Wind-solar power generation base power self-correction prediction method and device, electronic equipment and medium

Abstract

The invention discloses a wind and light power generation base power self-correction prediction method and device, electronic equipment and a medium, relates to the technical field of new energy power generation power prediction, and aims at effectively identifying extreme scenes with relatively large prediction errors by screening wind and light high prediction error days and determining key factors influencing prediction precision. And the key factors can obtain higher training weights in the model training process. The method comprises the following steps: selecting a plurality of sample days, and evaluating wind and light power generation data of each sample day to screen and obtain a wind and light height prediction error day; performing power analysis on the wind-solar power generation data of the wind-solar high prediction error day, and determining key factors influencing the prediction precision; constructing a weight index according to the key factors, and optimizing model parameters in combination with the weight index in a process of training a wind-solar power prediction model by using the wind-solar power generation data of the plurality of sample days to obtain the wind-solar power prediction model; and performing power prediction under the condition that the wind-solar power prediction demand is generated.

Description

Technical Field

[0001] This application relates to the field of new energy power generation prediction technology, and in particular to a self-correcting prediction method, device, electronic equipment and medium for wind and solar power generation bases. Background Technology

[0002] With the transformation of the global energy structure, the installed capacity of renewable energy is growing rapidly, and the proportion of new energy in the energy system is increasing. Among them, wind and solar power, as an important component of renewable energy, has been developed and utilized on a large scale due to its abundant resources and wide distribution. However, the output of wind and solar power is significantly intermittent and fluctuating, and its power output is affected by various factors such as meteorological conditions and geographical environment. As the penetration rate of new energy continues to rise, the accuracy of power forecasting for wind and solar power bases is crucial for the safe and stable operation of the power system, the orderly trading of the electricity market, and the efficient consumption of new energy. Improving the accuracy of power forecasting for wind and solar power bases can effectively reduce the operating costs of the power system and improve the utilization efficiency of new energy, which is a key link in achieving sustainable energy development.

[0003] In related technologies, some research focuses on building advanced prediction models. By introducing deep learning algorithms and signal processing techniques, historical data from wind and solar power plants is analyzed and mined to extract key features affecting power output, thereby establishing prediction models to forecast wind and solar power. Other studies attempt to use multi-model fusion methods, combining prediction models of different types and characteristics to fully leverage the advantages of each model and improve overall prediction accuracy. Furthermore, some research focuses on error correction of prediction results. By analyzing the distribution patterns and influencing factors of prediction errors, corresponding correction strategies are formulated to revise the initial prediction results, thereby improving prediction accuracy.

[0004] However, the applicant recognizes that, on the one hand, existing prediction models are poorly adapted to days with low power output and high fluctuations. In cold regions, frequent rainy weather in summer and windless conditions in winter are common. These special weather conditions can lead to a significant decrease and drastic fluctuation in wind and solar power output. Traditional prediction models often fail to accurately capture the changing patterns of wind and solar power output in these scenarios, resulting in a significant increase in prediction errors. On the other hand, the wind and solar power output curves and fluctuation characteristics differ on different dates, and existing models fail to accurately adjust the prediction results according to these differences, resulting in limited prediction accuracy. Furthermore, there are some typical days with large errors during the training process, leading to a large overall deviation in the prediction results, which cannot meet the requirements of high-precision power prediction and is difficult to meet the needs of practical engineering applications. Summary of the Invention

[0005] In view of this, this application provides a method, device, electronic device and medium for self-correcting prediction of power in wind and solar power generation bases. The main purpose is to solve the problems that existing models fail to accurately adjust the prediction results according to these differences, resulting in limited prediction accuracy. Furthermore, there are some typical days with large errors during the training process, which makes the overall prediction results have a large deviation and cannot meet the requirements of high-precision power prediction, thus failing to meet the needs of practical engineering applications.

[0006] According to the first aspect of this application, a self-correcting prediction method for power generation at wind and solar power bases is provided, the method comprising: Multiple sample days are selected, and the wind and solar power generation data for each sample day are evaluated to screen out days with high prediction errors for wind and solar power from the multiple sample days. Power analysis was performed on the wind and solar power generation data of the days with high prediction errors to determine the key factors affecting prediction accuracy. Based on the key factors, a weight index is constructed, and the model parameters are optimized in combination with the weight index during the process of training the wind and solar power prediction model using the wind and solar power generation data of the multiple sample days, so as to train the wind and solar power prediction model. When there is a demand for wind and solar power prediction, the aforementioned wind and solar power prediction model is used to predict power.

[0007] According to a second aspect of this application, a power self-correction prediction device for wind and solar power generation bases is provided, the device comprising: The error day screening module is used to select multiple sample days, evaluate the wind and solar power generation data of each sample day, and screen out the wind and solar high prediction error days from the multiple sample days. The key factor identification module is used to analyze the power parameters of the days with high prediction errors in wind and solar power, and to determine the key factors affecting the prediction accuracy. The weighted training module is used to construct weight indicators based on the key factors, and to optimize the model parameters in combination with the weight indicators during the process of training the wind and solar power prediction model using wind and solar power generation data from the multiple sample days, so as to train the wind and solar power prediction model. The prediction module is used to perform power prediction using the wind and solar power prediction model when a demand for wind and solar power prediction is generated.

[0008] According to a third aspect of this application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the first aspects above.

[0009] According to a fourth aspect of this application, a medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the first aspects above.

[0010] By employing the above technical solution, this application provides a method, device, electronic equipment, and medium for self-correcting power prediction of wind and solar power generation bases. This application selects multiple sample days and evaluates the wind and solar power generation data for each sample day to identify days with high prediction errors. Power analysis is performed on the wind and solar power generation data of these days to determine the key factors affecting prediction accuracy. Based on these key factors, weighting indicators are constructed. During the training of the wind and solar power prediction model using the wind and solar power generation data from multiple sample days, the model parameters are optimized using these weighting indicators to obtain the desired wind and solar power prediction model. When a wind and solar power prediction demand arises, the model is used for power prediction. By selecting days with high prediction errors, extreme scenarios with large prediction errors are effectively identified. Furthermore, by analyzing the power parameters, the key factors affecting prediction accuracy are clarified. Weighting indicators are synthesized based on these key factors, allowing them to obtain higher training weights during model training. This effectively improves the prediction performance of the trained wind and solar power prediction model in key scenarios, enhancing the overall accuracy and reliability of wind and solar power prediction. This provides a more accurate decision-making basis for grid dispatch and meets the needs of practical engineering applications.

[0011] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0012] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This paper illustrates a flowchart of a power self-correction prediction method for wind and solar power generation bases provided in an embodiment of this application. Figure 2 This paper illustrates a flowchart of another wind and solar power base power self-correction prediction method provided in an embodiment of this application. Figure 3 This illustration shows a structural schematic diagram of a wind and solar power generation base power self-correction prediction device provided in an embodiment of this application; Figure 4A schematic diagram of the device structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0013] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0014] This application provides a method for self-correcting prediction of power output at wind and solar power generation bases, such as... Figure 1 As shown, the method includes: S10: Select multiple sample days and evaluate the wind and solar power generation data for each sample day in order to screen out the days with high prediction errors for wind and solar power among the multiple sample days.

[0015] In this embodiment, firstly, multiple dates under different seasons and weather conditions are selected from the historical data storage system of the wind and solar power generation base as sample days. These sample days cover various typical operating conditions of wind and solar power generation. Then, the wind and solar power generation data for each sample day is evaluated. The wind and solar power generation data includes power curves, from which the actual power generation and the corresponding predicted power generation at each time point can be obtained. By evaluating the wind and solar power generation data for each sample day, days with high prediction errors can be screened from multiple sample days. Days with high prediction errors are extreme scenarios with large errors in predicted power, providing a targeted data foundation for subsequent analysis of the causes of prediction errors and optimization of the model, which helps to improve the prediction performance of the model in key scenarios.

[0016] For example, a wind and solar power generation base selected the 15th of each month over the past year, as well as dates corresponding to certain special weather events (such as heavy rain, strong winds, and intense sunlight), as sample days, totaling 50 sample days. By calculating the prediction error for each sample day, it was found that the errors on 8 of these dates were too large, and these 8 dates were selected as days with high prediction errors for wind and solar power.

[0017] In step S10, which involves selecting multiple sample days and evaluating the wind and solar power generation data for each sample day to identify days with high prediction errors for wind and solar power generation, the following steps are taken: S11: Select multiple sample days, obtain the wind and solar power generation data for each sample day, and preprocess the wind and solar power generation data for each sample day.

[0018] In this embodiment, multiple dates under different time periods and weather conditions are selected from the historical data records of the wind and solar power generation base as sample days. For each sample day, corresponding wind and solar power generation data is collected, which includes the raw wind and solar power generation power data and meteorological data for the corresponding sample day. The raw wind and solar power generation power data records the actual power output of wind and solar power generation on that date, and the meteorological data includes information such as wind speed and solar irradiance.

[0019] Subsequently, the wind and solar power generation data for each sample day were preprocessed, as follows: First, a data quality check is performed, systematically scanning the wind and solar power generation data for each sample day to assess its completeness and consistency, checking for missing data, incorrect formatting, and other issues. Then, outlier detection and removal are performed using a physical threshold method. Based on the physical operating characteristics of the wind and solar power generation equipment and the physical limits of natural conditions, an absolute threshold range is set. For example, regarding power constraints, for wind power plants, the conditions shown in Formula 1 below must be met: Formula 1:

[0020] In Formula 1 Indicates the rated power of wind power. This indicates the power at each time point in the wind and solar power generation data that needs to be determined.

[0021] For wind power data, the following conditions, as shown in Formula 2, must be met: Formula 2:

[0022] In Formula 2 Indicates real-time wind speed. Indicates the cut-in wind speed. This indicates the cut-out wind speed.

[0023] For photovoltaic data, the conditions shown in Formula 3 below must be met: Formula 3:

[0024] in, Indicates real-time irradiance. This indicates the highest historical irradiance in the area.

[0025] In practical applications, if data points in wind and solar power generation data exceed the ranges set in the above conditions, they are judged as outliers and directly removed.

[0026] Next, a time alignment process is performed to resample all data to the same fixed time interval (e.g., 15 minutes). For power data, forward padding or linear interpolation is used, and for meteorological data, linear interpolation is used. The timestamp of one data source (usually power data) is used as the reference, and the data from the other data source is matched and aligned by finding the nearest timestamp, ensuring that each power data point has its corresponding meteorological data point.

[0027] In this way, comprehensive data collection and meticulous preprocessing through the above process can effectively solve the quality problems of the original data, remove outliers, and align different types of data in time, thereby obtaining a high-quality, highly consistent, and time-synchronized standardized dataset. This provides a reliable data foundation for subsequent accurate analysis of prediction errors and model training. For example, a wind and solar power generation base selected 30 dates from different seasons and weather conditions over the past six months as sample days. After collecting the original wind and solar power generation data and meteorological data for each sample day, some missing data was found during data quality checks, which were repaired through forward imputation. In outlier detection, a small number of outlier data points exceeding the set thresholds for wind speed, power, and irradiance were removed, and time alignment was completed.

[0028] S12: Calculate the normalized mean absolute error for each sample day using the following formula 4.

[0029] Formula 4:

[0030] in, This represents the calculated normalized mean absolute error. Indicates the first The actual power value for each sample day can be obtained from the preprocessed wind and solar power generation data; This represents the maximum value among the true power values ​​for multiple sample days. Indicates the first Predicted power values ​​for each sample day; Indicates the first The total number of sampling points on a given sample day is calculated by resampling all data to the same fixed time interval during data preprocessing. Therefore, the total number of sampling points divided according to this fixed time interval within a 24-hour day is... For example, if a sampling interval is 15 minutes, and a 24-hour day contains a total of 96 15-minute intervals, then... It is 96.

[0031] Specifically, by substituting the data corresponding to each sample day into Formula 4, the normalized mean absolute error for each sample day is calculated to measure the absolute average of the prediction error. This method is insensitive to outliers and robustly reflects the overall deviation level for each sample day. For example, for one of the 30 sample days mentioned above, the true power value sequence for that day is obtained from the preprocessed data, and combined with the corresponding predicted power value sequence, the prediction error is determined. Then, the sample day's... It is 0.15.

[0032] S13: Calculate the normalized root mean square error for each sample day using the following formula 5.

[0033] Formula 5:

[0034] in, This represents the calculated normalized root mean square error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​for multiple sample days. Indicates the first The predicted power value for the nth sample day represents the power value for the nth sample day. The total number of sampling points on each sample day.

[0035] Calculating the normalized root mean square error (RMSE) allows for a more sensitive capture of the prediction error distribution, facilitating a more detailed understanding of its characteristics. This is particularly beneficial for assessing extreme error scenarios, providing crucial information for a comprehensive evaluation of prediction errors. For example, for the same sample day, the above formula, combined with existing actual power values ​​and predicted power values, can be used... and Calculate the sample day It is 0.18.

[0036] S14: Using the normalized mean absolute error and normalized mean absolute error of each sample day, construct a corresponding comprehensive error evaluation index for each sample day.

[0037] In the prediction error evaluation of this application embodiment, the mean absolute error (MAE) and root mean square error (RMSE) have different statistical characteristics and can reflect prediction accuracy from different perspectives. The MAE measures the absolute average of the prediction error; it is insensitive to outliers and can robustly reflect the overall bias level of the prediction model. The RMSE, on the other hand, amplifies the impact of larger errors through squaring, and can more sensitively capture the distribution of prediction errors, especially exhibiting a higher penalty for extreme error values. Therefore, to comprehensively utilize the advantages of these two error indicators, this application embodiment uses the normalized mean absolute error (MSE) and the normalized mean absolute error (RMSE) for each sample day to construct a corresponding comprehensive error evaluation index for each sample day. The specific process is as follows: For each sample day, the corresponding comprehensive error evaluation index is calculated using the following formula 6. Formula 6:

[0038] in, Represents the calculated first... The comprehensive error evaluation index corresponding to each sample day and This represents the preset weighting coefficient. and The value can be 0.5. Indicates the first Normalized mean absolute error for each sample day Indicates the first The normalized mean absolute error for each sample day. In practical applications, the weighting coefficients can be adjusted according to specific needs; if more attention is paid to the impact of extreme errors, they can be appropriately increased. If greater emphasis is placed on overall forecast stability, then the value can be appropriately increased. .

[0039] In this way, by constructing a comprehensive error assessment index, we can more comprehensively and accurately evaluate the prediction error of each sample day, providing a more scientific basis for selecting days with high prediction errors in wind and solar forecasts. For example, for the previously calculated... and For the sample day, calculate the comprehensive error evaluation index for that sample day according to Formula 6. It is 0.165.

[0040] S15: Based on the comprehensive error evaluation index corresponding to each sample day, sort the multiple sample days in descending order to obtain the sorting results.

[0041] In this embodiment, based on the comprehensive error evaluation index corresponding to each sample day, a sorting algorithm (such as bubble sort, quick sort, etc.) is used to sort multiple sample days in descending order. Sample days with larger comprehensive error evaluation index values ​​are ranked first, and those with smaller values ​​are ranked last, thus highlighting sample days with larger prediction errors. This facilitates quick location of extreme scenarios with large prediction errors and makes it easier to subsequently filter out days with high prediction errors in wind and solar power.

[0042] S16: Select the sample days ranked in the top preset position from the sorting results as the days with high prediction error for wind and solar power.

[0043] In this embodiment, the top m dates in the ranking results are selected as days with high prediction errors for wind and solar power, where m can be the top 20% of the total days. The selected days with high prediction errors represent abnormal scenarios with large prediction errors. By selecting these dates, we can focus on scenarios with large prediction problems for in-depth analysis, providing a basic sample set for subsequent error analysis and model correction. This helps to improve the model's prediction performance in key scenarios, thereby improving the overall accuracy and reliability of wind and solar power prediction.

[0044] For example, the top 20% of the 30 sample days are 6 sample days. The 6 sample days with the largest comprehensive error evaluation index after sorting are selected as the high prediction error days for wind and solar power, and used for subsequent model optimization analysis.

[0045] S20: Perform power analysis on wind and solar power generation data for days with high prediction errors to determine the key factors affecting prediction accuracy.

[0046] In this embodiment, for the selected days with high prediction errors in wind and solar power generation, power analysis is performed on the wind and solar power generation data of these days to quantitatively analyze the relationship between power parameters and prediction errors. Based on the analysis results, factors that significantly impact prediction errors are identified as key factors affecting prediction accuracy. This allows for a quantitative analysis of the causes of prediction errors, clearly identifying which factors have the greatest impact on wind and solar power generation prediction accuracy. This provides a scientific basis for subsequent construction of weighting indicators and model optimization, enabling targeted model optimization that adjusts to key factors and improves the model's prediction accuracy.

[0047] For example, an analysis of the eight selected days with high wind and solar power prediction errors revealed a strong correlation between the average daily power output and the prediction error. Therefore, the average daily power output was determined to be a key factor affecting prediction accuracy.

[0048] In step S20, power analysis is performed on the wind and solar power generation data for days with high prediction errors to determine the key factors affecting prediction accuracy, including: S21: The following formula is used to calculate the power output of wind and solar power generation data on days with high prediction errors, so as to obtain the average daily output value on days with high prediction errors.

[0049] In this embodiment, for the selected days with high prediction errors in wind and solar power generation, power curves are extracted from these days. Power calculations are performed on the wind and solar power generation data for these days to conduct multi-dimensional feature extraction and quantitative analysis of the power curves, establishing a complete power characteristic evaluation system and obtaining the average daily output value for these days. The average daily output value is a core indicator for measuring the daily power generation capacity of the power generation system, comprehensively considering the power output throughout the day. The specific calculation formula is shown in Formula 7 below. Formula 7:

[0050] in, This represents the calculated average daily output value. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value of the day with the highest prediction error for wind and solar power.

[0051] In this way, by accurately calculating the average daily power output on days with high prediction errors in wind and solar power generation, we can grasp the average power generation level of the wind and solar power system on these special dates as a whole. This provides basic data for subsequent analysis of factors affecting prediction accuracy and helps to more comprehensively understand the actual operating characteristics of wind and solar power generation under scenarios with large prediction errors. For example, a wind and solar power base has identified five days with high prediction errors. For one of these dates, the total number of sampling points on that day is known. =96 (with a sampling interval of 15 minutes), collect the actual power values ​​of each sampling point on the same day. Substituting this into Formula 7, the average daily output for that day is calculated to be 1500kW.

[0052] S22: The average daily output value is taken as a key factor affecting the accuracy of the forecast.

[0053] In this embodiment, the calculated average daily power output is included as a key consideration when analyzing factors affecting prediction accuracy. Since the power output of wind and solar power generation is affected by various factors, and the average daily power output reflects the overall power generation capacity of the power generation system on a specific date, it may be closely related to prediction error. For example, when the actual power generation capacity differs significantly from the conventional power generation mode on which the prediction model is based, the prediction error may increase. Therefore, the average daily power output is used as a key factor, and a weighting index can be synthesized based on it to give features related to the average daily power output higher training weights in parameter updates and other aspects during model training. The weighting index is used to measure the importance of different factors in model training. By giving higher weights to key factors, the model can pay more attention to these factors that have a greater impact on prediction accuracy.

[0054] Thus, by clearly identifying the average daily power output as a key factor affecting prediction accuracy through the above process, a clear direction is provided for subsequent synthesis of weight indicators based on key factors. This allows for more targeted optimization of model training, improving the model's prediction performance for wind and solar power generation in key scenarios, thereby enhancing overall prediction accuracy and reliability, and better meeting the needs of practical engineering applications such as grid dispatching. Continuing with the example of the five days with high prediction errors for wind and solar power, analysis revealed that the prediction errors were relatively larger on days with lower average daily power output. When constructing the weight indicators later, incorporating the average daily power output as a key factor allows the model to pay greater attention to features related to low average daily power output during training, thereby improving the model's prediction accuracy in these special scenarios.

[0055] In another alternative implementation, the Fluctuation Intensity (FI) index is an important indicator for quantifying power instability and can also be used as a core analytical indicator. Specifically, the following formula 8 can be used to calculate the power output of wind and solar power generation data on days with high prediction errors, thereby obtaining the Fluctuation Intensity Index for those days. Formula 8:

[0056] in, This represents the calculated volatility index. A larger value indicates more severe power fluctuations. This represents the average daily power output calculated for days with high forecasting error in wind and solar power. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value for days with high prediction errors in wind and solar power forecasts. Indicates the first The actual power value of the day with the highest prediction error for wind and solar power.

[0057] Furthermore, based on identifying the days with high wind and solar power prediction errors and extracting their power characteristics, statistical correlation analysis was used to quantitatively study the intrinsic relationship between daily average power output, fluctuation intensity index, and prediction errors. The specific process is as follows: First, Formula 9 is used as the first correlation coefficient between the calculated daily average power output on days with high wind and solar forecasting error and the forecasting error on those days. Formula 9:

[0058] in, This represents the first correlation coefficient, which measures the degree of linear correlation between the average daily output and the prediction error. Its value ranges from -1 to 1, with positive values ​​indicating positive correlation and negative values ​​indicating negative correlation. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The average daily power output on days with high forecast error in wind and solar power. This represents the average daily power output value for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. This represents the average value of the comprehensive error assessment index for all days with high prediction error in wind and solar power.

[0059] Meanwhile, Formula 10 below is used as the second correlation coefficient between the fluctuation intensity index and the prediction error on the day with high prediction error for wind and solar power. Formula 10:

[0060] in, This represents the second correlation coefficient, which measures the degree of linear correlation between the volatility intensity index and the prediction error. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The fluctuation intensity index of days with high prediction errors in wind and solar power. This represents the average fluctuation intensity index for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. This represents the average value of the comprehensive error assessment index for all days with high forecast error in wind and solar power. In practical applications, the analysis results obtained through the above process show that the average daily output and the prediction error are negatively correlated, while the fluctuation intensity index and the prediction error are positively correlated. Therefore, based on the first correlation coefficient, a first descriptive information can be generated to describe the negative correlation between the average daily output and the prediction error, and based on the second correlation coefficient, a second descriptive information can be generated to describe the positive correlation between the fluctuation intensity index and the prediction error. In other words, the prediction results are not ideal on low output days and high fluctuation days.

[0061] Finally, by using the fluctuation intensity index, first descriptive information, and second descriptive information of the days with high prediction errors in wind and solar power, a basis for model adjustment is generated. When the wind and solar power prediction model is trained, the model adjustment basis is used to label the wind and solar power prediction model, providing a scientific basis for the subsequent design of a targeted weighted training mechanism.

[0062] Thus, by introducing the fluctuation intensity index through the above process and using statistical correlation analysis to study its relationship with prediction error and the relationship between average daily power output and prediction error, we can gain a deeper understanding of the factors affecting wind and solar power prediction errors, identify situations where predictions are not ideal on low-output days and high-fluctuation days, and provide more targeted adjustment basis for model training. This helps to optimize the wind and solar power prediction model, improve the prediction accuracy and reliability of the model in various scenarios, and better meet the actual needs of grid dispatching. For example, a wind and solar power base selected 10 days with high prediction errors. First, the fluctuation intensity index and average daily power output of each day with high prediction errors were calculated according to the formula. Then, through statistical correlation analysis, the first correlation coefficient between the average daily power output and the comprehensive error evaluation index was calculated to be -0.6, indicating that the average daily power output and prediction error are negatively correlated; the second correlation coefficient between the fluctuation intensity index and the comprehensive error evaluation index was 0.7, indicating that the fluctuation intensity index and prediction error are positively correlated. Based on this, it is concluded that the prediction results are not ideal under the conditions of low output and high fluctuation. This information is used to generate the basis for model adjustment, and is labeled in the subsequent training of the wind and solar power prediction model so as to optimize the model in a targeted manner.

[0063] S30: Based on key factors, construct weight indicators, and optimize the model parameters by combining the weight indicators during the training of the wind and solar power prediction model using wind and solar power generation data from multiple sample days, so as to obtain the wind and solar power prediction model.

[0064] In this embodiment, weighting indicators are constructed based on the identified key factors. The construction of weighting indicators can employ various methods, such as allocating them based on the magnitude of the correlation between the key factors and the prediction error; the higher the correlation coefficient, the higher the weight allocated. Alternatively, weights can be determined by comprehensively evaluating the importance of different key factors in conjunction with expert experience. This embodiment does not impose specific limitations on these methods.

[0065] After constructing the weighting indicators, they need to be integrated into the model training algorithm during the training of the wind and solar power prediction model using wind and solar power generation data from multiple sample days. Specifically, during each model parameter update, the training contribution of different samples is weighted according to the key factor weights corresponding to the sample data. This makes the model training focus more on key samples with greater prediction difficulty, that is, those samples where key factors have a significant impact. In this way, by designing an adaptive weight allocation mechanism, the importance assessment of samples is innovatively integrated into the model training process, making the model parameter updates more inclined to optimize the prediction accuracy of key samples. This allows the model to focus on factors and scenarios that have a significant impact on the prediction results during training, thereby effectively improving the prediction performance of the trained wind and solar power prediction model in key scenarios and improving the overall prediction accuracy.

[0066] For example, the average daily power output is identified as a key factor. For each sample day, a corresponding weight index related to the average daily power output is calculated. When training the wind and solar power prediction model, sample data with large changes in the average daily power output are given higher weights when updating the model parameters. This makes the model more fully trained on these key samples, thereby improving the model's ability to predict similar key scenarios.

[0067] In step S30, based on the key factor analysis results, a weight index reflecting the importance of the samples is constructed, and higher training weights are assigned to days with low output and high fluctuations. That is, the weight index is constructed according to the key factors, including: S31: Calculate the power standard deviation for each sample day using the following formula 11.

[0068] The power standard deviation is an indicator that measures the dispersion of the daily power value relative to the daily average output value. The larger the value, the greater the fluctuation of power within that day. It can be calculated using the following formula 11: Formula 11:

[0069] in, Indicates the first The standard deviation of power for each sample day. Indicates the first The total number of sampling points on each sample day Indicates the first Sample day Power value at time, Indicates the first The average daily output value for each sample day.

[0070] By calculating the power standard deviation, the fluctuation of wind and solar power generation within each sample day can be quantitatively described, providing a crucial data foundation for the subsequent construction of weighted indicators and helping to more accurately reflect the power change characteristics of different sample days. For example, for one sample day, the total number of sampling points for that day is known. =96, power values ​​collected at various times. The average daily output value for that day has been calculated. Substitute the values ​​into Formula 11 above to calculate the power standard deviation for that sample day. It is 200kW.

[0071] S32: The power standard deviation and average daily output value for each sample day are calculated using the following formula 12 to obtain the weight index.

[0072] The weighting metric is used to measure the importance of each sample day in model training. Sample days with a large power standard deviation and relatively low average daily output are given higher weights because these days often correspond to low output and high fluctuations, making prediction more difficult. The weighting metric can be calculated using the following formula 12: Formula 12:

[0073] in, Indicates that for the first The weighted index calculated for each sample day Indicates the first The standard deviation of power for each sample day. Indicates the first The average daily output value for each sample day. This represents a preset minimum constant. Generally, 1×10 is taken. -5 ,set up The main purpose is to prevent the average daily output value A division by zero error occurs when the value is close to 0.

[0074] In this way, by constructing weighting indicators, the sample days can be weighted according to power fluctuations and average daily output. This allows days with low output and high fluctuations to receive higher training weights during model training, helping the model better learn the power change patterns under these key scenarios. This effectively improves the prediction performance of the trained wind and solar power prediction model in key scenarios, thereby enhancing the overall accuracy and reliability of wind and solar power prediction. For example, for a sample day with a calculated power standard deviation of 200kW and an average daily output of 1500kW, substituting into Formula 12, where... Calculate the weight index for this sample day. The same method was used to calculate the weights for other sample days so that they could be used in subsequent model training.

[0075] During model training, weighting metrics need to be integrated into the training process. Through a weighted loss function and gradient optimization strategy, the model focuses more on the prediction accuracy of key samples. Specifically, in step S30, which involves training the wind and solar power prediction model using wind and solar power generation data from multiple sample days, the model parameters are optimized using weighting metrics to train the wind and solar power prediction model. This includes: S33: Determine the preset loss function for training the wind and solar power prediction model, integrate the weight index into the preset loss function, and obtain the weighted loss function shown in the following formula.

[0076] The preset loss function is a fundamental function used to measure the difference between the model's predicted values ​​and the true values. In this embodiment, the loss function in the form of mean squared error is used as the preset loss function, which is then used as the basis for weighted design. Specifically, sample weights need to be incorporated into the loss function design. The weight vector of each sample is multiplied element-wise with the prediction error vector. When calculating the overall loss, the program replaces the traditional arithmetic mean with a weighted summation method to obtain the weighted loss function shown in Formula 13 below: Formula 13:

[0077] in, This represents the total model loss calculated using the weighted loss function. Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day Indicates the first The predicted power value for each sample day, Indicates the first The actual power value for each sample day.

[0078] In this way, by integrating the weight index into the loss function to construct a weighted loss function, the model can pay different attention to the prediction error of different samples according to the importance of the samples during training, and prioritize reducing the prediction error of key samples, thereby improving the prediction accuracy of the model in key scenarios.

[0079] S34: During the training of the wind and solar power prediction model using wind and solar power generation data from multiple sample days, the model parameters of the wind and solar power prediction model are adjusted according to the following formula until the preset convergence condition is met, thus obtaining the wind and solar power prediction model.

[0080] In the gradient calculation stage, a weighted optimization strategy is adopted, adjusting the parameter update magnitude according to the sample weights. During backpropagation, the gradient generated by each sample is multiplied by its corresponding weight coefficient, as detailed in Formula 14 below: Formula 14:

[0081] in, This represents the objective function used to guide the direction of model parameter updates. Regarding model parameters gradient, Indicates model parameters, Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day Represents the weighted loss function Regarding model parameters The gradient.

[0082] During backpropagation, the gradient generated by each sample is multiplied by its corresponding weight coefficient, allowing samples with larger weights to play a greater role in parameter updates. The model parameters are continuously adjusted until the preset convergence condition is met, ultimately yielding the wind and solar power prediction model. This weighted optimization strategy allows the model to focus more on key samples during training, optimizing its predictive ability for key scenarios and effectively improving the prediction performance of the wind and solar power prediction model in critical situations, thereby enhancing overall prediction accuracy and reliability. Continuing with the example of the aforementioned wind and solar power base, during training, the parameters of the wind and solar power prediction model are adjusted according to the constructed weighted loss function and the gradient calculation formula. For samples with larger weights, their gradients are amplified during backpropagation, thus playing a more significant role in parameter updates. After multiple iterations of training, when the model's total loss reaches the preset convergence condition, the final wind and solar power prediction model is obtained. This model can more accurately predict wind and solar power generation, especially in critical scenarios such as low output and high fluctuations.

[0083] In practical applications, weighted training can be achieved through a bidirectional long short-term memory network architecture. This architecture can better capture long-term dependencies in the data. Combined with a weighted training strategy, the model focuses more on the prediction accuracy of key samples. Specifically, the gradient of high-weight samples has a larger proportion in parameter updates, making the model parameter updates more inclined to optimize the prediction accuracy of key samples. In addition, an adaptive weight allocation strategy that dynamically adjusts according to sample characteristics can be used, and gradient normalization technology can be adopted to prevent excessive weights from causing training instability. The specific process is as follows: First, a weight allocation curve based on the average daily output is constructed, with the principle that the lower the output level, the larger the weight coefficient. This is because low output levels often correspond to special scenarios for wind and solar power generation, such as weak winds and low sunlight, in which prediction is more difficult. Assigning higher weights can make the model focus more on the prediction accuracy in these scenarios. Second, a weight allocation curve based on the fluctuation intensity index is constructed, with the principle that the higher the fluctuation intensity, the larger the weight coefficient. The fluctuation intensity index reflects the severity of fluctuations in wind and solar power generation. Prediction errors are usually larger under high fluctuation conditions. Increasing its weight helps the model better learn the power change patterns in these scenarios. Finally, the weight coefficients are normalized to prevent excessively large weights from causing training instability and to ensure numerical stability during training. Gradient normalization is also employed to further guarantee training stability.

[0084] In this way, by adopting an adaptive weight allocation strategy and dynamically adjusting weights based on sample characteristics, the model can more accurately focus on key samples during training, such as low-output and high-fluctuation scenarios, effectively improving prediction performance in these key scenarios. Meanwhile, normalization and gradient normalization techniques ensure training stability, enabling reliable parameter optimization of the model.

[0085] S40: When there is a demand for wind and solar power prediction, use the wind and solar power prediction model to predict the power.

[0086] In this embodiment, when a demand for wind and solar power prediction arises, for example, when the power grid dispatching department needs to understand the power generation of wind and solar power bases in advance for a period of time in order to make reasonable power allocation, relevant power parameters for the current time and a period of time in the future are collected, including meteorological parameters and operating status parameters of power generation equipment, etc. These parameters are input into the trained wind and solar power prediction model. The model outputs the predicted value of wind and solar power generation for a period of time in the future based on the input parameters and the internally learned patterns.

[0087] In this way, by using the optimized and trained wind and solar power prediction model through the above process, the overall accuracy and reliability of wind and solar power prediction can be improved, providing a more accurate basis for grid dispatch, helping the grid to rationally arrange power generation plans and power transmission, meeting the needs of wind and solar power prediction in practical engineering applications, and ensuring the stable operation of the power system.

[0088] For example, if the power grid dispatching department needs to know the power generation of a wind and solar power generation base for the next 24 hours at 10:00 a.m. the next day, it will collect the meteorological parameters (wind speed, light intensity, etc.) and the operating status parameters of the power generation equipment at 10:00 a.m. that day and input them into the trained wind and solar power prediction model. The model will output the predicted power generation values ​​for each time point in the next 24 hours, providing an accurate reference for power grid dispatching.

[0089] In summary, the logical process of the self-correcting prediction method for wind and solar power generation bases proposed in this application is summarized as follows: Figure 2 As shown, the first step is to acquire raw data, including historical power generation data from wind and solar power bases and relevant meteorological data. This raw data forms the basis for subsequent analysis, but it often contains issues such as noise and missing values, thus requiring data preprocessing. After data preprocessing, the power feature analysis stage begins, aiming to extract key features affecting prediction accuracy and clarify the importance of different samples in the prediction. Next, the weights for low-output days and high-fluctuation days are calculated, assigning corresponding weights to each sample day. Then, the weight coefficients are normalized to prevent excessively large weights from causing training instability, ensuring numerical stability during training, and guaranteeing the rationality of weight allocation and the reliability of training.

[0090] During the weighted model training phase, normalized weight metrics are integrated into the loss function, and a bidirectional long short-term memory network architecture is used for training. This architecture can better capture long-term dependencies in the data. Combined with a weighting strategy, the model focuses more on key samples with higher weights during training, such as low-output and high-fluctuation scenarios. During backpropagation, the parameter update magnitude is adjusted according to the sample weights, with the gradients of high-weight samples accounting for a larger proportion in parameter updates, prompting the model parameters to be updated in the direction of optimizing the prediction accuracy of key samples. Finally, the trained wind and solar power prediction model is used to predict power and output the prediction results.

[0091] The method provided in this application identifies multiple sample days, evaluates the wind and solar power generation data for each sample day to screen out days with high prediction errors, performs power analysis on the wind and solar power generation data of these days to determine key factors affecting prediction accuracy, constructs weighting indicators based on these key factors, and optimizes the model parameters by incorporating these weighting indicators during the training of the wind and solar power prediction model using the wind and solar power generation data from multiple sample days. When a wind and solar power prediction demand arises, the model is used for power prediction. By screening for days with high prediction errors, extreme scenarios with large prediction errors are effectively identified. Furthermore, by analyzing the power parameters, key factors affecting prediction accuracy are identified, and weighting indicators are synthesized based on these key factors. This allows the key factors to obtain higher training weights during model training, effectively improving the prediction performance of the trained wind and solar power prediction model in key scenarios, enhancing the overall accuracy and reliability of wind and solar power prediction, providing more accurate decision-making basis for grid dispatch, and meeting the needs of practical engineering applications.

[0092] Furthermore, as Figure 1 To specifically implement the method, this application provides a power self-correction prediction device for wind and solar power generation bases, such as... Figure 3 As shown, the device includes: an error day screening module 301, a key factor identification module 302, a weighted training module 303, and a prediction module 304.

[0093] Error day screening module 301 is used to select multiple sample days, evaluate the wind and solar power generation data of each sample day, and screen out high prediction error days for wind and solar power from the multiple sample days. The key factor identification module 302 is used to analyze the power parameters of the days with high prediction errors in wind and solar power, and to determine the key factors affecting the prediction accuracy. The weighted training module 303 is used to construct weight indicators based on the key factors, and to optimize the model parameters in combination with the weight indicators during the process of training the wind and solar power prediction model using wind and solar power generation data from the multiple sample days, so as to train the wind and solar power prediction model. The prediction module 304 is used to perform power prediction using the wind and solar power prediction model when a wind and solar power prediction demand is generated.

[0094] In a specific application scenario, the error day screening module 301 is used to select the plurality of sample days, obtain the wind and solar power generation data for each sample day, and preprocess the wind and solar power generation data for each sample day. The wind and solar power generation data for each sample day includes the original wind and solar power generation data and meteorological data for the corresponding sample day. The normalized average absolute error for each sample day is calculated using the following formula.

[0095] in, This represents the calculated normalized mean absolute error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​of the multiple sample days. Indicates the first The predicted power value for each sample day, Indicates the first The total number of sampling points on each sample day; The normalized root mean square error for each sample day is calculated using the following formula.

[0096] in, This represents the calculated normalized root mean square error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​of the multiple sample days. Indicates the first The predicted power value for the nth sample day represents the power value for the nth sample day. The total number of sampling points on each sample day; Using the normalized mean absolute error and normalized mean absolute error of each sample day, a corresponding comprehensive error evaluation index is constructed for each sample day; based on the comprehensive error evaluation index corresponding to each sample day, the multiple sample days are sorted in descending order to obtain the sorting result; the sample days ranked in the top preset position in the sorting result are selected as the wind and solar high prediction error days.

[0097] In a specific application scenario, the error day screening module 301 is used to calculate the corresponding comprehensive error evaluation index for each sample day using the following formula.

[0098] in, Represents the calculated first... The comprehensive error evaluation index corresponding to each sample day and This represents the preset weighting coefficient. Indicates the first Normalized mean absolute error for each sample day Indicates the first Normalized mean absolute error for each sample day.

[0099] In specific application scenarios, the key factor identification module 302 is used to calculate the power output of wind and solar power generation data on days with high prediction errors using the following formula, thereby obtaining the average daily output value on those days.

[0100] in, This represents the calculated average daily output value. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value of a day with high prediction error in wind and solar power; The average daily output value is considered as the key factor affecting the accuracy of the prediction.

[0101] In specific application scenarios, the device further includes: The correlation analysis module is used to calculate the power output of wind and solar power generation data on days with high prediction errors using the following formula, thereby obtaining the fluctuation intensity index for those days.

[0102] in, This represents the calculated volatility index. This represents the average daily power output calculated for the day with the highest wind and solar power prediction error. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value for days with high prediction errors in wind and solar power forecasts. Indicates the first The actual power value for each day with high prediction error in wind and solar power forecasting; the first correlation coefficient between the average daily power output and the prediction error for each day with high prediction error in wind and solar power forecasting is calculated using the following formula.

[0103] in, This represents the first correlation coefficient. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The average daily power output on days with high forecast error in wind and solar power. This represents the average daily power output value for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. This represents the average of the comprehensive error assessment index for all days with high forecast errors in wind and solar power. The following formula is used to calculate the second correlation coefficient between the fluctuation intensity index and the forecast error for each day with high forecast errors in wind and solar power.

[0104] in, This represents the second correlation coefficient. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The fluctuation intensity index of days with high prediction errors in wind and solar power. This represents the average fluctuation intensity index for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. The average value of the comprehensive error assessment index for all days with high prediction errors in wind and solar power is represented. Based on the first correlation coefficient, first descriptive information is generated to describe the negative correlation between the average daily power output and the prediction error, and based on the second correlation coefficient, second descriptive information is generated to describe the positive correlation between the fluctuation intensity index and the prediction error. Using the fluctuation intensity index of the days with high prediction errors in wind and solar power, the first descriptive information, and the second descriptive information, a model adjustment basis is generated, and when the wind and solar power prediction model is trained, the model adjustment basis is used to label the wind and solar power prediction model.

[0105] In specific application scenarios, the weighted training module 303 is used to calculate the power standard deviation for each sample day using the following formula.

[0106] in, Indicates the first The standard deviation of power for each sample day. Indicates the first The total number of sampling points on each sample day Indicates the first Sample day Power value at time, Indicates the first The average daily output value for each sample day; the power standard deviation and average daily output value for each sample day are calculated using the following formula to obtain the weighting index.

[0107] in, Indicates that for the first The weighted index calculated for each sample day Indicates the first The standard deviation of power for each sample day. Indicates the first The average daily output value for each sample day. This represents a preset minimum constant.

[0108] In a specific application scenario, the weighted training module 303 is used to determine a preset loss function for training the wind and solar power prediction model, and integrate the weight index into the preset loss function to obtain the weighted loss function shown in the following formula.

[0109] in, This represents the total model loss calculated using the weighted loss function. Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day Indicates the first The predicted power value for each sample day, Indicates the first The actual power values ​​for each sample day; during the training of the wind and solar power prediction model using the wind and solar power generation data from the multiple sample days, the model parameters of the wind and solar power prediction model are adjusted according to the following formula until the preset convergence condition is met, thus obtaining the wind and solar power prediction model.

[0110] in, This represents the objective function used to guide the direction of model parameter updates. Regarding model parameters gradient, Indicates model parameters, Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day The weighted loss function represents... Regarding model parameters The gradient.

[0111] The apparatus provided in this application selects multiple sample days and evaluates the wind and solar power generation data for each sample day to identify days with high prediction errors. Power analysis is then performed on the wind and solar power generation data of these days to determine key factors affecting prediction accuracy. Based on these key factors, weighting indices are constructed. During the training of the wind and solar power prediction model using the data from multiple sample days, the model parameters are optimized using these weighting indices to obtain the desired model. When a wind and solar power prediction requirement arises, the model is used for power prediction. By selecting days with high prediction errors, extreme scenarios with large prediction errors are effectively identified. Furthermore, by analyzing the power parameters, key factors affecting prediction accuracy are identified, and weighting indices are synthesized based on these key factors. This allows the key factors to receive higher training weights during model training, effectively improving the prediction performance of the trained wind and solar power prediction model in key scenarios. This enhances the overall accuracy and reliability of wind and solar power prediction, providing more accurate decision-making support for grid dispatch and meeting the needs of practical engineering applications.

[0112] It should be noted that other corresponding descriptions of the functional units involved in the self-correcting prediction device for wind and solar power generation bases provided in this application embodiment can be found in the following references. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0114] The above embodiments and the technical features in the embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

[0116] In an exemplary embodiment, see Figure 4Furthermore, an electronic device is provided, comprising a bus, a processor, a memory, and a communication interface. It may also include input / output interfaces and a display device, wherein the various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the self-correcting prediction method for wind and solar power generation bases described in the above embodiments.

[0117] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the self-correcting prediction method for power generation bases of the wind and solar power generation bases.

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented in hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) and includes several instructions to cause an electronic device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0119] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application.

[0120] Those skilled in the art will understand that the modules in the apparatus of the implementation scenario can be distributed within the apparatus of the implementation scenario as described, or they can be located in one or more apparatuses different from this implementation scenario with corresponding changes. The modules of the above-mentioned implementation scenario can be combined into one module, or they can be further divided into multiple sub-modules.

[0121] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of the implementation scenario.

[0122] The above disclosures are only a few specific implementation scenarios of this application. However, this application is not limited to these. Any variations that can be conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A self-correcting prediction method for power generation at wind and solar power bases, characterized in that, include: Multiple sample days are selected, and the wind and solar power generation data for each sample day are evaluated to screen out days with high prediction errors for wind and solar power from the multiple sample days. Power analysis was performed on the wind and solar power generation data of the days with high prediction errors to determine the key factors affecting prediction accuracy. Based on the key factors, a weight index is constructed, and the model parameters are optimized in combination with the weight index during the process of training the wind and solar power prediction model using the wind and solar power generation data of the multiple sample days, so as to train the wind and solar power prediction model. When there is a demand for wind and solar power prediction, the aforementioned wind and solar power prediction model is used to predict power.

2. The method according to claim 1, characterized in that, The step of selecting multiple sample days and evaluating the wind and solar power generation data for each sample day to screen out days with high prediction errors for wind and solar power from the multiple sample days includes: Select the multiple sample days, obtain the wind and solar power generation data for each sample day, and preprocess the wind and solar power generation data for each sample day. The wind and solar power generation data for each sample day includes the original wind and solar power generation data and meteorological data for the corresponding sample day. The normalized mean absolute error for each of the sample days is calculated using the following formula. in, This represents the calculated normalized mean absolute error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​of the multiple sample days. Indicates the first The predicted power value for each sample day, Indicates the first The total number of sampling points on each sample day; The normalized root mean square error for each sample day is calculated using the following formula. in, This represents the calculated normalized root mean square error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​of the multiple sample days. Indicates the first The predicted power value for the nth sample day represents the power value for the nth sample day. The total number of sampling points on each sample day; Using the normalized mean absolute error and normalized mean absolute error of each sample day, a corresponding comprehensive error evaluation index is constructed for each sample day. Based on the comprehensive error evaluation index corresponding to each sample day, the multiple sample days are sorted in descending order to obtain the sorting result; The sample days ranked at the top preset number in the sorting results are selected as the predicted error days for wind and solar high.

3. The method according to claim 2, characterized in that, The step involves using the normalized mean absolute error and the normalized mean absolute error of each sample day to construct a corresponding comprehensive error evaluation index for each sample day, including: For each sample day, the corresponding comprehensive error evaluation index is calculated using the following formula. in, Represents the calculated first... The comprehensive error evaluation index corresponding to each sample day and This represents the preset weighting coefficient. Indicates the first Normalized mean absolute error for each sample day Indicates the first Normalized mean absolute error for each sample day.

4. The method according to claim 1, characterized in that, The power analysis of wind and solar power generation data on days with high prediction errors is performed to determine the key factors affecting prediction accuracy, including: The following formula is used to calculate the power output of wind and solar power generation data on days with high prediction errors, thereby obtaining the average daily power output value for those days. in, This represents the calculated average daily output value. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value of a day with high prediction error in wind and solar power; The average daily output value is considered as the key factor affecting the accuracy of the prediction.

5. The method according to claim 4, characterized in that, The method further includes: The following formula is used to calculate the power output of wind and solar power generation data on days with high prediction errors, thereby obtaining the fluctuation intensity index for those days. in, This represents the calculated volatility index. This represents the average daily power output calculated for the day with the highest wind and solar power prediction error. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value for days with high prediction errors in wind and solar power forecasts. Indicates the first The actual power value of a day with high prediction error in wind and solar power; The following formula is used to calculate the first correlation coefficient between the average daily power output and the prediction error on the day with the highest wind and solar power prediction error. in, This represents the first correlation coefficient. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The average daily power output of the wind and solar power system on days with high forecast error. This represents the average daily power output value for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. This represents the average value of the comprehensive error assessment index for all days with high forecast error in wind and solar power. The second correlation coefficient between the fluctuation intensity index and the prediction error of the wind and solar high prediction error day is calculated using the following formula. in, This represents the second correlation coefficient. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The fluctuation intensity index of days with high prediction errors in wind and solar power. This represents the average fluctuation intensity index for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. This represents the average value of the comprehensive error assessment index for all days with high forecast error in wind and solar power. Based on the first correlation coefficient, first descriptive information is generated to describe the negative correlation between the average daily output value and the prediction error, and based on the second correlation coefficient, second descriptive information is generated to describe the positive correlation between the fluctuation intensity index and the prediction error. Using the fluctuation intensity index of the day with high prediction error of wind and solar power, the first descriptive information and the second descriptive information, a model adjustment basis is generated, and when the wind and solar power prediction model is trained, the model adjustment basis is used to label the wind and solar power prediction model.

6. The method according to claim 1, characterized in that, The construction of weighting indicators based on the key factors includes: The power standard deviation for each sample day is calculated using the following formula. in, Indicates the first The standard deviation of power for each sample day. Indicates the first The total number of sampling points on each sample day Indicates the first Sample day Power value at time, Indicates the first The average daily output value for each sample day; The weighting index is obtained by calculating the power standard deviation and average daily output value for each sample day using the following formula. in, Indicates that for the first The weighted index calculated for each sample day Indicates the first The standard deviation of power for each sample day. Indicates the first The average daily output value for each sample day. This represents a preset minimum constant.

7. The method according to claim 1, characterized in that, In the process of training the wind and solar power prediction model using the wind and solar power generation data from the multiple sample days, the model parameters are optimized by combining the weight index to train the wind and solar power prediction model, including: A preset loss function is determined for training the wind and solar power prediction model. The weight index is then integrated into the preset loss function to obtain the weighted loss function shown in the following formula. in, This represents the total model loss calculated using the weighted loss function. Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day Indicates the first The predicted power value for each sample day, Indicates the first The actual power value for each sample day; During the training of the wind and solar power prediction model using the wind and solar power generation data from the multiple sample days, the model parameters of the wind and solar power prediction model are adjusted according to the following formula until the preset convergence condition is met, thus obtaining the wind and solar power prediction model. in, This represents the objective function used to guide the direction of model parameter updates. Regarding model parameters gradient, Indicates model parameters, Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day The weighted loss function represents... Regarding model parameters The gradient.

8. A power self-correction prediction device for wind and solar power generation bases, characterized in that, include: The error day screening module is used to select multiple sample days, evaluate the wind and solar power generation data of each sample day, and screen out the wind and solar high prediction error days from the multiple sample days. The key factor identification module is used to analyze the power parameters of the days with high prediction errors in wind and solar power, and to determine the key factors affecting the prediction accuracy. The weighted training module is used to construct weight indicators based on the key factors, and to optimize the model parameters in combination with the weight indicators during the process of training the wind and solar power prediction model using wind and solar power generation data from the multiple sample days, so as to train the wind and solar power prediction model. The prediction module is used to perform power prediction using the wind and solar power prediction model when a demand for wind and solar power prediction is generated.

9. The apparatus according to claim 8, characterized in that, The error day screening module is used to select the plurality of sample days, obtain the wind and solar power generation data for each sample day, and preprocess the wind and solar power generation data for each sample day. The wind and solar power generation data for each sample day includes the original wind and solar power generation data and meteorological data for the corresponding sample day. The normalized mean absolute error for each sample day is calculated using the following formula. in, This represents the calculated normalized mean absolute error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​of the multiple sample days. Indicates the first The predicted power value for each sample day, Indicates the first The total number of sampling points on each sample day; The normalized root mean square error for each sample day is calculated using the following formula. in, This represents the calculated normalized root mean square error. Indicates the first The actual power value for each sample day This represents the maximum value among the true power values ​​of the multiple sample days. Indicates the first The predicted power value for the nth sample day represents the power value for the nth sample day. The total number of sampling points on each sample day; Using the normalized mean absolute error and normalized mean absolute error of each sample day, a corresponding comprehensive error evaluation index is constructed for each sample day; based on the comprehensive error evaluation index corresponding to each sample day, the multiple sample days are sorted in descending order to obtain the sorting result; the sample days ranked in the top preset position in the sorting result are selected as the wind and solar high prediction error days.

10. The apparatus according to claim 9, characterized in that, The error day screening module is used to calculate the corresponding comprehensive error evaluation index for each sample day using the following formula. in, Represents the calculated first... The comprehensive error evaluation index corresponding to each sample day and This represents the preset weighting coefficient. Indicates the first Normalized mean absolute error for each sample day Indicates the first Normalized mean absolute error for each sample day.

11. The apparatus according to claim 8, characterized in that, The key factor identification module is used to calculate the power output of wind and solar power generation data on days with high prediction errors using the following formula, thereby obtaining the average daily output value on those days. in, This represents the calculated average daily output value. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value of a day with high prediction error in wind and solar power; The average daily output value is considered as the key factor affecting the accuracy of the prediction.

12. The apparatus according to claim 11, characterized in that, The device further includes: The correlation analysis module is used to calculate the power output of wind and solar power generation data on days with high prediction errors using the following formula, thereby obtaining the fluctuation intensity index for those days. in, This represents the calculated volatility index. This represents the average daily power output calculated for the day with the highest wind and solar power prediction error. Indicates the first The total number of sampling points on the day with high prediction error for wind and solar power. Indicates the first The actual power value for days with high prediction errors in wind and solar power forecasts. Indicates the first The actual power value for each day with high prediction error in wind and solar power forecasting; the first correlation coefficient between the average daily power output and the prediction error for each day with high prediction error in wind and solar power forecasting is calculated using the following formula. in, This represents the first correlation coefficient. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The average daily power output of the wind and solar power system on days with high forecast error. This represents the average daily power output value for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. This represents the average of the comprehensive error assessment index for all days with high forecast errors in wind and solar power. The following formula is used to calculate the second correlation coefficient between the fluctuation intensity index and the forecast error for each day with high forecast errors in wind and solar power. in, This represents the second correlation coefficient. This indicates the number of days with high forecast errors for wind and solar power. Indicates the first The fluctuation intensity index of days with high prediction errors in wind and solar power. This represents the average fluctuation intensity index for all days with high forecast errors in wind and solar power. Indicates the first The comprehensive error assessment index corresponding to each day with high prediction error in wind and solar power. The average value of the comprehensive error assessment index for all days with high prediction errors in wind and solar power is represented. Based on the first correlation coefficient, first descriptive information is generated to describe the negative correlation between the average daily power output and the prediction error, and based on the second correlation coefficient, second descriptive information is generated to describe the positive correlation between the fluctuation intensity index and the prediction error. Using the fluctuation intensity index of the days with high prediction errors in wind and solar power, the first descriptive information, and the second descriptive information, a model adjustment basis is generated, and when the wind and solar power prediction model is trained, the model adjustment basis is used to label the wind and solar power prediction model.

13. The apparatus according to claim 8, characterized in that, The weighted training module is used to calculate the power standard deviation for each sample day using the following formula. in, Indicates the first The standard deviation of power for each sample day. Indicates the first The total number of sampling points on each sample day Indicates the first Sample day Power value at time, Indicates the first The average daily output value for each sample day; the power standard deviation and average daily output value for each sample day are calculated using the following formula to obtain the weighting index. in, Indicates that for the first The weighted index calculated for each sample day Indicates the first The standard deviation of power for each sample day. Indicates the first The average daily output value for each sample day. This represents a preset minimum constant.

14. The apparatus according to claim 8, characterized in that, The weighted training module is used to determine a preset loss function for training the wind and solar power prediction model, and integrates the weight index into the preset loss function to obtain the weighted loss function shown in the following formula. in, This represents the total model loss calculated using the weighted loss function. Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day Indicates the first The predicted power value for each sample day, Indicates the first The actual power values ​​for each sample day; during the training of the wind and solar power prediction model using the wind and solar power generation data from the multiple sample days, the model parameters of the wind and solar power prediction model are adjusted according to the following formula until the preset convergence condition is met, thus obtaining the wind and solar power prediction model. in, This represents the objective function used to guide the direction of model parameter updates. Regarding model parameters gradient, Indicates model parameters, Indicates the first The total number of sampling points on each sample day Indicates that for the first The weighted index calculated for each sample day The weighted loss function represents... Regarding model parameters The gradient.

15. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

16. A medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images