An ultra-short-term prediction method for irradiance of photovoltaic power station based on AP clustering and Markov chain
By combining AP clustering and Markov chain methods, the irradiance data of photovoltaic power plants is clustered and predicted, which solves the problem of insufficient accuracy in ultra-short-term prediction in existing technologies, realizes higher accuracy in photovoltaic power generation prediction, and improves the stability of the power system and the absorption capacity of photovoltaic power generation.
Patent Information
- Application Number
- CN202210448980.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-04-26
AI Technical Summary
Existing ultra-short-term forecasting methods for photovoltaic power generation suffer from local optima and classification principles that are not adapted to ultra-short-term time scales, resulting in insufficient forecast accuracy and an inability to meet the ultra-short-term dispatching requirements of photovoltaic power generation.
We employ AP clustering and Markov chain-based methods to cluster photovoltaic power plant irradiance data. We select the most similar time period based on the weather conditions during the forecast period, and use Euclidean distance and Markov transition rate matrix to predict irradiance, ensuring the stability and accuracy of the number of clusters and their centers.
It improves the accuracy of photovoltaic irradiance prediction, reduces the uncertainty of photovoltaic power generation on the power grid, enhances the safe and stable operation of the power system and the absorption capacity of photovoltaic power generation, and reduces economic losses.
Smart Images

Figure CN114897229B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of ultra-short-term prediction of irradiance of photovoltaic power station, and specifically relates to an ultra-short-term prediction method of irradiance of photovoltaic power station based on the combination of AP clustering and Markov chain. Background Art
[0002] Photovoltaic power generation is easily affected by meteorological factors such as irradiance, temperature, and cloud cover. The output power is highly random and volatile, which has an adverse impact on the safe and stable operation of the power system and the power generation scheduling plan. Among them, solar irradiance has the greatest impact on photovoltaic power generation. Therefore, it is necessary to improve the irradiance prediction accuracy of photovoltaic power stations, and then improve the prediction accuracy of photovoltaic power generation, provide a reliable basis for grid scheduling, and expand the support of photovoltaic power generation to the grid.
[0003] Photovoltaic forecasts are divided into ultra-short-term, short-term, medium-term and long-term according to the time scale. Ultra-short-term forecasts use a 15-minute resolution to predict photovoltaic power generation in the next 15 minutes to 4 hours, which is of great significance for incorporating photovoltaic power generation into spot transactions in the power market, arranging energy storage system configuration, and formulating power dispatch plans.
[0004] At present, the ultra-short-term prediction methods of photovoltaic power generation are mainly divided into two categories: physical model and data-driven. Physical model prediction is to directly mathematically model the photovoltaic power generation process based on the conversion efficiency of photovoltaic power stations, the installation angle of photovoltaic arrays and local meteorological information. The physical model has high requirements for photovoltaic components and environmental meteorological parameters of power stations, and has the characteristics of complex modeling, poor scalability, and more suitable for medium- and long-term predictions. The data-driven method mainly uses specific algorithms to establish a mapping relationship between the output of photovoltaic power stations and historical conditions (often meteorological data). Common algorithms include Markov chains, support vector machines, neural networks, etc. Data-driven modeling is relatively simple and the prediction cost is relatively low, so it is widely used in photovoltaic power generation prediction.
[0005] Regarding the Markov chain method for predicting irradiance, the paper "Short-term prediction method for output power of photovoltaic power generation system based on Markov chain" proves that the random process of solar irradiance changing with sampling time has a single-step memory feature, so the discrete Markov chain can be used to describe the irradiance changes in the past. The paper "Simulation of photovoltaic output time series using improved Markov chain" considers seasonal characteristics, daily characteristics and weather characteristics, divides historical data into 96 categories of 12 months, two periods of morning and afternoon, and four types of weather: sunny, cloudy, showers, and heavy rain. Photovoltaic fluctuations are simulated by random sampling to generate photovoltaic output time series. The paper "Adaboost weather clustering ultra-short-term prediction method for photovoltaic power generation in microgrids" first classifies historical data based on four typical weather types: sunny, cloudy, overcast, and rainy, and then weights each step-length Markov chain according to the influence weight to predict the probability distribution matrix of the irradiance value at the next moment.
[0006] Based on the above literature, the Markov chain method is theoretically feasible for predicting photovoltaic power generation, but when applied to ultra-short-term prediction, the existing data processing methods are not adapted to the ultra-short-term time scale. When clustering historical data, the existing methods generally consider climate and weather factors and classify them according to the four seasons of the year, 12 months, or four typical weather types: sunny, cloudy, overcast, and rainy. These classification methods have three disadvantages: (1) The number of classifications is determined in advance, which is easy to fall into the local optimal situation; (2) The classification principle is difficult to clearly distinguish the irradiance sequence, such as the uncertainty of the date division of the four seasons, and the last day of a month and the first day of the next month have no obvious difference but are classified in different categories; (3) The classification principle is inconsistent with the ultra-short-term time scale. For example, if it is cloudy on a certain day, but there is sunshine for 4 hours, the irradiance curve is similar to that on a sunny day. In addition, under the above classification principle, similar days are mostly selected based on key weather characteristics. The key weather characteristics of similar days are obtained from weather forecasts, which are not necessarily consistent with the ultra-short-term 4-hour key weather characteristics. Therefore, it is necessary to propose an irradiance prediction method that adapts to the ultra-short-term time scale. Summary of the invention
[0007] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes an ultra-short-term prediction method for irradiance of photovoltaic power stations based on AP clustering and Markov chain, in order to globally optimally cluster the historical data of the ultra-short-term prediction period, and select the most similar period in combination with the weather conditions of the prediction period, so as to achieve better prediction results in any climate, weather and time, thereby providing a theoretical basis and scientific foundation for photovoltaic power grid connection, scheduling and energy storage configuration.
[0008] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is:
[0009] The ultra-short-term prediction method of irradiance of a photovoltaic power station based on AP clustering and Markov chain of the present invention is characterized in that it includes:
[0010] Step 1: Set the prediction period to [t1, t2, …, t n ], the moment before the prediction period is t0; where t n Indicates the nth prediction moment in the prediction time period;
[0011] Extract the full time period of all dates in the historical data [t0, t1, t2, …, t n ] and form a total of m days of full-time irradiance sequence I 1 ,I 2 ,…,I i ,…,I m , where I i represents the irradiance sequence of the i-th day, and represents the nth prediction time t n Irradiance sequence, i = 1, 2, ..., m;
[0012] Step 2: Use AP clustering algorithm to cluster the irradiance sequences of all dates in the full time period I 1 ,I 2 ,…,I m Perform clustering, calculate the number of clusters M and the date of the irradiance series included in each cluster;
[0013] Step 3: Calculate the total time period [t0, t1, t2, …, t n ]The average temperature at each moment, and the average temperature of each category over the entire time period in, represents the nth prediction time t in the kth category n The average temperature of , k=1,…,M;
[0014] Step 4: According to the weather forecast, calculate the weather forecast temperature for the entire time period of the day to be predicted [T0, T1, T2, …, T n ], where T n Indicates the nth prediction time t of the day to be predicted n Weather forecast temperature;
[0015] Step 5: Calculate the weather forecast temperature [T0, T1, T2, …, T n ] and the average temperature of the kth category over the entire time period The Euclidean distance D k , thereby calculating the weather forecast temperature for the entire time period of the predicted day [T0, T1, T2, …, T n ] and the average temperature of each class over the entire time period, and select the class with the smallest Euclidean distance as the most similar class;
[0016]
[0017] Step 6: Sort the irradiance of all dates at time t0 in the most similar category from small to large, and remove the same values to obtain the irradiance column vector containing x0 irradiance values in, Represents the x0th irradiance value of all different irradiances at time t0 in the most similar classification, sorted from small to large;
[0018] Step 7: Based on the measured irradiance value I0 at time t0 on the day to be predicted and the irradiance column vector C0 at time t0 in the most similar classification, calculate the irradiance probability distribution P at time t0 0 ;
[0019] Step 8: Define variable q and initialize q=1;
[0020] Step 9: Classify the most similar q The irradiance of all dates at the time is sorted from small to large, and after removing the same values, the irradiance of all dates including x is obtained. q irradiance column vector of irradiance values in, Indicates the most similar category t q All different irradiances at the time are sorted from small to large. q irradiance values;
[0021] Step 10: Calculate the most similar category t q-1 Time transfer to t q The irradiance transfer times matrix B at the time q-1,q and the irradiance transfer rate matrix A q-1,q ;
[0022] Step 11: Calculate t according to formula (2) q The probability distribution of irradiance at time P q ;
[0023] P q =P q-1 A q-1,q (2)
[0024] Step 12: Place the q The probability distribution of irradiance at time P q With t q The irradiance column vector C at time q After multiplication, we get t q Predicted irradiance at the time
[0025] Step 13: Determine whether q=n is true. If so, the predicted time period [t1, t2, …, t n ] irradiance prediction sequence Otherwise, assign q+1 to q and return to step 9 to execute sequentially.
[0026] The ultra-short-term prediction method for irradiance of a photovoltaic power station based on AP clustering and Markov chain described in the present invention is also characterized in that step 2 comprises:
[0027] Step 2.1: Set the damping coefficient to λ and the maximum number of iterations to N max The number of times the cluster center remains stable is N stable ;
[0028] Step 2.2: Calculate the similarity matrix S m×mThe value of the off-diagonal element s(i,j) in the i-th row and j-th column, let s(i,j) be the irradiance sequence for the entire time period of the i-th day The irradiance sequence of the whole period of day j The negative Euclidean distance between , and j≠i;
[0029] Step 2.3: Calculate the similarity matrix S m×m The value of the diagonal element s(i,i) in the i-th row and i-th column of , let s(i,i) be the similarity matrix S m×m The median of all off-diagonal elements in the i-th row of ;
[0030] Step 2.4: Construct the initial attraction matrix and belonging matrix, both of which are all zero matrices, and based on the similarity matrix S m×m Iterate the attraction matrix and the attribution matrix until the cluster center remains stable for N times. stable Or reach the maximum number of iterations N max until;
[0031] Step 2.5: The sum of the attraction matrix and the belonging matrix after the final iteration is taken as the cluster center matrix, and the number of rows whose diagonal elements of the cluster center matrix are greater than zero is selected as the cluster center number, and the irradiance sequence represented by the cluster center number is set as the cluster center, so as to calculate the number of cluster centers, which is the number of clusters M;
[0032] Step 2.6: Select the column number where the maximum value of each row in the cluster center matrix is located as the cluster center to which the irradiance sequence represented by the current row belongs, and count the dates of all irradiance sequences in each class.
[0033] The step 7 comprises:
[0034] Step 7.1: Assume that the initial irradiance probability distribution sequence P at time t0 is 0 is a row vector of x0 columns, whose elements are all 0;
[0035] Step 7.2: Compare the measured irradiance value I0 at time t0 on the day to be predicted with the irradiance column vector C0 at time t0 in the most similar category. If there is an e-th value in C0 Then the initial irradiance probability distribution sequence P 0 The e-th element of is reset to 1, and the irradiance probability distribution P at time t0 is obtained 0 , and execute step 8, otherwise, it means that there is no value in C0 equal to I0, and execute step 7.3, 1≤e≤x0;
[0036] Step 7.3: Find the two values closest to I0 and record them as the fth number and the gth number Arrange in order from smallest to largest:
[0037] Step 7.4: The irradiance probability distribution sequence P at time t0 0 The fth element of is reset to The gth element is reset to Thus, the irradiance probability distribution P at time t0 is obtained 0 .
[0038] The step 10 comprises:
[0039] Step 10.1: Assume that in the most similar category, t q-1 Column vector of irradiance at time t q Column vector of irradiance at time Then the irradiance transfer times matrix B q-1,q is an x q-1 Row x q A column matrix, where the element B(y,z) in the yth row and zth column is t q-1 The irradiance at the moment is And t q The irradiance at the moment is The number of occurrences of , 1≤y≤x q-1 ; 1≤z≤x q ;
[0040] Step 10.2: Let the transfer rate matrix A q-1,q is an x q-1 Row x q Column matrix, and use formula (3) to calculate the element A(y,z) in the y-th row and z-th column;
[0041]
[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0043] 1. The accurate photovoltaic power generation irradiance prediction of the present invention is the basis for accurate prediction of photovoltaic output, and can provide a numerical basis for the grid-connected operation and dispatching of photovoltaic power stations, reduce the uncertainty of input grid power, ensure the safe and stable operation of the power system, and effectively improve the absorption capacity of photovoltaic power generation, thereby reducing the economic losses caused by power curtailment and abandonment of photovoltaic power stations, and improving operational management efficiency.
[0044] 2. Before calculating the Markov transfer rate matrix, the present invention clusters the irradiance sequences in the same prediction period, avoiding the traditional classification method of using typical weather conditions of four seasons, 12 months, sunny, cloudy and rainy weather. The clustering method is more scientific and clear.
[0045] 3. The present invention uses the AP clustering method to cluster the irradiance, which can directly obtain the number of clusters and the cluster center, and the results of multiple calculations are the same, avoiding the clustering instability caused by the free selection of the number of clusters and the cluster center by the traditional K-means method.
[0046] 4. When clustering solar irradiance, the present invention directly clusters the irradiance data of the ultra-short-term prediction period (no more than 4 hours), rather than the traditional clustering of the full-day data of the prediction day, thereby avoiding the influence of the irradiance clustering due to the difference between the weather conditions on the prediction day and the weather conditions in the prediction period;
[0047] 5. The present invention selects the category whose average temperature of each category is closest to the Euclidean distance of the weather forecast temperature of the forecast time period as the most similar category, so that the category that is consistent with the temperature characteristics of the time period to be predicted can be objectively selected, avoiding the situation where the weather on the forecast day is inconsistent with the weather in the forecast period. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 is a flow chart of the method of the present invention;
[0049] Figure 2 This is a graph of the average temperature of the total time period of some categories in this embodiment;
[0050] Figure 3 This is a comparison chart of the prediction results of this embodiment. DETAILED DESCRIPTION
[0051] In this embodiment, all available measurement data (812 records in total) of the Alice Springs site of the DKA Solar Centre (Desert Knowledge Australia Solar Centre) in Australia from July 2019 to December 2021 are used as samples to predict the irradiance on January 12, 2022. An ultra-short-term prediction method for irradiance of photovoltaic power stations based on AP clustering and Markov chain is proposed. Figure 1 As shown, the following steps are included:
[0052] Step 1: Assume the prediction time period is [8:05, 8:10, …, 12:00], that is, t1 = 8:05, t n =12:00, the time interval is 5 minutes, there are n = 48 moments in total, the moment before the prediction time period is t0 = 8:00, and the full time period is [8:00, 8:05, 8:10, ..., 12:00]. Extract the irradiance sequence of the full time period of all dates in the historical data to form a total of m = 812 days of the full time period irradiance sequence I 1 ,I 2 ,…,I i ,…,I812 , where I i represents the irradiance sequence of the i-th day, and represents the nth prediction time t n Irradiance sequence; i = 1, 2, ..., m;
[0053] Step 2: Use AP clustering algorithm to cluster the irradiance sequences of all dates in the full time period I 1 ,I 2 ,…,I m Perform clustering, calculate the number of clusters M and the date of the irradiance series included in each cluster;
[0054] Step 2.1: Set the damping coefficient λ = 0.5 and the maximum number of iterations N max = 1000, the number of times the cluster center remains stable N stable =100;
[0055] Step 2.2: Calculate the similarity matrix S m×m The value of the off-diagonal element s(i,j) in the i-th row and j-th column, let s(i,j) be the irradiance sequence for the entire time period of the i-th day The irradiance sequence of the whole period of day j The negative Euclidean distance between , and j≠i;
[0056] Step 2.3: Calculate the similarity matrix S m×m The value of the diagonal element s(i,i) in the i-th row and i-th column of , let s(i,i) be the similarity matrix S m×m The median of all off-diagonal elements in the i-th row of ;
[0057] Step 2.4: Construct the initial attraction matrix and belonging matrix, both of which are all zero matrices, and based on the similarity matrix S m×m Iterate the attraction matrix and the attribution matrix until the cluster center remains stable for N times. stable = 100 or the maximum number of iterations N is reached max =1000;
[0058] Step 2.5: The sum of the attraction matrix and the belonging matrix after the final iteration is used as the cluster center matrix, and the number of rows whose diagonal elements of the cluster center matrix are greater than zero is selected as the cluster center number, and the irradiance sequence represented by the cluster center number is set as the cluster center, so as to calculate the number of cluster centers, that is, the cluster number M. In this embodiment, the cluster number M=68, and the cluster center numbers are shown in Table 1;
[0059] Table 1 Cluster center numbers in this embodiment
[0060] 84 91 92 124 136 137 148 149 153 176 188 236 60 248 253 346 380 386 387 388 389 390 391 392 393 394 401 403 769 416 429 256 453 457 458 463 469 471 479 444 481 482 484 486 498 541 549 573 291 614 639 643 732 733 734 735 736 737 741 742 743 744 749 753 775 777 765 795
[0061] Step 2.6: Select the column number where the maximum value of each row in the cluster center matrix is located as the cluster center to which the irradiance sequence represented by the current row belongs, and count the dates of all irradiance sequences in each class. Table 2 lists the irradiance sequence numbers and corresponding dates of some cluster centers, and the irradiance sequence numbers and corresponding dates contained in each class;
[0062] Table 2 Part of the cluster centers and the irradiation sequence numbers and corresponding dates
[0063]
[0064]
[0065] Step 3: Calculate the total time period [t0, t1, t2, …, t n ]The average temperature at each moment, and the average temperature of each category over the entire time period in, represents the nth prediction time t in the kth category n The average temperature of, k = 1, ..., M, Figure 2 The average temperatures for the total time period of the five categories in Table 1 are listed;
[0066] Step 4: According to the weather forecast, calculate the weather forecast temperature for the entire time period of the day to be predicted [T0, T1, T2, …, T n ], where T n Indicates the nth prediction time t of the day to be predicted n In this embodiment, the weather forecast temperature for the entire period from 8:00 am to 12:00 pm on January 12, 2022 is [32.7 32.8 … 39.2];
[0067] Step 5: Calculate the full-time weather forecast temperature [T0, T1, T2, …, T n ] and the average temperature of the kth category over the entire time period The Euclidean distance D k , thereby calculating the weather forecast temperature for the entire time period of the predicted day [T0, T1, T2, …, T n ] and the average temperature of each category over the entire time period, and select the category with the smallest Euclidean distance as the most similar category. In this embodiment, the most similar category is category 8, and its Euclidean distance is D8=4.4;
[0068]
[0069] Step 6: Sort the irradiance of all dates at t0 = 8:00 in the most similar category from small to large, and remove the same values to obtain the irradiance column vector containing x0 irradiance values in, It represents the x0th irradiance value of all different irradiances at time t0 in the most similar classification, sorted from small to large. In this embodiment, there are 2 irradiance sequences at time t0=8:00, which are Then x0=2,
[0070] Step 7: Based on the measured irradiance value I0=431 at t0=8:00 on the predicted day, January 12, 2022, and the irradiance column vector at t0=8:00 in the most similar classification The irradiance probability distribution P at t0 = 8:00 is calculated 0 ;
[0071] Step 7.1: Let t0 = the initial irradiance probability distribution sequence P at 8:00 0 is a row vector with x0=2 columns, all of which are 0, that is, P 0 = [0 0];
[0072] Step 7.2: Compare the measured irradiance value I0=431 at t0=8:00 on the predicted day with the irradiance column vector at t0=8:00 in the most similar category If C0 does not have a value equal to I0, proceed to step 7.3;
[0073] Step 7.3: Find the two values in C0 that are closest to I0 and record them as the f=1th number and the g=2nd number Arrange in order from smallest to largest:
[0074] Step 7.4: Set the irradiance probability distribution sequence P at t0 = 8:00 0 The f=1th element of is reset to The g=2th element is reset to Thus, the probability distribution of irradiance at t0 = 8:00 is obtained
[0075] Step 8: Define variable q and initialize q=1;
[0076] Step 9: Classify the most similar q The irradiance of all dates at the time is sorted from small to large, and the same values are eliminated to obtain the dates containing x q irradiance column vector of irradiance values in, Indicates the most similar category t q All different irradiances at the time are sorted from small to large. q irradiance value, in this embodiment, x1=2,
[0077] Step 10: Calculate the most similar category t q-1 Time transfer to t q The irradiance transfer times matrix B at the time q-1,q and the irradiance transfer rate matrix A q-1,q ;
[0078] Step 10.1: Assume that in the most similar category, t q-1 Column vector of irradiance at time t q Column vector of irradiance at time Then the irradiance transfer times matrix B q-1,q is an x q-1 Row x q A column matrix, where the element B(y,z) in the yth row and zth column is t q-1 The irradiance at the moment is And t q The irradiance at the moment is The number of occurrences of , 1≤y≤x q-1 ; 1≤z≤x q In this embodiment, B(1,1)=1,
[0079] Step 10.2: Let the transfer rate matrix A q-1,q is an x q-1 Row x q The matrix is formed by using the formula to calculate the element A(y,z) at the y-th row and the z-th column. In this embodiment, A(1,1)=1.
[0080]
[0081] Step 11: Calculate t according to the formula q The probability distribution of irradiance at time P q In this embodiment,
[0082] P q =P q-1 A q-1,q (3)
[0083] Step 12: Place the q The probability distribution of irradiance at time P q With tq The irradiance column vector C at time q After multiplication, we get t q Predicted irradiance at the time In this embodiment, the irradiance probability distribution at time t1 is The irradiance sequence at time t1 Multiply them to get the predicted irradiance value at time t1
[0084] Step 13: Determine whether q=n is true. If so, the predicted time period [t1, t2, …, t n ] irradiance prediction sequence Otherwise, after assigning q+1 to q, return to step 9 and execute sequentially. In this embodiment, q=1, n=48, q=n does not hold. After assigning q+1=2 to q, return to step 9 and execute sequentially.
[0085] Step 14: After 48 iterations, q = n holds, which means that the predicted time period [t1, t2, …, t n ] irradiance prediction sequence The comparison between the predicted results and the measured values is as follows Figure 3 shown.
Claims
1. A method for ultra-short-term prediction of irradiance of photovoltaic power station based on AP clustering and Markov chain, characterized in that: include: Step 1: Set the forecast period to , the moment before the prediction period is moment; among them, Indicates the nth prediction moment in the prediction time period; Extract the full time period of all dates in historical data The irradiance sequence is composed of Irradiance series for the entire day ,in, Indicates The daily irradiance sequence, and , Indicates the nth prediction moment The irradiance sequence is ; Step 2: Use AP clustering algorithm to cluster the irradiance series of all dates in all time periods Perform clustering and calculate the number of clusters and the dates of the irradiance series included in each category; Step 3: Calculate the total time period in each category The average temperature at each moment, to obtain the average temperature of each category over the entire time period ,in, Indicates The nth prediction moment in each category The average temperature, ; Step 4: According to the weather forecast, calculate the weather forecast temperature for the entire time period of the day to be predicted ,in, Indicates the nth prediction time of the day to be predicted Weather forecast temperature; Step 5: Calculate the weather forecast temperature for the entire time period of the forecast day according to formula (1) With Average temperature for all time periods of each category The Euclidean distance between , thereby calculating the weather forecast temperature for the entire time period of the predicted day The Euclidean distance between the average temperature of each class over the entire time period, and the class with the smallest Euclidean distance is selected as the most similar class; (1) Step 6: Classify the most similar The irradiance of all dates at the time is sorted from small to large, and the same values are removed to obtain irradiance column vector of irradiance values ,in, Indicates the most similar category All different irradiances at the moment are sorted in ascending order. irradiance values; Step 7: According to the forecast day Measured irradiance value at time and most similar categories Column vector of irradiance at time , calculated Probability distribution of irradiance at time ; Step 8: Define variable q and initialize it ; Step 9: Classify the most similar The irradiance of all dates at the time is sorted from small to large, and after removing the same values, the irradiance of all dates at the time is obtained. irradiance column vector of irradiance values ,in, Indicates the most similar category All different irradiances at the moment are sorted in ascending order. irradiance values; Step 10: Calculate the most similar category Time transfer to The irradiance transfer times matrix at time and the irradiance transfer rate matrix ; Step 11: Calculate according to formula (2) Probability distribution of irradiance at time ; (2) Step 12: Probability distribution of irradiance at time and Column vector of irradiance at time After multiplication, we get Predicted irradiance at the time ; Step 13: Judgement Is it true? If so, it means the predicted time period is obtained. Irradiance prediction sequence Otherwise, Assign to Then, return to step 9 and execute sequentially.
2. The ultra-short-term prediction method for irradiance of photovoltaic power station based on AP clustering and Markov chain according to claim 1 is characterized in that: The step 2 comprises: Step 2.1: Set the damping coefficient to The maximum number of iterations is The number of times the cluster center remains stable is ; Step 2.2: Calculate the similarity matrix The off-diagonal elements in row i and column j The value of For the Irradiance series for the entire day With Irradiance series for the entire day The negative Euclidean distance between ; Step 2.3: Calculate the similarity matrix The diagonal elements of the i-th row and i-th column of The value of is the similarity matrix Middle The median of all off-diagonal elements of a row; Step 2.4: Construct the initial attraction matrix and belonging matrix, both of which are all zero matrices, and based on the similarity matrix Iterate the attraction matrix and the attribution matrix until the number of times the cluster center remains stable is reached. Or the maximum number of iterations is reached until; Step 2.5: The sum of the attraction matrix and the belonging matrix after the final iteration is used as the cluster center matrix, and the number of rows whose diagonal elements of the cluster center matrix are greater than zero is selected as the cluster center number. The irradiance sequence represented by the cluster center number is the cluster center, and the number of cluster centers is calculated as the number of clusters. ; Step 2.6: Select the column number where the maximum value of each row in the cluster center matrix is located as the cluster center to which the irradiance sequence represented by the current row belongs, and count the dates of all irradiance sequences in each class.
3. The ultra-short-term prediction method for irradiance of photovoltaic power station based on AP clustering and Markov chain according to claim 1 is characterized in that: The step 7 comprises: Step 7.1: Set The initial irradiance probability distribution sequence at time is a The row vector of the column, whose elements are all 0; Step 7.2: Compare the forecasted day Measured irradiance value at time and most similar categories Column vector of irradiance at time ,like There is Values , then the initial irradiance probability distribution sequence No. elements are reset to 1, and we get Probability distribution of irradiance at time , and execute step 8, otherwise, it means There is no equal values and execute step 7.
3. ; Step 7.3: Find out the The two closest values are recorded as Number and Number , arranged in order from smallest to largest , , ; Step 7.4: Irradiance probability distribution sequence at time No. Elements are reset to , No. Elements are reset to , thus obtaining Probability distribution of irradiance at time .
4. The ultra-short-term prediction method for irradiance of photovoltaic power station based on AP clustering and Markov chain according to claim 1 is characterized in that: The step 10 comprises: Step 10.1: Assume that in the most similar category, Column vector of irradiance at time , Column vector of irradiance at time , then the irradiance transfer times matrix is a OK Column matrix, where the Line Elements of a column yes The irradiance at the moment is and The irradiance at the moment is The number of occurrences of ; ; Step 10.2: Let the transfer rate matrix is a OK Column matrix, and use formula (3) to calculate the Line Elements of a column ; (3)。
