Photovoltaic power interval prediction method and system

Abstract

The invention relates to a photovoltaic power interval prediction method and system, and belongs to the field of new energy power generation. Calculating an initial generation power prediction value of a prediction day according to meteorological factor data and historical power actual data of the prediction day, and dividing a set time period before the to-be-predicted day according to a set time interval to form a similar time period library; selecting similar time period data of which the correlation degree is greater than a threshold value according to the correlation between each piece of meteorological factor data of the prediction day and each piece of historical meteorological factor data in the similar time period library, and determining a prediction error distribution function according to the power actual data and the power prediction data corresponding to the selected plurality of similar time periods, and obtaining a photovoltaic power generation prediction error confidence interval according to the prediction error distribution function and a set significance level, and obtaining a photovoltaic power generation power interval prediction result according to the initial power generation power prediction value and the prediction error confidence interval. The prediction method is high in prediction accuracy and simple in prediction process.

Description

Technical Field

[0001] The invention relates to a photovoltaic power interval prediction method and system, and belongs to the field of new energy power generation. Background Art

[0002] At present, more and more new energy power stations are being built, and the proportion of new energy is gradually increasing. New energy power generation mainly includes wind power generation, hydropower generation and photovoltaic power generation, among which photovoltaic power generation is the most widely used. However, due to the strong dependence of photovoltaic power generation on the meteorological environment, its output has very strong randomness and volatility, which in turn poses new challenges to the safe and stable operation of the power system. With the grid connection of a large number of photovoltaic sites, the volatility and randomness of photovoltaic power generation have an increasingly serious impact on large-scale photovoltaic grid-connected power generation. In order to improve the voltage's ability to absorb sunlight, reduce the economic losses caused to photovoltaic owners due to power restrictions, and increase the return on investment of photovoltaic power stations, it is necessary to predict the power of new energy power generation.

[0003] Currently, the point prediction method is often used to predict photovoltaic power generation. The point prediction method can only accurately predict the power generation power at a certain moment. For the photovoltaic power within a period of time, the prediction error is relatively large and the prediction result is not accurate. Summary of the invention

[0004] The purpose of the present invention is to provide a photovoltaic power interval prediction method and system to solve the problem of inaccurate prediction of existing point prediction methods.

[0005] To achieve the above object, the solution of the present invention includes:

[0006] A photovoltaic power interval prediction method of the present invention comprises the following steps:

[0007] 1) Obtain historical data for a set time period before the day to be measured, the historical data including historical actual power data, corresponding power forecast data and various historical meteorological factor data, and obtain various meteorological factor data on the forecast day, and calculate the initial power forecast value of the forecast day according to the various meteorological factor data on the forecast day and the historical actual power data;

[0008] 2) Divide the historical data of the set time period before the test date into the set time interval, and obtain the historical data of each time period after division to form a similar time period library;

[0009] 3) Select multiple similar time periods based on the similarity between the meteorological factor data of the forecast day and the historical meteorological factor data in the similar time period database;

[0010] 4) Determine the prediction error distribution function based on the actual power data and power forecast data in multiple selected similar time periods, obtain the photovoltaic power generation prediction error confidence interval based on the prediction error distribution function and the set significance level, and obtain the interval prediction result of photovoltaic power generation from the initial power generation prediction value and the prediction error confidence interval.

[0011] Beneficial effects: The photovoltaic power interval prediction method of the present invention determines similar time periods of each time period on the prediction day, calculates the prediction error of each similar time period according to the actual power data in the similar time period and the power prediction data of the corresponding time period; uses the prediction error of each time period to obtain the confidence interval of the prediction error, and corrects the initial prediction value obtained by the point prediction through the error confidence interval to obtain the photovoltaic power generation power interval prediction result. The present invention combines the confidence interval and similar time periods to make an accurate interval prediction of the photovoltaic power generation power in the future, which can not only assist the power grid in long-term scheduling of supply and demand balance and help users reduce assessments, but also assist in decision-making on the investment benefits of photovoltaic station construction.

[0012] Furthermore, when determining the confidence interval of the photovoltaic power generation prediction error, the prediction error level is determined according to the prediction error and the startup capacity corresponding to similar time periods, and the error level distribution is obtained using the prediction error level, and then the error level confidence interval is obtained, and the prediction error confidence interval is calculated according to the startup capacity on the prediction day and the error level confidence interval.

[0013] Beneficial effect: The present invention further takes into account the startup capacity, that is, taking into account the fact that the startup capacity of new energy sources may change during historical data and the forecast period, such as maintenance, power rationing, or power station expansion, the startup capacity has an impact on the forecast error. When the startup capacity is taken into account, the accuracy of the forecast can be further improved.

[0014] Furthermore, the confidence interval of the photovoltaic power generation prediction error is the product of the startup capacity on the prediction day and the confidence interval of the prediction error level.

[0015] Furthermore, the historical meteorological factor data and the meteorological factor data of the forecast day are obtained after screening using the Pearson correlation coefficient.

[0016] Beneficial effect: The Pearson correlation coefficient is used to filter out meteorological factor data with small correlation coefficients, and only meteorological data with large correlation coefficients are considered, which facilitates subsequent calculations without reducing the accuracy of predictions.

[0017] Furthermore, the meteorological factor data are standardized data, including at least two of radiation, sunshine time, average wind speed, average temperature, relative humidity and atmospheric transparency.

[0018] Beneficial effect: The more fully meteorological factors are considered, the more accurate the forecast will be.

[0019] Furthermore, the process of determining similar time periods is as follows: similarity is calculated using the meteorological factor data of the forecast day and the meteorological factor data of each time period in the similar time period library; the similarity is obtained using the meteorological factor data of each time period in the similar time period library, the meteorological factor data of the forecast day and the weight of each meteorological factor data.

[0020] Furthermore, the similarity is cosine similarity.

[0021] Furthermore, the weight of each meteorological factor data is obtained by using the Pearson correlation coefficient of each meteorological factor data in a similar time period library.

[0022] A photovoltaic power interval prediction system of the present invention comprises a processor, wherein the processor is used to execute instructions to implement the photovoltaic power interval prediction method as described above.

[0023] Beneficial effects: The prediction system of the present invention has a simple structure and a simple and clear processing logic. It can combine confidence intervals and similar time periods to make accurate interval predictions on photovoltaic power generation in the future. It can not only assist the power grid in long-term scheduling of supply and demand balance and help users reduce assessments, but also assist in decision-making on the investment benefits of photovoltaic station construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic flow chart of a multi-scale photovoltaic power generation prediction method based on confidence intervals in a method embodiment of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0026] Photovoltaic power interval prediction method embodiment:

[0027] The idea of ​​the present invention is to provide a photovoltaic power interval prediction method, which uses the interval estimation method to correct the existing photovoltaic power point prediction results, and uses the error value of the photovoltaic power history data to obtain the distribution function of the photovoltaic power prediction result error, thereby realizing the photovoltaic power prediction interval result.

[0028] like Figure 1A photovoltaic power interval prediction method is shown, firstly, by obtaining a similar period of the prediction day within a set time period before the prediction day, and obtaining an initial prediction result, obtaining a similar period of the prediction time period according to the historical meteorological factor data within the set time period and the meteorological data of the prediction day, determining a prediction error distribution function according to the actual power data of the similar period and the corresponding power prediction data, obtaining a photovoltaic power generation prediction error confidence interval according to the prediction error distribution function and the set significance level, and obtaining a photovoltaic power generation prediction result by combining the initial power generation prediction result with the power generation error prediction confidence interval. The following is a specific implementation method.

[0029] Obtain historical data for a set time period before the day to be tested, including historical actual power data, corresponding power forecast data, and historical meteorological factor data, and obtain meteorological factor data on the forecast day. Calculate the initial power generation forecast value on the forecast day based on the meteorological factor data on the forecast day and historical actual power data.

[0030] Specifically, this embodiment selects actual power data, power forecast data and standardized historical meteorological factor data of a set time period before the test date (for example, 60 days before the prediction date) to form an overall data set A (the accuracy of time in data A is less than 5 minutes), and the meteorological factor data type is the indicator obtained by screening according to the Pearson coefficient.

[0031] The historical meteorological factor data in this embodiment includes at least two of radiation, sunshine time, average wind speed, average temperature, relative humidity and atmospheric transparency. Since the dimensions and magnitudes of various meteorological data vary greatly, it is necessary to normalize each indicator data to the interval [0, 1], and the normalization formula is as follows:

[0032]

[0033] Where: x j is the jth factor value, x max is the maximum value of the jth factor in the meteorological data set, x min is the minimum value of the jth factor in the meteorological data set, x ij It is the standardized value of the jth factor on the i-th meteorological day.

[0034] After the historical meteorological factor data is standardized, the Pearson coefficients of different meteorological factors and historical actual power generation data are calculated. Among them, the Pearson correlation coefficient reflects the degree of correlation between two variables and is defined as:

[0035]

[0036] Where: X，Yis the Pearson correlation coefficient; X and Y are different sets of variables; cov(X, Y) is the covariance of X and Y; σ X and σ Y are the standard deviations of X and Y respectively; E(·) is the mathematical expectation. From this, we can get the weights of different meteorological characteristics.

[0037] In addition, the historical meteorological factor data and the meteorological factor data of the forecast day are obtained after screening using the Pearson correlation coefficient. The meteorological factor data with a Pearson coefficient correlation less than a threshold value are eliminated, and the Pearson coefficients corresponding to the remaining meteorological factor data are combined to obtain a meteorological factor weight vector, and the meteorological characteristic quantity of each historical meteorological factor data is obtained according to the meteorological factor weight vector. Specifically, for example, after obtaining the Pearson correlation coefficient of each meteorological factor data, the indicators with weak correlation whose correlation coefficient is less than 0.1 are eliminated, and the Pearson correlation coefficients of the remaining meteorological factor indicators are combined to obtain the meteorological factor weight vector a. As another embodiment, the meteorological factor data can also be screened in the form of expert scoring.

[0038] Secondly, the historical data of the set time period before the test date is divided into set time intervals, and the historical data of each time period after division constitutes a similar time period library.

[0039] In this embodiment, the overall data set A is divided into n time periods at set time intervals to form a similar time period library, that is, to form a data set [A 1 , A 2 , …, A n For example, the overall data set A is divided into n time periods at intervals of 1 hour to form data sets [A 1 , A 2 , …, A n ].

[0040] Each data set A i The corresponding one-hour historical meteorological data is included in the standardized processing. The meteorological factor weight vector a is calculated using the Pearson correlation coefficient to calculate each data set A. i The meteorological characteristic quantity W i , get the data set [W 1 , W 2 , …, W n ].

[0041] Then, based on the correlation between the meteorological factor data of the forecast day and the historical meteorological factor data in the similar time period library, similar time period data with a correlation greater than a threshold are selected.

[0042] The process of determining similar time periods is as follows: the similarity is calculated using the meteorological factor data of the forecast day and the meteorological factor data of each time period in the similar time period library; the similarity is obtained based on the meteorological factor data of each time period in the similar time period library, the meteorological factor data of the forecast day, and the weight of each meteorological factor data. The weight of each meteorological factor data is obtained using the Pearson correlation coefficient of each meteorological factor data in the similar time period library. In the process of calculating weights and selecting similar time periods, this embodiment can also select similar time periods by calculating the meteorological characteristic quantities of each meteorological factor data. Specifically, meteorological factors will also change in different time periods of the same day. In order to make the meteorological characteristic quantities more timely, this embodiment divides a day into multiple time periods and assigns different weights to meteorological factors with strong correlation. Based on the traditional method of selecting similar days, this embodiment selects each similar time period in a day, and calculates the meteorological characteristic quantities W of daily meteorological factors in each time period in a day as the basis for selecting similar time periods. The meteorological characteristic quantity W of each time period throughout the day is defined as:

[0043] W i =aX 1i +aX 2i +…+aX ni

[0044] Where: W i is the meteorological characteristic quantity of the whole day on the i-th day; X 1i , X 1i , …, X ni is the n different meteorological characteristic vectors (such as temperature, irradiance, wind speed, etc.) on the i-th day, and a is the weight vector, which is the Pearson coefficient of the corresponding period.

[0045] When selecting similar days, this embodiment adopts the cosine similarity method, which measures the size of the difference by the cosine value of the angle between two vectors in the vector space. The specific formula is as follows:

[0046]

[0047] Where D oi The closer the value is to 1, the higher the W ik and W ok The more similar the two are.

[0048] In addition, each dataset A i The corresponding actual power data and power forecast data within one hour are included in the data set A, so each data set A can be calculated. i The error of photovoltaic power point prediction in the dataset E is obtained. i , forming the initial error data set [E 1 ′, E, ′…, E n ′].

[0049] Finally, the prediction error distribution function is determined based on the actual power data and power forecast data corresponding to multiple selected similar time periods. According to the prediction error distribution function and the set significance level, the photovoltaic power generation prediction error confidence interval is obtained, and the photovoltaic power generation power interval prediction result is obtained from the initial power generation power prediction value and the prediction error confidence interval.

[0050] Considering that the startup capacity of new energy sources may change during the statistical period of historical data and the forecast period (such as maintenance, power rationing, or power station expansion, etc.), when determining the error confidence interval, the forecast error level is determined based on the forecast error and startup capacity corresponding to similar time periods, and the error distribution function is obtained using the forecast error level, and then the error level confidence interval is obtained. The forecast error confidence interval is calculated based on the startup capacity on the forecast day and the error level confidence interval. The photovoltaic power generation error level is the quotient of the photovoltaic power generation error and the photovoltaic startup capacity, and then the error data set sample [E 1 , E 2 ,…,E n ].

[0051] In addition, if the aging of photovoltaic modules and other technological updates are taken into account, corresponding conversions can be made.

[0052] The photovoltaic power generation prediction error in this embodiment obeys a normal distribution function with a mean of 0, that is, the distribution function T of the photovoltaic power generation error level is:

[0053]

[0054] According to the actual needs, the significance level α is given. According to the set significance level, we have:

[0055]

[0056] Further:

[0057]

[0058] Then the confidence interval of the mathematical expectation μ with a confidence level of 1-α is Among them, the confidence interval can be calculated by setting the significance level. The smaller the set significance level, the larger the corresponding confidence interval. It is usually more appropriate to set the confidence interval at around 90% to 99.9%.

[0059] In the formula, n is the selected similar period, is the mean of the distribution function T of the photovoltaic power generation error level, which is 0 here; σ is the error level data set [E 1 , E 2 ,…,E n], which is also the unbiased estimate of the standard deviation of the distribution function T of the photovoltaic power generation error level, It can be obtained by querying the normal distribution quantile table. After substituting the above data, the confidence interval of the mathematical expectation μ of the normal distribution can be calculated.

[0060] According to the confidence interval of the mathematical expectation μ of the obtained normal distribution Assume that the error confidence interval calculated after the parameters are entered is [θ 1 ,θ 2 ], by considering the startup capacity, the confidence interval of the photovoltaic power generation prediction error is the product of the startup capacity on the prediction day and the confidence interval of the prediction error level. The confidence interval of the photovoltaic power generation prediction error is:

[0061]

[0062] Where C is the startup capacity within the forecast day, θ 1 is the lower confidence limit, θ 2 is the upper confidence limit.

[0063] If the result of the current initial power generation point prediction is P, the result of the final photovoltaic power generation range prediction is: [Cθ 1 +P,Cθ 2 +P].

[0064] Combining the above prediction process, the photovoltaic power interval prediction method of this embodiment determines the similar time periods of each time period on the prediction day, and calculates the prediction error level of each similar time period according to the actual power data in the similar time period and the power prediction data of the corresponding time period. The prediction error of each time period is used to obtain the confidence interval of the prediction error, and the photovoltaic power generation power interval prediction result is obtained based on the confidence interval. The present invention combines the confidence interval and the similar time period to make an accurate interval prediction of the photovoltaic power generation power in the future period, which can not only assist the power grid in long-term scheduling of supply and demand balance, help users reduce assessments, but also assist in decision-making on the investment benefits of photovoltaic site construction.

[0065] Photovoltaic power interval prediction system embodiment:

[0066] A photovoltaic power interval prediction system in this embodiment includes a memory, a processor and an internal bus, and the processor and the memory communicate and exchange data with each other through the internal bus. The memory includes at least one memory capable of storing historical power actual data, corresponding power prediction data, and various historical meteorological factor data, and various meteorological factor data of the prediction day. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the photovoltaic power interval prediction method introduced in the method embodiment of the present invention.

[0067] That is to say, the method in the above method embodiment should be understood as a process of implementing the photovoltaic power interval prediction method by computer program instructions. These computer program instructions can be provided to a processor so that the processor executes these instructions to generate functions specified in the above method flow.

[0068] The processor may be a microprocessor MCU, a programmable logic device FPGA or other processing device.

[0069] The memory may be various memories that use electrical energy to store information, such as RAM, ROM, etc.; it may also be various memories that use magnetic energy to store information, such as hard disks, floppy disks, magnetic tapes, magnetic core memories, bubble memories, USB flash drives, etc.; it may also be various memories that use optical methods to store information, such as CDs, DVDs, etc.; of course, it may also be other types of memories, such as quantum memories, graphene memories, etc.

[0070] Specific implementation methods are given above, but the present invention is not limited to the described implementation methods. The basic idea of ​​the present invention lies in the above basic scheme. For ordinary technicians in this field, it does not take creative work to design various deformed models, formulas, and parameters according to the teachings of the present invention. Changes, modifications, substitutions, and variations of the implementation methods without departing from the principles and spirit of the present invention still fall within the scope of protection of the present invention.

Claims

1. A photovoltaic power interval prediction method, It is characterized in that The following steps are involved: 1) Obtain historical data for a set time period before the day to be measured, the historical data including historical actual power data, corresponding power forecast data and various historical meteorological factor data, and obtain various meteorological factor data on the forecast day, and calculate the initial power forecast value of the forecast day according to the various meteorological factor data on the forecast day and the historical actual power data; 2) Divide the historical data of the set time period before the test date into the set time interval, and obtain the historical data of each time period after division to form a similar time period library; 3) Select multiple similar time periods based on the similarity between the meteorological factor data of the forecast day and the historical meteorological factor data in the similar time period database; 4) Determine the prediction error distribution function based on the actual power data and power forecast data in multiple selected similar time periods, obtain the photovoltaic power generation prediction error confidence interval based on the prediction error distribution function and the set significance level, and obtain the interval prediction result of photovoltaic power generation from the initial power generation prediction value and the prediction error confidence interval.

2. The photovoltaic power interval prediction method according to claim 1, It is characterized in that When determining the confidence interval of the photovoltaic power generation prediction error, the prediction error level is determined according to the prediction error and the startup capacity corresponding to similar time periods, the prediction error level distribution is obtained using the prediction error level, and then the error level confidence interval is obtained, and the prediction error confidence interval is calculated according to the startup capacity on the prediction day and the error level confidence interval.

3. The photovoltaic power interval prediction method according to claim 2, It is characterized in that The confidence interval of the photovoltaic power generation forecast error is the product of the startup capacity on the forecast day and the confidence interval of the forecast error level.

4. The photovoltaic power interval prediction method according to claim 1, It is characterized in that The historical data of meteorological factors and the meteorological factors of the forecast day are obtained after screening using the Pearson correlation coefficient.

5. The photovoltaic power interval prediction method according to claim 4, It is characterized in that The meteorological factor data include at least two of radiation, sunshine time, average wind speed, average temperature, relative humidity and atmospheric transparency.

6. The photovoltaic power interval prediction method according to claim 1, It is characterized in that The process of determining similar time periods is as follows: the meteorological factor data of the forecast day is used to calculate the similarity with the meteorological factor data of each time period in the similar time period library; the similarity is obtained based on the meteorological factor data of each time period in the similar time period library, the meteorological factor data of the forecast day and the weight of each meteorological factor data.

7. The photovoltaic power interval prediction method according to claim 6, It is characterized in that The similarity is cosine similarity.

8. The photovoltaic power interval prediction method according to claim 6, It is characterized in that The weight of each meteorological factor data is obtained using the Pearson correlation coefficient of each meteorological factor data in the similar time period library.

9. A photovoltaic power interval prediction system, comprising a processor, It is characterized in that The processor is used to execute instructions to implement the photovoltaic power interval prediction method as described in any one of claims 1 to 8.

Images