Photovoltaic power generation power prediction method based on a mixture of gaussian kernel function modeling

Abstract

The present application relates to the technical field of photovoltaic power prediction, and particularly relates to a photovoltaic power prediction method based on a mixed Gaussian kernel function modeling. The method comprises: performing correlation analysis on meteorological factors, screening out the meteorological factor with the strongest correlation with photovoltaic power output, and recording the meteorological factor as a key meteorological factor; combining regression residuals and an isolated forest method to jointly clean photovoltaic data obtained, to obtain a data set; constructing a mixed Gaussian kernel prediction model with the key meteorological factor as input; using an adaptive filtering augmented gradient iteration algorithm to perform parameter identification on the mixed Gaussian kernel prediction model with the data set, until the model parameters meet a preset estimation precision, to obtain a mixed Gaussian kernel prediction model under final parameter estimation; and in a prediction stage, inputting the key meteorological factor at a time to be predicted into the mixed Gaussian kernel prediction model to predict photovoltaic power generation. The present application can improve the prediction accuracy of photovoltaic power generation under real complex scenarios and related noise interference.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power prediction technology, and specifically to a photovoltaic power prediction method based on Gaussian mixture kernel function modeling. Background Technology

[0002] Photovoltaic power generation, with its outstanding advantages such as clean production and flexible deployment (it can be built centrally or distributed), has become an important part of the pollution-free and resource-recycling field of renewable energy, and is recognized as one of the clean energy forms with the greatest potential for large-scale development. With the continuous and rapid growth of installed capacity, large-scale, high-proportion photovoltaic grid connection has become a clear trend in power system development.

[0003] Photovoltaic power output is directly affected by meteorological factors such as solar irradiance, ambient temperature, and wind speed. These factors vary with weather conditions, causing photovoltaic power output to exhibit randomness, fluctuation, and intermittency. Among these, solar irradiance is the main meteorological factor determining the variation in photovoltaic power generation, and there is usually a significant nonlinear mapping relationship between it and photovoltaic power output. This uncertainty in output makes it difficult for photovoltaic power plants to provide stable power support, posing serious challenges to the operation and scheduling of the power system, including supply and demand balance, frequency stability, voltage regulation, and reserve capacity configuration.

[0004] The fundamental way to effectively address the risks of large-scale photovoltaic grid connection lies in accurately grasping the changing patterns of photovoltaic power output and obtaining spatiotemporal distribution information of power in advance. High-precision photovoltaic power prediction models can provide scientific basis for grid dispatching agencies, facilitating the optimization of unit combinations, the allocation of reserve capacity, and the formulation of inter-regional power transmission plans in advance, thereby reducing the impact of photovoltaic fluctuations on system security and improving the capacity for renewable energy absorption.

[0005] Therefore, establishing an accurate and reliable photovoltaic power prediction model is of great theoretical and practical value for ensuring the safety and economy of the power system under high-proportion photovoltaic access. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling, which can improve the prediction accuracy of photovoltaic power generation under real complex scenarios and related noise interference.

[0007] To solve the above-mentioned technical problems, the technical solution of the present invention is: a photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling, comprising: S1. Conduct correlation analysis on meteorological factors, screen out the meteorological factors with the strongest correlation with photovoltaic power output, and record them as key meteorological factors; S2. Combine regression residuals and the isolated forest method to jointly clean the acquired photovoltaic data, obtain the photovoltaic power prediction dataset, and divide it into training set and test set; S3. Construct a mixture Gaussian kernel prediction model with key meteorological factors as input; S4. Using the adaptive filtering augmented gradient iterative algorithm, the parameters of the mixed Gaussian kernel prediction model are identified using the photovoltaic power prediction training set until the model parameters meet the preset convergence accuracy, and the final parameter estimates that meet the preset convergence conditions are obtained. S5. In the prediction stage, the key meteorological factors of the time to be predicted are input into the hybrid Gaussian kernel prediction model after the parameters are determined, so as to obtain the predicted value of photovoltaic power generation at the time to be predicted.

[0008] Furthermore, a correlation analysis was conducted on the meteorological factors; specifically, the Pearson correlation coefficient was used to perform a correlation analysis on the meteorological factors.

[0009] Furthermore, the key meteorological factor is solar irradiance.

[0010] Furthermore, an adaptive filtering augmented gradient iterative algorithm is used to identify the parameters of the mixed Gaussian kernel prediction model using the photovoltaic power prediction training set until the model parameters meet the preset convergence accuracy, thus obtaining the final parameter estimates that satisfy the preset convergence conditions; specifically including: Step S41: Start the Gaussian mixture kernel prediction model and initialize it; Step S42: Use key meteorological factors from a set of data in the photovoltaic power prediction training set as input data, and use the measured value of photovoltaic power generation at the corresponding time as output data. Step S43, construct the information vector; Step S44: Construct the filtered information vector; Step S45: Determine the iteration step size, update the parameter estimates, and calculate the estimates of colored noise and white noise; Step S46: Determine whether the model meets the accuracy requirements. If not, return to step S42. If it does, obtain the parameter estimation results, that is, determine the final parameter estimates that meet the preset convergence conditions.

[0011] Furthermore, the expression for the Gaussian mixture kernel prediction model is:

[0012] Among them, parameters , and ( ) represents the feature parameters of the Gaussian mixture model. Indicates the input to the prediction model. This represents the output of the prediction model. This indicates colored noise interference.

[0013] The colored noise interference item Represented as: , , Right now , in, Represents the unit delay operator ( ); Indicates the order of the colored noise model; The mean is zero and the variance is . Random white noise, ( ) indicates the first Parameters of the colored noise model.

[0014] To use the constructed hybrid Gaussian kernel prediction model for photovoltaic power prediction, the correspondence between the parameters in the model and the key parameters for photovoltaic prediction is defined as follows: ( ) represents the linear weighting coefficient of photovoltaic power output with respect to different light levels; ( The sensitivity coefficient of photovoltaic power to changes in sunlight; ( The reference center represents the level of illumination. ( () represents the moving average coefficient of photovoltaic noise. It is white noise.

[0015] Furthermore, the unknown feature parameters in the Gaussian mixture kernel function model are defined as a parameter vector:

[0016]

[0017]

[0018]

[0019] To facilitate model training, parameter vectors are integrated. and A parameter vector Integrate parameter vectors and A parameter vector ,Right now

[0020]

[0021] Define the information vector as:

[0022]

[0023]

[0024] After processing, the identification model of the Gaussian mixture kernel prediction model is obtained, and its expression is:

[0025] In step S41, the identification model is activated.

[0026] Furthermore, step S43 specifically includes: constructing and ;

[0027] in, Sampling time; For data length; For iteration variables; For the first The iteration yielded Parameter estimates; For the first The iteration yielded Parameter estimates; For the first The information vector obtained from the next iteration; For noise The Sub-iteration estimation; For the first The noise information vector obtained from the next iteration.

[0028] Furthermore, step S44 specifically includes: designing a data filter. ,calculate , , , and ; Design a filter ,in This represents the structural expression for colored noise. Without altering the system's input-output relationship, the original identification model is filtered, transforming the noise-affected model into a filtered and noise-affected model. This reduces the interference of colored noise on parameter estimation, improving the accuracy and stability of model parameter identification. The filtered output is defined. Filtered input and information vector as follows:

[0029] Multiply both ends of the original identification model by a filter. , on observation data and information vector After filtering, the filtered identification model can be obtained as follows:

[0030] Using noise model parameter vectors Estimate Constructing filters Estimate for:

[0031] in, Indicates the first Colored noise structure polynomial in the next iteration The estimated value.

[0032] Estimation using filters replace The unknown quantity can be obtained. , and Estimate , and for:

[0033] definition and The formula is:

[0034] in, This represents the sensitivity estimation vector relative to the Gaussian kernel nonlinear parameters; express and The augmented information vector is used to uniformly update the parameters of the Gaussian mixture kernel model.

[0035] Furthermore, in step S45, in order to effectively improve the convergence speed and parameter identification accuracy while ensuring stable convergence of the algorithm, an appropriate iteration step size needs to be selected. Define the iteration step size. and The formula is:

[0036] in, Represents the parameter vector The iteration step size; Represents the parameter vector The iteration step size.

[0037] definition and The formula is:

[0038] in, Indicates the first The colored noise estimate obtained in the next iteration; Indicates the first The white noise estimate obtained in the next iteration.

[0039] Based on the structure of the Gaussian mixture kernel prediction model, the adaptive filter augmented gradient iterative (AF-EGI) algorithm can be derived. The parameter estimation vector of the Gaussian mixture kernel model is shown. and Iterative updates are performed according to the following formula:

[0040] in, Indicates the first The parameter estimation vector of the mixture Gaussian kernel model obtained in the next iteration; Indicates the first The parameter estimation vector of the colored noise model obtained in the next iteration.

[0041] Furthermore, step S46 specifically includes: Determine whether the model meets the accuracy requirements. If yes, return to step S42; otherwise, obtain... , and The parameter estimation result is the final parameter estimate that satisfies the preset convergence condition.

[0042] Furthermore, in step S5, the Gaussian kernel mixture prediction model after the parameters are determined is:

[0043] in, Indicates the first Predicted photovoltaic output power at any given time; Indicates the first The solar irradiance input acquired at any time; and Indicates the first The iteration yielded and The final estimated value; and Indicates the first The iteration yielded and The final estimated value; and Indicates the first The iteration yielded and The final estimated value; Indicates the order of the colored noise model; Indicates the first The final estimated values ​​of the parameters of the colored noise model; Indicates the first The estimated historical prediction error at any given time.

[0044] Without loss of generality, assume that the initial conditions are satisfied: Based on this, the historical prediction error estimate is defined as:

[0045] When the first After obtaining the measured photovoltaic power value at time 1, the prediction error estimate for that time is calculated and used for photovoltaic power prediction at subsequent times.

[0046] By adopting the above technical solution, the present invention has the following beneficial effects: This invention addresses the strong nonlinearity and random fluctuations in photovoltaic power output caused by the coupled influence of multiple factors such as irradiance, temperature, meteorological conditions, and equipment operating conditions. It employs a Gaussian mixture kernel function to locally, smoothly, and superimpose a high-precision approximation of the complex nonlinear relationship between key input meteorological factors and output power. The Gaussian kernel function, with its excellent local response and nonlinear expression capabilities, can accurately characterize the nonlinear dynamic behavior of photovoltaic power as irradiance changes. Simultaneously, this invention fully considers that errors in real-world systems are not ideal white noise but rather colored noise with significant time correlation. Colored noise modeling is integrated into the prediction model framework, enabling the model to describe the time correlation and fluctuation persistence of prediction errors, achieving effective characterization of the dynamic characteristics of related noise. This unified modeling strategy for nonlinear characteristics and noise influence allows the constructed Gaussian mixture kernel prediction model to be effectively applied to the identification of complex nonlinear dynamic systems affected by related noise.

[0047] This invention proposes an adaptive filtering augmented gradient iterative algorithm based on the Gaussian mixture kernel prediction model to assist in parameter identification of the prediction model. This method employs batch data processing, which fully utilizes observational data, thereby achieving stronger noise resistance and higher parameter estimation accuracy. By introducing the adaptive filtering augmented gradient iterative method into the parameter identification stage of the prediction process, the biased influence of colored noise on parameter estimation can be effectively suppressed, ensuring the model's prediction performance in complex noise environments. The final parameter estimation model obtained after identification can not only accurately characterize the nonlinear response of photovoltaic power to irradiance but also capture the time-domain dependence of prediction errors, thus improving the model's prediction accuracy in real-world complex scenarios.

[0048] In summary, this invention, from a system identification perspective, addresses the inherent limitations of traditional linear models in effectively approximating complex nonlinear objects. It establishes a prediction model centered on the fusion of Gaussian mixture kernel functions and colored noise, and combines this with an adaptive filtering augmented gradient iterative algorithm to form a complete parameter identification and prediction framework. This technical solution can accurately grasp the dynamic trends of photovoltaic power output, effectively addressing the nonlinearity, randomness, and volatility in photovoltaic power output. It provides reliable power prediction support for photovoltaic power plant operation and maintenance, grid dispatching decisions, and the safe and stable operation of the power system, possessing significant engineering application value. Attached Figure Description

[0049] Figure 1 This is a flowchart of the photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling of the present invention; Figure 2 This is a flowchart of the adaptive filtering augmented gradient iterative algorithm of the present invention; Figure 3 This is a structural diagram of the hybrid Gaussian kernel prediction model of the present invention; Figure 4 Clean the graphs for the dataset; Figure 5 Error histograms of the AEGI and AF-EGI algorithms on the test set; Figure 6 Regression plots of the AEGI and AF-EGI algorithms on the test set; Figure 7 This is a comparison graph showing the predicted outputs and actual values ​​of the AEGI and AF-EGI algorithms on the test set. Detailed Implementation

[0050] The present invention will be further described below by way of preferred embodiments, but the invention is not limited to the scope of the embodiments described herein.

[0051] A photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling, such as Figure 1 and Figure 2 As shown, it includes: Step S1: Perform correlation analysis on meteorological factors and screen out the meteorological factors that are most strongly correlated with photovoltaic power output, and record them as key meteorological factors.

[0052] The formula for calculating the Pearson correlation coefficient is as follows:

[0053] In the formula, Pearson correlation coefficient; For a certain meteorological factor, the first One sample; The highest historical photovoltaic output power One sample; and This is the average value; This represents the total number of samples.

[0054] The calculation results (see Table 1) show that the correlation coefficient between solar radiation intensity and photovoltaic output power is the largest, meaning that photovoltaic power is most strongly correlated with solar irradiance. Based on this, solar irradiance is selected as the meteorological factor input model affecting photovoltaic power.

[0055] Table 1. Correlation Statistical Table of Photovoltaic Output Power and Influencing Factors

[0056] Step S2: Combine regression residuals and the isolated forest method to jointly clean the acquired photovoltaic data, obtain the photovoltaic power prediction dataset, and divide it into training and test sets, specifically including: Step S21: The regression residual method is an anomaly detection method based on a prediction model. It uses a quadratic polynomial to fit the nonlinear relationship between irradiance and power, calculates the fitting residuals for each sample point, and uses 3 times the standard deviation of the residuals as a threshold to identify outliers that deviate from the fitting trend. The residuals for each observation point are calculated as follows: . use The principle is used as the standard to judge abnormalities.

[0057] . In the formula, P actual This represents the measured power value. P predicted This represents the power value fitted using a quadratic polynomial. It is the standard deviation of the residuals, and values ​​that satisfy the above formula are marked as outliers.

[0058] Step S22: Isolation Forest is a tree-based unsupervised anomaly detection method capable of handling multidimensional and complex distributed data. This method isolates sample points by randomly partitioning the feature space. Because anomalies are sparsely distributed and far from normal data clusters, they are easier to separate quickly. In photovoltaic data cleaning, anomalies can be detected directly from the irradiance-power two-dimensional feature space.

[0059] Isolated Forest Anomaly Score for:

[0060] The above formula Indicates sample The average path length in a tree. Indicates sample size The constant normalized coefficient, ,in, .

[0061] The anomaly labeling results from the two methods described above are combined, and the labeled samples are then confirmed and removed to obtain the cleaned photovoltaic power prediction dataset. This step combines regression residuals and isolated forests for joint data cleaning, taking into account both trend deviation and outlier distribution anomalies. This effectively reduces the interference of outlier samples on subsequent model parameter identification, ensuring the reliability and consistency of the training data. The cleaned data is as follows: Figure 4 As shown.

[0062] Step S3: Construct a mixture Gaussian kernel prediction model with key meteorological factors as input.

[0063] Step S31: Referring to the established model structure and the constructed information vector, the complete Gaussian mixture kernel prediction model expression is obtained as follows:

[0064] Among them, parameters , and ( ) represents the feature parameters of the Gaussian mixture model. Indicates the input to the prediction model. This represents the output of the prediction model. This indicates colored noise interference.

[0065] The colored noise interference item Represented as: , , Right now , in, Represents the unit delay operator ( ); Indicates the order of the colored noise model; The mean is zero and the variance is . Random white noise, ( ) indicates the first Parameters of the colored noise model.

[0066] To use the constructed hybrid Gaussian kernel prediction model for photovoltaic power prediction, the correspondence between the parameters in the model and the key parameters for photovoltaic prediction is defined as follows: ( ) represents the linear weighting coefficient of photovoltaic power output with respect to different light levels; ( The sensitivity coefficient of photovoltaic power to changes in sunlight; ( The reference center represents the level of illumination. ( () represents the moving average coefficient of photovoltaic noise. It is white noise. The structure of the Gaussian mixture kernel prediction model is as follows: Figure 3 As shown.

[0067] Step S32: In order to effectively capture the nonlinear response of photovoltaic power to irradiance, the unknown characteristic parameters in the Gaussian mixture kernel function prediction model are defined as a parameter vector:

[0068]

[0069]

[0070]

[0071] To facilitate model training, parameter vectors are integrated. and A parameter vector Integrate parameter vectors and A parameter vector ,Right now

[0072]

[0073] It should be noted that the definition The purpose is to simplify the subsequent formula expression so that the adaptive filtering augmented gradient iterative algorithm can make a unified estimate of the above parameters.

[0074] Define the information vector as:

[0075]

[0076]

[0077] in, By using Gaussian kernel nonlinear mapping of irradiance, the model is able to capture the nonlinear variation of photovoltaic power under different irradiance levels. By introducing time-correlated noise information, a more refined description of the random fluctuations in photovoltaic output is achieved, thereby improving prediction accuracy. These two elements constitute an augmented information vector. This vector not only utilizes the nonlinear characteristics of the current illumination, but also incorporates historical noise information.

[0078] Step S33: After rearranging the above equations, the complete identification model of the Gaussian mixture kernel prediction model is obtained, and its expression is:

[0079] in, The irradiance on a certain day, The photovoltaic power output for this day is predicted by the hybrid Gaussian kernel prediction model. Colored noise is considered interference. White noise for the identification process.

[0080] In the Gaussian mixture kernel prediction model, the information vector and Merged into augmented information vectors And set the corresponding parameter vector. For this model, the augmented information vector includes not only terms related to system parameters, but also terms of unmeasurable noise. This allows for the joint estimation of system parameters and noise parameters. To address the issue that noise cannot be directly measured, this invention uses noise estimates. Replace the unmeasurable noise term .

[0081] Step S4: Using the adaptive filtering augmented gradient iterative algorithm, the parameters of the mixed Gaussian kernel prediction model are identified using the photovoltaic power prediction training set until the model parameters meet the preset convergence accuracy, and the final parameter estimates that meet the preset convergence conditions are obtained.

[0082] definition and as follows:

[0083] in, Sampling time, For data length, For iteration variables, For the first The iteration yielded Parameter estimates For the first The iteration yielded Parameter estimates For the first The information vector obtained from the next iteration. For noise The Subsequent iteration estimation For the first The noise information vector obtained from the next iteration.

[0084] Design a filter ,in This represents the structural expression for colored noise. Without changing the system's input-output relationship, the original identification model is filtered, transforming the identification model under colored noise interference into a filtered identification model and a noise identification model. This reduces the interference of colored noise on the parameter estimation process and improves the accuracy and stability of model parameter identification.

[0085] Define the filter input Filtered output and information vector as follows:

[0086] Multiply both ends of the original identification model by a filter. , on observation data and information vector After filtering, the filtered identification model can be obtained as follows:

[0087] Using noise model parameter vectors Estimate Constructing filters Estimate for:

[0088] in, Indicates the first Colored noise structure polynomial in the next iteration The estimated value.

[0089] Estimation using filters replace The unknown quantity can be obtained. , and Estimate , and for:

[0090] definition and The formula is:

[0091] in, This represents the sensitivity estimation vector relative to the Gaussian kernel nonlinear parameters. express and The augmented information vector is used to uniformly update the parameters of the Gaussian mixture kernel model.

[0092] To ensure stable convergence of the algorithm while effectively improving convergence speed and parameter identification accuracy, an appropriate iteration step size needs to be selected. Define the iteration step size. and The formula is:

[0093] Among them, Represents the computational parameter vector The iteration step size. Represents the computational parameter vector The iteration step size.

[0094] definition and as follows:

[0095] in, Indicates the first The colored noise estimate obtained in the next iteration Indicates the first The white noise estimate obtained in the next iteration.

[0096] Based on the establishment of the Gaussian mixture kernel prediction model structure, an adaptive filtering augmented gradient iterative (AF-EGI) algorithm is derived to optimize the prediction model parameters. (Gaussian mixture kernel model parameter estimation vector) and Iterative updates are performed according to the following formula:

[0097] in, The algorithm indicates that it has gone through the first... The result obtained after the second iteration The estimated value. Indicates the first The parameter estimation vector of the colored noise model obtained in the next iteration.

[0098] The two formulas here are used for iteratively updating model parameters using the AF-EGI algorithm. Once the algorithm converges, from... Extract the weights, sensitivity coefficients, and central parameters of the Gaussian mixture kernel model, and then correlate them with... These two formulas are then substituted into the final prediction model to calculate the predicted photovoltaic output power. Therefore, they are the key steps between parameter training and final prediction.

[0099] Summarizing the above derivation process, the AF-EGI algorithm can be obtained, which includes the following steps: Step S41: Start the identification algorithm (identification model). Step S42, let the number of iterations be... Set initial values, data length, and model precision; Step S43: Obtain the light irradiance of a specific day in the training set. As input data, the measured value of photovoltaic power generation at the corresponding time. As output data; Step S44, Construction and ; Step S45: Design a data filter ,calculate , , and ; Step S46: Determine the iteration step size, update the parameter estimate, and calculate... and ; Step S47: Determine whether the model meets the accuracy requirements. ,but Return to step S43; otherwise, obtain , and Once the parameter estimation results are obtained, the process ends.

[0100] Indicates the preset convergence accuracy of the parameters; , and These are all parameters to be identified in the final photovoltaic power prediction model. Specifically, For the first The estimated vector of Gaussian kernel weight parameters obtained in the next iteration determines the degree of contribution of each Gaussian kernel to the photovoltaic power output; For the first The Gaussian kernel nonlinear parameter vector obtained in the second iteration The estimated value is used to determine the shape and range of influence of the curve of power changing with irradiance; This represents the parameter estimation vector for the colored noise model. After the algorithm converges iteratively, the final estimates of the above parameters are substituted into the Gaussian mixture kernel prediction model to determine the specific form of the prediction model, which can then be applied to practical applications of photovoltaic power prediction.

[0101] Step S5: In the prediction stage, the key meteorological factors of the time to be predicted are input into the hybrid Gaussian kernel prediction model after the parameters are determined, so as to obtain the predicted value of photovoltaic power generation at the time to be predicted.

[0102] After the AF-EGI algorithm identifies the parameters and determines the final parameter estimates, the predicted value of the photovoltaic output power can be obtained. as follows:

[0103] in, Indicates the first Predicted photovoltaic output power at any given time; Indicates the first The solar irradiance input value acquired at any time. and This represents the final estimated values ​​of the weight parameters of the two Gaussian kernel functions; and This represents the final estimated value of the sensitivity coefficients of the two Gaussian kernel functions; and This represents the final estimated value of the central parameters of the two Gaussian kernel functions; Indicates the order of the colored noise model; Indicates the first The final estimated values ​​of the parameters of the colored noise model; Indicates the first The historical prediction error estimate for a given time.

[0104] Without loss of generality, assume that the initial conditions are satisfied: Based on this, the historical prediction error estimate is defined as:

[0105] When the first After obtaining the measured photovoltaic power value at time 1, the prediction error estimate for that time is calculated and used for photovoltaic power prediction at subsequent times.

[0106] Specifically, to identify these parameters, this embodiment introduces a data filter. By changing the structure of the interference noise model, the influence of colored noise on parameter estimation is effectively reduced, and a batch processing method is used to fully utilize the data information. Based on this, the identified mixed Gaussian kernel prediction model can not only characterize the nonlinear response of photovoltaic power to irradiance, but also capture the time-domain dependence of prediction errors, thereby improving prediction accuracy.

[0107] The following specific embodiments verify the solutions involved in the above embodiments. This invention employs the following methods in the algorithm derivation process: A colored noise model of order 1 is used to ensure the applicability of the proposed parameter identification method to different orders of colored noise models. In specific embodiments, a comprehensive consideration is taken into account... A second-order colored noise model is constructed. This order setting is only used to illustrate one specific embodiment of the present invention and does not constitute a limitation on the order of the colored noise model or the scope of protection of the present invention.

[0108] To verify the practical applicability of the proposed modeling method, the Gaussian mixture kernel prediction model was applied to photovoltaic power prediction on a real photovoltaic dataset. To evaluate the effectiveness of the model in the constructed prediction scenario, actual power generation data from a photovoltaic power plant was used to validate the algorithm. This dataset includes variables such as solar irradiance and photovoltaic output power, used for model training and prediction performance evaluation. Long-term prediction experiments were conducted to evaluate the prediction performance and generalization ability of the proposed model under real noise and complex environments. To evaluate the effectiveness of the AF-EGI algorithm in photovoltaic power generation prediction, it was compared with the traditional adaptive augmented gradient iterative (AEGI) algorithm. Both algorithms were used to identify the unknown parameters in the Gaussian mixture kernel prediction model to obtain final parameter estimates that satisfy preset convergence conditions. Then, the identified models were applied to photovoltaic prediction, and the prediction performance of the models under the two algorithms was compared. The invention is further illustrated below with reference to the accompanying figures and example simulations.

[0109] It should be noted that, in the process of predicting photovoltaic power, in order to comprehensively evaluate the predictive performance of the model, this invention uses the following four indicators for evaluation: root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 ) and Mean Absolute Percentage Error (MAPE). The specific calculation formula is:

[0110] in Represents the total number of samples. Indicates the first The actual photovoltaic power of each sample This represents the corresponding predicted value. This represents the average value of the actual photovoltaic power.

[0111] By comprehensively evaluating the above multiple indicators, we can more objectively and accurately compare the performance of the model in the prediction task, and verify the effectiveness and reliability of the photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling in the task.

[0112] Example: This example provides a photovoltaic (PV) power prediction method based on Gaussian mixture kernel function modeling. It includes conducting PV power prediction experiments using a large-sample, long-term dataset to evaluate the model's generalization ability across different time periods. The dataset used in this example collects data on varying light intensity to help the model learn the regular changes in PV power generation and improve its predictive ability. The model uses irradiance as the input variable and PV output power as the target variable. During training, the model learns the mapping relationship between input and output; during testing, the model predicts PV output power based on irradiance data.

[0113] RMSE and R on the training and test sets 2 A comparison of MAE and MAPE is shown in Table 2.

[0114] Table 2: RMSE and R on the training and test sets 2 Comparison of MAE and MAPE

[0115] A comparison of the error histograms of the AEGI and AF-EGI algorithms on the test set is attached. Figure 5 As shown; a comparison of the regression plots of the AEGI and AF-EGI algorithms on the test set is shown below. Figure 6 As shown; a comparison of the predicted outputs and actual values ​​of the AEGI and AF-EGI algorithms on the test set is shown below. Figure 7 As shown in the figure, the statistical and visualization results demonstrate that the AF-EGI method and the model it identifies can effectively capture the changing patterns of photovoltaic power, improve the overall prediction accuracy, and perform well in various comprehensive evaluation indicators. It not only has a good ability to characterize the changes in the shape of the power curve, but also has better generalization performance.

[0116] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling, characterized in that, include: S1. Conduct correlation analysis on meteorological factors, screen out the meteorological factors with the strongest correlation with photovoltaic power output, and record them as key meteorological factors; S2. Combine regression residuals and the isolated forest method to jointly clean the acquired photovoltaic data, obtain the photovoltaic power prediction dataset, and divide it into training set and test set; S3. Construct a mixture Gaussian kernel prediction model with key meteorological factors as input; S4. Using the adaptive filtering augmented gradient iterative algorithm, the parameters of the mixed Gaussian kernel prediction model are identified using the photovoltaic power prediction training set until the model parameters meet the preset convergence accuracy. The final parameter estimates that meet the preset convergence conditions are obtained, and the mixed Gaussian kernel prediction model under the final parameter estimates is obtained. S5. In the prediction stage, the key meteorological factors at the time to be predicted are input into the mixed Gaussian kernel prediction model under the final parameter estimate to obtain the predicted value of photovoltaic power generation at the time to be predicted.

2. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 1, characterized in that, Step S1 specifically includes: performing correlation analysis on meteorological factors. Specifically, this involves using the Pearson correlation coefficient to perform correlation analysis on the meteorological factors.

3. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 2, characterized in that, Step S3 specifically includes: the key meteorological factor is solar irradiance.

4. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 3, characterized in that, Step S4 specifically includes: using an adaptive filtering augmented gradient iterative algorithm, the parameters of the mixed Gaussian kernel prediction model are identified using the photovoltaic power prediction training set until the model parameters meet the preset convergence accuracy, thus obtaining the final parameter estimates that meet the preset convergence conditions; specifically including: Step S41: Start the Gaussian mixture kernel prediction model and initialize it; Step S42: Use key meteorological factors from a set of data in the photovoltaic power prediction training set as input data, and use the measured value of photovoltaic power generation at the corresponding time as output data. Step S43, construct the information vector; Step S44: Construct the filtered information vector; Step S45: Determine the iteration step size, update the parameter estimates, and calculate the estimates of colored noise and white noise; Step S46: Determine whether the model meets the accuracy requirements. If not, return to step S42. If it does, obtain the parameter estimation results, that is, determine the final parameter estimates that meet the preset convergence conditions.

5. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 4, characterized in that, The expression for the Gaussian mixture kernel prediction model is: ； in, Indicates the input to the prediction model. This represents the output of the prediction model. This indicates colored noise interference. Represented as: ， ， Right now ， in, Represents the unit delay operator ( ); Indicates the order of the colored noise model; The mean is zero and the variance is . Random white noise, ( ) indicates the first Parameters of the colored noise model; , and ( ) represents the feature parameters of the Gaussian mixture model. ( ) represents the linear weighting coefficient of photovoltaic power output with respect to different light levels; ( The sensitivity coefficient of photovoltaic power to changes in sunlight; ( The reference center represents the level of illumination. ( () represents the moving average coefficient of photovoltaic noise. It is white noise.

6. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 5, characterized in that, In the Gaussian mixture kernel model, the unknown feature parameters are defined as a parameter vector: ； Integrated parameter vector: ； Define the information vector as: ； After processing, the identification model of the Gaussian mixture kernel prediction model is obtained, and its expression is: ； In step S41, the identification model is activated.

7. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 6, characterized in that, Step S43 specifically includes: constructing and ; ； in, Sampling time; For data length; For iteration variables; For the first The iteration yielded Parameter estimates; For the first The iteration yielded Parameter estimates; For the first The information vector obtained from the next iteration; For noise The Sub-iteration estimation; For the first The noise information vector obtained from the next iteration.

8. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 7, characterized in that, Step S44 specifically includes: designing a data filter. ,calculate , , , and ; Design a filter ,in The structural expression representing colored noise; Define filter output Filtered input and information vector : ； The original identification model is multiplied by the filter at both ends. , on observation data and information vector After filtering, the filtered identification model is obtained as follows: ； Using noise model parameter vectors Estimate Constructing filters Estimate for: ； in, Indicates the first Colored noise structure polynomial in the next iteration The estimated value; Estimation using filters replace The unknown quantity can be obtained. , and Estimate , and for: ； definition and The formula is: ； in, This represents the sensitivity estimation vector relative to the Gaussian kernel nonlinear parameters; express and The augmented information vector is used to uniformly update the parameters of the Gaussian mixture kernel model.

9. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 8, characterized in that, In step S45, the iteration step size is defined. and The formula is: ； in, Represents the parameter vector The iteration step size; Represents the parameter vector The iteration step size; definition and The formula is: ； in, Indicates the first The colored noise estimate obtained in the next iteration; Indicates the first The white noise estimate obtained in the next iteration; Gaussian kernel mixture model parameter estimation vector and The formula for iterative updates is: ； in, Indicates the first The parameter estimation vector of the mixture Gaussian kernel model obtained in the second iteration Indicates the first The parameter estimation vector of the colored noise model obtained in the next iteration.

10. The photovoltaic power generation prediction method based on Gaussian mixture kernel function modeling according to claim 9, characterized in that, Step S46 specifically includes: determining whether the model meets the accuracy requirements; if... If yes, return to step S42; otherwise, obtain... , and The parameter estimation result is the final parameter estimate that satisfies the preset convergence condition. The final Gaussian kernel mixture prediction model based on the parameter estimates is as follows: ； in, Indicates the first Predicted photovoltaic output power at any given time; Indicates the first The solar irradiance input acquired at any time; and Indicates the first The iteration yielded and The final estimated value; and Indicates the first The iteration yielded and The final estimated value; and Indicates the first The iteration yielded and The final estimated value; Indicates the order of the colored noise model; Indicates the first The final estimated values ​​of the parameters of the colored noise model; Indicates the first The estimated historical prediction error at any given time.

Images