability density function by utilizing the correlation between nighttime and daytime shown by the consumption load, and obtains the optimal weight by solving the optimization problem, thereby achieving a preliminary estimation of the photovoltaic output power curve.

[0164] This method corrects the original capacity estimation results based on the original data, enhancing the robustness of the original photovoltaic decomposition algorithm. This method achieves high capacity estimation accuracy and maintains good adaptability even with a small number of missing values. It also improves the photovoltaic decomposition algorithm, increasing its computational speed and accuracy.

Claims

1. An improved photovoltaic decomposition method based on capacity estimation, characterized by: The following steps are involved: (A) Generate candidate capacity sequence; (B) Obtaining the capacity curve based on the candidate capacity sequence; (C) Capacity estimation based on capacity curve; (D) Constructing probability density function; (E) Generate typical samples; (F) Perform maximum likelihood estimation to obtain preliminary results of PV decomposition; (G) Get the confidence coefficient; (H) Use the confidence coefficient to correct the preliminary results and obtain the final photovoltaic decomposition results; Step (A) generates a candidate capacity sequence. The specific process is as follows: First, take the net load power data as input; Then, the net load nighttime power extreme value and the net load daytime power extreme value are counted and recorded on a monthly basis; Finally, the candidate capacity sequence is formed by the additive combination of the net load nighttime power extreme value and the net load daytime power extreme value.

2. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (B) obtains the capacity curve based on the candidate capacity sequence. The specific process is as follows: First, sort the obtained candidate capacity sequences; Then, a capacity curve is established based on the sequence number and capacity.

3. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (C) estimates the capacity based on the capacity curve. The specific process is as follows: First, based on the capacity curve, the changing trend of the capacity curve is obtained; Then, the capacity is estimated using the changing trend of the capacity curve.

4. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (D) constructs the probability density function. The specific process is as follows: First, obtain data information of some significant users; Then, the monthly nighttime and daytime electricity consumption of the consumer load is extracted based on the data information; Finally, a probability density function is constructed based on the monthly nighttime and daytime electricity consumption.

5. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (E) generates typical samples. The specific process is as follows: Photovoltaic output power curves are clustered to form typical samples.

6. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (F) performs maximum likelihood estimation to obtain preliminary results of photovoltaic decomposition. The specific process is as follows: First, linear combination is solved for typical samples; Then, after obtaining the optimal weights, the preliminary results of PV decomposition are obtained.

7. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (G) obtains the confidence coefficient. The specific process is as follows: First, based on the data information of some observable users, the real value of capacity is obtained; Then, we obtain the capacity estimate of the considerable users; Finally, the confidence coefficient is obtained by comparing the estimated capacity with the true value of the capacity.

8. The improved photovoltaic decomposition method based on capacity estimation according to claim 1, characterized in that: Step (H) uses the confidence coefficient to correct the preliminary results to obtain the final photovoltaic decomposition results. The specific process is as follows: First, the capacity values ​​in the preliminary results are obtained; Then, the capacity estimate obtained using the capacity estimation method; Then, the capacity estimate and the capacity value are processed to obtain a threshold value; Then, the capacity estimation is compared with the threshold to determine whether it is necessary; Finally, the confidence coefficient is used to correct the part that needs correction.

Images