Ultra-short-term photovoltaic power prediction method based on self-adaptive two-dimensional modeling and closed-loop correction

Abstract

The invention belongs to the technical field of photovoltaic power prediction methods, and particularly relates to an ultra-short-term photovoltaic power prediction method based on adaptive two-dimensional modeling and closed-loop correction, which comprises the following steps of: acquiring multi-dimensional data and preprocessing features, acquiring static parameters and historical operation data of a photovoltaic power station, and normalizing the static parameters and the historical operation data; constructing a time period and irradiation two-dimensional dynamic K coefficient matrix, dividing time period and irradiation intensity clustering intervals, and calculating corresponding K coefficients; calculating the initial prediction generation power of the prediction day based on the K coefficient matrix; introducing proportion-integral closed-loop deviation compensation, and calculating a deviation value, an accumulated deviation value and a unit compensation coefficient of a day before the prediction day; and in combination with the unit compensation coefficient and the initial predicted power generation power, applying non-physical constraint to obtain final predicted power generation power.

Description

Technical Field

[0001] This invention belongs to the technical field of photovoltaic power prediction methods, and particularly relates to an ultra-short-term photovoltaic power prediction method with adaptive two-dimensional modeling and closed-loop correction. Background Technology

[0002] Ultra-short-term photovoltaic (PV) power forecasting is a core supporting technology for PV power plant grid-connected dispatch, electricity market trading, and renewable energy consumption. Its forecasting accuracy directly impacts the stability and economic efficiency of the power grid. PV output is influenced by multiple environmental factors, including solar irradiance, ambient temperature, and solar incidence angle, exhibiting strong volatility, randomness, and intermittency. Especially within ultra-short-term timeframes (minutes), rapid changes in irradiance lead to drastic fluctuations in PV power, posing a significant challenge to accurate forecasting. Existing forecasting methods often build models based on a single dimension, focusing only on time period divisions or considering the influence of irradiance alone. These methods fail to fully exploit the coupling relationship between time period and irradiance, resulting in insufficient adaptability of models to complex and changing environmental scenarios. Key parameters such as the K-coefficient are difficult to dynamically match actual operating conditions, leading to significant forecasting errors.

[0003] Traditional forecasting methods generally lack effective closed-loop bias compensation mechanisms, relying solely on historical data for static modeling. This makes them unable to correct systematic and random fluctuation errors generated during the forecasting process in real time. While some methods introduce bias correction, they often employ single proportional adjustments or simple historical bias averaging, failing to balance rapid response to short-term fluctuations with the cumulative offsetting of long-term systematic errors. Furthermore, they do not adequately consider the impact of differences in installed capacity of different power plants on the universality of the forecasting model, and their normalization methods are imperfect. This results in poor model transferability across different types of photovoltaic power plants, making it difficult to meet the comprehensive requirements of forecasting accuracy, adaptability, and universality in practical engineering projects. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned technical problems by providing an ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction.

[0005] In view of this, the present invention provides an ultra-short-term photovoltaic power prediction method with adaptive two-dimensional modeling and closed-loop correction, comprising the following steps:

[0006] Step 1: Multidimensional data acquisition and feature preprocessing, obtaining static parameters and historical operating data of the photovoltaic power station and performing normalization processing;

[0007] Step 2: Construct a dynamic K-coefficient matrix with two dimensions: time period and irradiance, divide the time period and irradiance intensity into cluster intervals and calculate the corresponding K-coefficients;

[0008] Step 3: Calculate the initial predicted power generation for the prediction day based on the K coefficient matrix;

[0009] Step 4: Introduce proportional-integral closed-loop deviation compensation, and calculate the deviation amount, cumulative deviation amount, and unit compensation coefficient for the day before the forecast date.

[0010] Step 5: Combining the unit compensation coefficient and the initial predicted power generation, apply non-physical constraints to obtain the final predicted power generation.

[0011] Preferably, the normalization process in step one includes calculating the actual power generation per unit area. The calculation formula is:

[0012] ;

[0013] in, This represents the absolute value of the actual power generation.

[0014] This refers to the installed capacity on the DC side.

[0015] Preferably, the irradiance intensity clustering interval in step two is divided into four continuous intervals, namely:

[0016] [0,500]W / m², [501,1000]W / m², [1001,1500]W / m², [1501,2000]W / m².

[0017] The ultra-short-term photovoltaic power prediction method based on adaptive dual-dimensional modeling and closed-loop correction according to claim 1 is characterized in that: the formula for calculating the K coefficient in step two is:

[0018] ;

[0019] in, That is, only historical data where the irradiance intensity falls within the corresponding interval are statistically analyzed;

[0020] This represents the sum of the actual power generated per unit area in history under this scenario;

[0021] This represents the sum of historically measured irradiance in this scenario.

[0022] Preferably, the deviation in step four The calculation formula is:

[0023] ;

[0024] in, This represents the average actual power generation for all minutes within the irradiance range during this period on day D-1.

[0025] This represents the average of the predicted power generation after minute corrections for this scenario, with all data falling within the corresponding irradiance range. .

[0026] Preferably, the cumulative deviation in step four Cumulative bias is the accumulation of historical biases, used to capture the persistence of prediction errors. The calculation formula is as follows:

[0027] ;

[0028] in, This represents the cumulative deviation for the corresponding time period and irradiation interval scenario on day D-2.

[0029] This represents the deviation for this scenario on day D-1.

[0030] Preferably, the unit compensation coefficient in step four Combining proportional and integral terms, the unit compensation coefficient for this scenario is calculated to achieve dual correction for short-term fluctuations and long-term systematic errors. The calculation formula is as follows:

[0031] ;

[0032] The data belongs to the interval R. The preset proportional and integral correction coefficients can be finely adjusted according to the characteristics of the power plant;

[0033] This represents the average of all minute-level initial predicted power generation for this scenario on day D-1.

[0034] Preferably, the formula for calculating the predicted power generation after unit scale correction in step five is as follows:

[0035] ; .

[0036] Preferably, the formula for calculating the final predicted power generation in step five is:

[0037] ;

[0038] The beneficial effects of this invention are as follows: By constructing a dynamic K-coefficient matrix with two dimensions of time period and irradiance, this invention divides a day into 24-hour time periods and four continuous irradiance intensity intervals, accurately capturing the coupling influence of factors such as temperature and solar incidence angle on photovoltaic output under different scenarios, enabling the K-coefficient to dynamically match the actual operating state. Combined with normalization processing, it eliminates the differences in installed capacity of different power plants, significantly improving the initial prediction accuracy and the model's ability to transfer across power plants, and solving the problem of insufficient adaptability of traditional single-dimensional modeling.

[0039] By introducing a proportional-integral closed-loop deviation compensation mechanism, the unit compensation coefficient is obtained by calculating the deviation amount of the day before the forecast date and the cumulative deviation amount, combined with the preset correction coefficient. This enables a rapid response to short-term fluctuation errors and effective offsetting of long-term system deviations. Furthermore, non-physical constraints ensure that the predicted power is non-negative, ultimately significantly improving the accuracy and stability of ultra-short-term photovoltaic power forecasting. This provides reliable data support for grid connection dispatch and power market transactions, and helps to efficiently absorb new energy. Attached Figure Description

[0040] Figure 1 This is the photovoltaic power generation prediction and actual curve of this invention;

[0041] Figure 2 This is a flowchart illustrating the predicted power of this invention.

[0042] Figure 3 This is a flowchart of the K coefficient calculation process of the present invention;

[0043] Figure 4 This is a flowchart of the compensation coefficient of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0045] In this invention, normalization is a process to eliminate differences in the installed capacity of different photovoltaic power plants. The core of this process is to calculate the actual power generation per unit size so that the model can be adapted to power plants of different specifications.

[0046] Two-dimensional dynamic K-coefficient matrix: The coefficient matrix is ​​dynamically calculated with a 24-hour time period and four irradiance intensity intervals as two dimensions. The K coefficient represents the average power generation corresponding to a unit irradiance intensity.

[0047] Proportional-integral closed-loop deviation compensation: This mechanism combines a proportional term (for rapid response to short-term fluctuation errors) and an integral term (to offset long-term systematic errors) to correct the prediction results through feedback of the previous day's deviation.

[0048] Unit compensation coefficient: A correction coefficient calculated based on the previous day's deviation and cumulative deviation, used to adjust the initial forecast power and improve forecast accuracy.

[0049] Non-physical constraints: Ensure that the prediction results meet the actual constraints (such as photovoltaic power cannot be negative) and avoid prediction values ​​that do not conform to physical laws.

[0050] The photovoltaic power generation prediction method proposed in this invention specifically includes the following steps;

[0051] Step S1: Multidimensional data acquisition and feature preprocessing;

[0052] S1.1: Obtain the static parameters of the photovoltaic power plant: DC side installed capacity (unit: This parameter is an inherent configuration parameter of the power plant and is used to subsequently eliminate the impact of installed capacity on the generality of the prediction model.

[0053] S1.2: Obtain historical operating data: Collect historical minute-level operating data of the photovoltaic power station, including actual power generation. (Take the absolute value, unit: ) and measured irradiance (unit: The data collection frequency is 1 minute per data point, with a total of 1440 data points per day.

[0054] S1.3: Data Normalization Process: To eliminate the impact of differences in installed capacity of different power plants on the prediction model, the actual power generation per unit area is calculated. (unit: The calculation formula is as follows:

[0055] ;

[0056] in, This represents the absolute value of the actual power generation. This refers to the installed capacity on the DC side.

[0057] Step S2: Construct a dynamic K-coefficient matrix with a two-dimensional "time period-irradiation" relationship;

[0058] The K-coefficient is used to characterize the average power generation per unit of irradiance within a specific time period and irradiance range. It comprehensively reflects the impact of factors such as temperature and solar incidence angle on photovoltaic output. Figure 3 The flowchart for calculating the K coefficient is shown below. The specific construction process is as follows:

[0059] S2.1: Time Segmentation: Divide the day into 24 consecutive hourly time segments (e.g., 00-01, 01-02, ..., 23-24), with each time segment lasting 1 hour.

[0060] S2.2: Irradiance Intensity Clustering Interval Division: For each hourly time period, four consecutive irradiance intensity intervals are divided, as follows:

[0061] Interval 1 ( coefficient): ;

[0062] Interval 2 ( coefficient): ;

[0063] Interval 3 ( coefficient): ;

[0064] Interval 4 ( coefficient): ;

[0065] S2.3: Calculate the K coefficient: for each hourly time period Each irradiation intensity range Based on historical normalized data, calculate the K coefficient for this combined scenario. The calculation formula is as follows:

[0066] ;

[0067] in, That is, only historical data where the irradiance intensity falls within the corresponding interval are statistically analyzed; This represents the sum of the actual power generated per unit area in history under this scenario. This represents the sum of historically measured irradiance in this scenario.

[0068] The K coefficient is essentially the average power generation per unit of irradiance within a specific hour and irradiance interval, comprehensively reflecting the influence of factors such as temperature and irradiance angle during that period.

[0069] Step S3: Initial predicted power calculation (for prediction date D);

[0070] For any minute of day D Calculate the initial predicted power generation using the following steps:

[0071] S3.1: Determine the prediction time parameters: Obtain the time of that minute on day D. Measured radiation intensity (unit: ), and determine the hour period to which that moment belongs.

[0072] S3.2: Matching K coefficient: based on the hourly time period to which the current moment belongs. and measured irradiance The interval to which it belongs Find the corresponding K coefficient from the dynamic K coefficient matrix .

[0073] S3.3: Calculate the initial predicted power generation:

[0074] First, calculate the initial predicted power generation per unit area. (unit: ):

[0075] ;

[0076] Then, combining the DC-side installed capacity, the initial predicted power is calculated. (unit: ):

[0077] ;

[0078] Step S4: Introduce proportional-integral (PI) closed-loop bias compensation, such as... Figure 4 As shown.

[0079] To eliminate systematic forecast bias, bias feedback from day D-1 (the day before the forecast date) is introduced to construct a PI closed-loop compensation mechanism. The specific process is as follows:

[0080] S4.1: Calculate the deviation For each hourly period on day D-1 and each irradiation zone The average deviation between the predicted and actual values ​​in this scenario is calculated using the following formula:

[0081] ;

[0082] in, This represents the average actual power generation for all minutes within the "time period-irradiance interval" scenario on day D-1. This represents the average of the predicted power generation after minute corrections for this scenario, with all data falling within the corresponding irradiance range. .

[0083] S4.2: Calculate the cumulative deviation Cumulative bias is the accumulation of historical biases, used to capture the persistence of prediction errors. The calculation formula is as follows:

[0084] ;

[0085] in, This represents the cumulative deviation under the "time period - irradiation interval" scenario corresponding to day D-2 (two days before the forecast date). This represents the deviation for this scenario on day D-1.

[0086] S4.3: Calculate the unit compensation coefficient Combining proportional and integral terms, the unit compensation coefficient for this scenario is calculated to achieve dual correction for short-term fluctuations and long-term systematic errors. The calculation formula is as follows:

[0087] ;

[0088] The data belongs to the interval R. The preset proportional and integral correction factors can be fine-tuned according to the characteristics of the power plant. This represents the average of all minute-level initial predicted power generation for this scenario on day D-1.

[0089] Step S5: Correct the predicted power calculation (for prediction date D);

[0090] Based on the initial predicted power, a unit compensation coefficient is introduced for correction, and a non-physical constraint (power generation cannot be negative) is applied to obtain the final predicted power generation. The specific calculation formula is as follows:

[0091] First, calculate the predicted power generation after unit size correction. (unit: ):

[0092] ;

[0093] Then, combining the DC-side installed capacity, the initial predicted power is calculated. (unit: ):

[0094] ;

[0095] The max() function is used to ensure that the predicted power is non-negative, which is consistent with the physical characteristics of photovoltaic power generation.

[0096] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction, characterized by: Includes the following steps: Step 1: Multidimensional data acquisition and feature preprocessing, obtaining static parameters and historical operating data of the photovoltaic power station and performing normalization processing; Step 2: Construct a dynamic K-coefficient matrix with two dimensions: time period and irradiance, divide the time period and irradiance intensity into cluster intervals and calculate the corresponding K-coefficients; Step 3: Calculate the initial predicted power generation for the prediction day based on the K coefficient matrix; Step 4: Introduce proportional-integral closed-loop deviation compensation, and calculate the deviation amount, cumulative deviation amount, and unit compensation coefficient for the day before the forecast date. Step 5: Combining the unit compensation coefficient and the initial predicted power generation, apply non-physical constraints to obtain the final predicted power generation.

2. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: The normalization process in step one includes calculating the actual power generation per unit area. The calculation formula is: ； in, This represents the absolute value of the actual power generation. This refers to the installed capacity on the DC side.

3. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: Step two divides the irradiance intensity clustering intervals into four continuous intervals, namely: [0,500]W / m², [501,1000]W / m², [1001,1500]W / m², [1501,2000]W / m².

4. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: The formula for calculating the K coefficient in step two is as follows: ； in, That is, only historical data where the irradiance intensity falls within the corresponding interval are statistically analyzed; This represents the sum of the actual power generated per unit area in history under this scenario; This represents the sum of historically measured irradiance in this scenario.

5. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: Deviation in step four The calculation formula is: ； in, This represents the average actual power generation for all minutes within the irradiance range during this period on day D-1. This represents the average of the predicted power generation after minute corrections for this scenario, with all data falling within the corresponding irradiance range. .

6. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 5, characterized in that: Cumulative deviation in step four Cumulative bias is the accumulation of historical biases, used to capture the persistence of prediction errors. The calculation formula is as follows: ； in, This represents the cumulative deviation for the corresponding time period and irradiation interval scenario on day D-2. This represents the deviation for this scenario on day D-1.

7. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: Step 4 Unit compensation coefficient Combining proportional and integral terms, the unit compensation coefficient for this scenario is calculated to achieve dual correction for short-term fluctuations and long-term systematic errors. The calculation formula is as follows: ； The data belongs to the interval R. The preset proportional and integral correction coefficients can be finely adjusted according to the characteristics of the power plant; This represents the average of all minute-level initial predicted power generation for this scenario on day D-1.

8. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: The formula for calculating the predicted power generation after unit scale correction in step five is as follows: 。 9. The ultra-short-term photovoltaic power prediction method with adaptive dual-dimensional modeling and closed-loop correction according to claim 1, characterized in that: The formula for calculating the final predicted power generation in step five is as follows: 。

Images