A method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM

By combining the VMD-KELM method with grey relational analysis and an improved GMM clustering algorithm, and utilizing VMD variational mode decomposition and KELM kernel extreme learning machine, the accuracy problem of photovoltaic power generation prediction under extreme weather conditions was solved, achieving more efficient photovoltaic power generation prediction and grid stability.

CN118568529BActive Publication Date: 2026-11-10NANJING NORMAL UNIVERSITY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410622778.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-17
Publication Date
2026-11-10
Estimated Expiration
2044-05-17

AI Technical Summary

Technical Problem

Existing photovoltaic power generation forecasting methods do not take meteorological factors into account in a comprehensive manner, resulting in low forecast accuracy. In particular, the output of photovoltaic power plants fluctuates greatly under extreme weather conditions, affecting the stability of the power grid.

Method used

The VMD-KELM-based method is adopted to extract meteorological elements affecting photovoltaic power generation through grey relational analysis, improve the GMM clustering algorithm for data clustering, use VMD variational mode decomposition for stabilization processing, and use KELM kernel extreme learning machine for high-dimensional inner product operation to reconstruct the prediction results.

Benefits of technology

It improves the accuracy and efficiency of photovoltaic power generation forecasting, especially under extreme weather conditions, enabling more precise forecasting of photovoltaic power output and maintaining the stability and efficient operation of the power system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118568529B_ABST
    Figure CN118568529B_ABST
Patent Text Reader

Abstract

This invention discloses a method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM. The method includes: performing grey relational analysis on environmental factors affecting photovoltaic power generation to extract typical meteorological elements affecting photovoltaic power generation efficiency; selecting similar days based on the GMM clustering algorithm and improving the GMM clustering algorithm by introducing iterative degrees of freedom to extract differential load characteristics of power generation, and improving the iterative degrees of freedom during the extraction process; using VMD variational mode decomposition to perform Fourier decomposition on the power prediction data of the photovoltaic power station for stationarization processing; and using KELM kernel extreme learning machine to transform the complex spatial problem into a high-dimensional inner product operation problem, modeling and predicting the mode function at each scale. This invention improves the accuracy of meteorological data classification; uses VMD to decompose photovoltaic data, reducing data non-stationarity; and uses KELM to restructure the subsequences, improving the model's prediction accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of photovoltaic forecasting for new energy power generation and grid connection, and particularly to a method for forecasting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM. Background Technology

[0002] The development and utilization of new energy sources has become a major research direction for the present and for a long time to come. New energy power generation, such as wind and solar power, can develop and utilize these renewable energy sources. New energy power generation not only increases the overall power output of the power generation process but also reduces pollution and improves the Earth's ecological environment. However, new energy power generation technologies are not yet mature, and their intermittent and fluctuating characteristics affect power generation output. Weather also has a significant impact. Therefore, to address the uncertainties in new energy power generation, accurate prediction of power generation probability is crucial. However, current power generation prediction methods are not comprehensive enough; some methods do not consider meteorological factors, and some methods suffer from low prediction accuracy. In recent years, with the intensification of global warming, extreme weather events have become more frequent. Because photovoltaic power plant output is highly susceptible to environmental factors, its output is intermittent and fluctuating. Under extreme weather conditions, photovoltaic power plants may experience a sharp drop in power generation or even shut down. Large-scale grid connection of photovoltaic power plants can have a significant impact on system stability.

[0003] To improve the accuracy of photovoltaic power output forecasting and maintain the stable and efficient operation of the power system, a photovoltaic power generation forecasting method with high accuracy, forecasting efficiency, and universality is proposed, which is crucial for the economic dispatch of the power grid. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a method for predicting photovoltaic power generation under major extreme weather conditions based on VMD-KELM.

[0005] Technical solution: The present invention includes the following steps:

[0006] (1) A grey relational analysis was conducted on the environmental factors affecting photovoltaic power generation to extract typical meteorological elements that affect photovoltaic power generation efficiency;

[0007] (2) Radiation, ambient temperature and wind speed are the main meteorological factors affecting photovoltaic output. Similar days are selected according to the GMM clustering algorithm, and iterative degrees of freedom are introduced to improve the GMM clustering algorithm. Then, based on the typological meteorological data and combined with the demand factor method, the differential load characteristics of power generation are extracted. During the extraction process, the iterative degrees of freedom are improved, and meteorological clustering under extreme weather conditions is selected to establish a sample of similar photovoltaic power days.

[0008] (3) Use VMD variational mode decomposition to perform Fourier decomposition on the power prediction data of photovoltaic power plants and perform stabilization processing.

[0009] (4) Based on the KELM kernel extreme learning machine, the complex spatial problem is transformed into a high-dimensional inner product operation problem. The modal function of each scale is modeled and predicted. Finally, the predicted subsequence is reconstructed to realize the prediction of photovoltaic power output under severe weather conditions.

[0010] Further, step (1) includes calculating the grey relational degree between cloud cover, relative humidity, ambient temperature, wind speed, air pressure, radiation and photovoltaic power generation efficiency. From the quantitative value of the grey relational degree, it is found that radiation has the greatest impact on photovoltaic power generation efficiency, followed by wind speed, ambient temperature, cloud cover, relative humidity and air pressure.

[0011] Furthermore, step (2) of improving the GMM clustering algorithm includes the following steps:

[0012] (2.1) Using radiation, ambient temperature and wind speed as the main meteorological factors affecting photovoltaic output, an initial meteorological data matrix was established to analyze the correlation of the data;

[0013] (2.2) Extract the data with the highest similarity;

[0014] (2.3) Process meteorological data to obtain five types of meteorological data: lightning, precipitation, clear clouds, wind and snow, and visual impairment. Select meteorological clusters under extreme weather conditions, and select five weather types: thunderstorm, hail solid precipitation, clear sky, snowstorm and sandstorm to establish daily samples of photovoltaic power similarity.

[0015] Further, step (2.2) includes:

[0016] Optimal parameters are introduced to determine the similarity of meteorological data. Euclidean distance is used to calculate the similarity between meteorological data, and the most suitable distance metric and optimal similarity threshold are selected. However, the clustering process suffers from variations in the distribution of meteorological data, which disturbs the clustering results. Therefore, the optimal parameters for the extraction process are improved, and the improved optimal parameters are expressed as follows:

[0017]

[0018] Where α represents the optimal parameter; θ represents the dimension of the meteorological data; and e represents the natural constant. Then, a similarity threshold is determined, and the meteorological data is categorized. The GMM clustering algorithm is introduced to cluster and classify the meteorological data. During the clustering process, iterative degrees of freedom are introduced to help determine the optimal clustering result. The clustering classification model is constructed as follows:

[0019]

[0020] Among them, a ij (·) represents similarity; m represents the amount of meteorological data collected; c represents the degrees of freedom for clustering iteration; the degrees of freedom for the data clustering process are improved, and the improved degrees of freedom for clustering iteration are:

[0021]

[0022] Where M represents the correlation coefficient of meteorological data, and X represents the maximum log-likelihood function.

[0023] Further, step (3) includes:

[0024] (3.1) The original signal f is decomposed into K mode functions with finite bandwidth. Each mode function is subjected to Hilbert transform for signal analysis and its one-sided spectrum is obtained.

[0025] (3.2) Adjust each mode function u k The center frequency ω of (t) k An exponential term is introduced, and the spectrum of each mode is modulated accordingly to adapt to the fundamental frequency bandwidth;

[0026] (3.3) The fundamental bandwidth of each mode function is estimated by calculating the gradient 2 norm of the demodulated signal, and a variational model with constraints is constructed as follows:

[0027]

[0028]

[0029] Among them, {u k} and {ω k} are the modality set and the centrality set, respectively; is the partial derivative operator; * is the convolution operator; δ(t) is the unit impulse function;

[0030] (3.4) Step: Update the functional: For all ω≥0, update the functional. and ω k Construct the following model:

[0031]

[0032]

[0033] in, The current remaining amount Wiener filtering; f(ω) is the center frequency of the power spectrum of the current mode function. and They are f(t), u(t), λ(t) and respectively. Fourier transform.

[0034] Furthermore, the transformation of the constrained variational model into an unconstrained variational model in step (3.3) includes: leveraging the advantages of the quadratic penalty term and the Lagrange multiplier method, an augmented Lagrangian function is introduced, as shown below:

[0035]

[0036] Furthermore, step (4) maps the original data to a high-dimensional kernel space using KELM, and directly implements the dot product operation between samples using the kernel function. The KELM model includes an input layer, a hidden layer, and an output layer.

[0037] Furthermore, the input layer is used to receive raw data, which consists of n features, represented as an input vector X = [x1, x2, ..., xn]. n ].

[0038] Furthermore, the hidden layer is implemented using the kernel function K(x) i x j ) Calculate the mapping value between the input data x and each hidden layer neuron, where x j These represent the centers or sample points of the hidden layer neurons. These mapping values ​​constitute the output of the hidden layer neurons. The kernel function is shown in the following equation:

[0039]

[0040] Where g represents the kernel parameter, and each neuron in the hidden layer has a weight vector w and a bias term b. These weights and bias terms are typically randomly initialized and then determined during training by solving linear equations. For N different samples, the output of the hidden layer neuron is represented as:

[0041] Hw=T

[0042]

[0043] Where H is the output of the hidden layer; w is the weight vector; T is the target output matrix; and M is the number of neurons in the hidden layer.

[0044] Furthermore, the output layer receives the output of the hidden layer, which is an M-dimensional vector. The output of the output layer is the predicted output result, and its function is as follows:

[0045]

[0046] Where I is the identity matrix and y(x) is the final output of KELM.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: by introducing iterative degrees of freedom to perform more accurate clustering of the original meteorological data, the accuracy of meteorological data classification is improved; by using VMD to decompose photovoltaic data, the non-stationarity of the data is reduced, and by using KELM to restructure the subsequences, the prediction accuracy of the model is improved; in the prediction model, not only is the impact of environmental factors on the output of photovoltaic power plants considered, but also the changes in photovoltaic power generation under extreme weather conditions, which have not been considered in previous studies. Attached Figure Description

[0048] Figure 1 This is a flowchart of the present invention;

[0049] Figure 2 Here is a flowchart of the grey relational analysis process;

[0050] Figure 3 This is a diagram of the KELM model structure. Detailed Implementation

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0052] like Figure 1 The flowchart shown is a process of the present invention, including the following steps:

[0053] (1) First, a grey relational analysis was conducted on the environmental factors affecting photovoltaic power generation to extract typical meteorological elements that affect photovoltaic power generation efficiency;

[0054] (2) Based on step 1, using radiation, ambient temperature, and wind speed as the main meteorological factors affecting photovoltaic output, similar days are selected according to the GMM clustering algorithm. Iterative degrees of freedom are introduced to improve the GMM clustering algorithm. Then, based on the typological meteorological data and combined with the demand factor method, the differential load characteristics of power generation are extracted. During the extraction process, the iterative degrees of freedom are improved, and meteorological clusters under extreme weather conditions are selected to establish a sample of similar photovoltaic power days.

[0055] like Figure 2 The flowchart shown is a grey relational analysis program used to calculate the grey relational degree between cloud cover, relative humidity, ambient temperature, wind speed, air pressure, radiation and photovoltaic power generation efficiency. From the quantitative values ​​of the grey relational degree, we can see that radiation has the greatest impact on photovoltaic power generation efficiency, followed by wind speed, ambient temperature, cloud cover, relative humidity and air pressure.

[0056] Improving the GMM clustering algorithm involves the following steps:

[0057] (2.1) Using radiation, ambient temperature and wind speed as the main meteorological factors affecting photovoltaic output, an initial meteorological data matrix was established to analyze the correlation of the data;

[0058] (2.2) Extracting the maximum similarity data; Optimal parameters are introduced to determine the similarity of meteorological data. Euclidean distance is used to calculate the similarity between meteorological data, and the most suitable distance metric and optimal similarity threshold are selected. The clustering process suffers from variations in the distribution of meteorological data, which disturbs the clustering results. Therefore, the optimal parameters for the extraction process are improved. The improved optimal parameters are expressed as follows:

[0059]

[0060] Where α represents the optimal parameter; θ represents the dimension of the meteorological data; and e represents the natural constant. Then, a similarity threshold is determined, and the meteorological data is categorized. The GMM clustering algorithm is introduced to cluster and classify the meteorological data. During the clustering process, iterative degrees of freedom are introduced to help determine the optimal clustering result. The clustering classification model is constructed as follows:

[0061]

[0062] Among them, a ij (·) represents similarity; m represents the amount of meteorological data collected; c represents the degrees of freedom for clustering iteration; the degrees of freedom for the data clustering process are improved, and the improved degrees of freedom for clustering iteration are:

[0063]

[0064] Where M represents the correlation coefficient of meteorological data, and X represents the maximum log-likelihood function.

[0065] (2.3) Process meteorological data to obtain five types of meteorological data: lightning, precipitation, clear clouds, wind and snow, and visual impairment. Select meteorological clusters under extreme weather conditions, and select five weather types: thunderstorm, hail solid precipitation, clear sky, snowstorm and sandstorm to establish daily samples of photovoltaic power similarity.

[0066] (3) Based on step (2), use VMD (variational mode decomposition) to perform Fourier decomposition on the power prediction data of the photovoltaic power station and perform stabilization processing;

[0067] (3.1) The original signal f is decomposed into K mode functions with finite bandwidth. Each mode function is subjected to Hilbert transform for signal analysis and its one-sided spectrum is obtained.

[0068] (3.2) Adjust each mode function u k The center frequency ω of (t) kAn exponential term is introduced, and the spectrum of each mode is modulated accordingly to adapt to the fundamental frequency bandwidth;

[0069] (3.3) The fundamental bandwidth of each mode function is estimated by calculating the gradient 2 norm of the demodulated signal, and a variational model with constraints is constructed as follows:

[0070]

[0071]

[0072] Among them, {u k} and {ω k} are the modality set and the centrality set, respectively; is the partial derivative operator; * is the convolution operator; δ(t) is the unit impulse function; transforming a constrained variational model into an unconstrained variational model involves: leveraging the advantages of the quadratic penalty term and the Lagrange multiplier method, an augmented Lagrangian function is introduced, as shown below:

[0073]

[0074] (3.4) Step: Update the functional: For all ω≥0, update the functional. and ω k Construct the following model:

[0075]

[0076]

[0077] in, The current remaining amount Wiener filtering; f(ω) is the center frequency of the power spectrum of the current mode function. and f(t), u(t), λ(t) and respectively Fourier transform.

[0078] (4) Based on step (3), the complex spatial problem is transformed into a high-dimensional inner product operation problem according to KELM (Kernel Extreme Learning Machine). The modal function of each scale is modeled and predicted. Finally, the predicted subsequence is reconstructed to realize the prediction of photovoltaic power output under severe weather conditions.

[0079] KELM maps the original data to a high-dimensional kernel space, and directly implements the dot product operation between samples using kernel functions. The KELM model includes an input layer, hidden layers, and an output layer. Figure 3 The diagram shown is a structural diagram of the KELM model.

[0080] The input layer receives raw data and consists of n features, which are represented as an input vector X = [x1, x2, ..., xn]. n ].

[0081] The hidden layer is passed through the kernel function K(x) i x j ) Calculate the mapping value between the input data x and each hidden layer neuron, where x j These represent the centers or sample points of the hidden layer neurons. These mapping values ​​constitute the output of the hidden layer neurons. The kernel function is shown in the following equation:

[0082]

[0083] Where g represents the kernel parameter, and each neuron in the hidden layer has a weight vector w and a bias term b. These weights and bias terms are typically randomly initialized and then determined during training by solving linear equations. For N different samples, the output of the hidden layer neuron is represented as:

[0084] Hw=T

[0085]

[0086] Where H is the output of the hidden layer; w is the weight vector; T is the target output matrix; and M is the number of neurons in the hidden layer.

Claims

1. A method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM, characterized in that: Includes the following steps: (1) Conduct grey relational analysis on the environmental factors affecting photovoltaic power generation and extract typical meteorological elements that affect photovoltaic power generation efficiency; (2) Radiation, ambient temperature and wind speed are the main meteorological factors affecting photovoltaic output. Similar days are selected according to the GMM clustering algorithm, and iterative degrees of freedom are introduced to improve the GMM clustering algorithm. Then, based on the typological meteorological data and combined with the demand factor method, the differential load characteristics of power generation are extracted. During the extraction process, the iterative degrees of freedom are improved, and meteorological clustering under extreme weather conditions is selected to establish a sample of similar photovoltaic power days. (3) Use VMD variational mode decomposition to perform Fourier decomposition on the power prediction data of photovoltaic power plants and perform stabilization processing; (4) Based on the KELM kernel extreme learning machine, the complex spatial problem is transformed into a high-dimensional inner product operation problem. The modal function of each scale is modeled and predicted. Finally, the predicted subsequence is reconstructed to realize the prediction of photovoltaic power output under severe weather conditions. The improvement of the GMM clustering algorithm in step (2) includes the following steps: (2.1) Using radiation, ambient temperature and wind speed as the main meteorological factors affecting photovoltaic output, an initial meteorological data matrix was established to analyze the data correlation; (2.2) Extract the data with the highest similarity; (2.3) Process meteorological data to obtain five types of meteorological data: lightning, precipitation, clear clouds, wind and snow, and visual impairment. Select meteorological clusters under extreme weather conditions, and select five weather types: thunderstorm, hail solid precipitation, clear sky, snowstorm and sandstorm to establish a daily sample of photovoltaic power similarity. Step (2.2) includes: Optimal parameters are introduced to determine the similarity of meteorological data. Euclidean distance is used to calculate the similarity between meteorological data, and the most suitable distance metric and optimal similarity threshold are selected. The clustering process suffers from variations in the distribution of meteorological data, which disturbs the clustering results. Therefore, the optimal parameters for the extraction process are improved. The improved optimal parameters are expressed as follows: , in, Indicates the optimal parameters; Indicates the dimension of meteorological data; e The natural constant is represented; then, a similarity threshold is determined, and meteorological data are classified; the GMM clustering algorithm is introduced to cluster and classify the meteorological data. During the clustering process, iterative degrees of freedom are introduced to help determine the optimal clustering result, and the clustering classification model is constructed as follows: , Among them, a ij (·) indicates similarity; m This indicates the amount of meteorological data collected; c This represents the degrees of freedom for clustering iteration; the iterative degrees of freedom in the data clustering process are improved, and the improved clustering iterative degrees of freedom are: , Where M represents the correlation coefficient of meteorological data, and X represents the maximum log-likelihood function.

2. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 1, characterized in that: Step (1) includes calculating the grey relational degree between cloud cover, relative humidity, ambient temperature, wind speed, air pressure, radiation and photovoltaic power generation efficiency. From the quantitative value of the grey relational degree, it is found that radiation has the greatest impact on photovoltaic power generation efficiency, followed by wind speed, ambient temperature, cloud cover, relative humidity and air pressure.

3. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 1, characterized in that: Step (3) includes: (3.1) The original signal f Decomposed into K Each mode function with a finite bandwidth is subjected to Hilbert transform for signal analysis, and its one-sided spectrum is obtained. (3.2) Adjust each mode function center frequency An exponential term is introduced, and the spectrum of each mode is modulated accordingly to adapt to the fundamental frequency bandwidth; (3.3) The fundamental bandwidth of each mode function is estimated by calculating the gradient 2 norm of the demodulated signal, and a variational model with constraints is constructed as follows: , , in, and These are the modality set and the centrality set, respectively. It is a partial derivative operator; This is the convolution operator; It is a unit impulse function; (3.4) Step to update the functional: for all Update functional as well as Construct the following model: , , in, The current remaining amount Wiener filtering; The center frequency of the power spectrum of the current mode function; , , and They are respectively , , and Fourier transform.

4. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 1, characterized in that: The step (3.3) of transforming the constrained variational model into an unconstrained variational model includes: leveraging the advantages of the quadratic penalty term and the Lagrange multiplier method, an augmented Lagrangian function is introduced, as shown below: 。 5. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 1, characterized in that: Step (4) maps the original data to a high-dimensional kernel space using KELM, and directly implements the dot product operation between samples using the kernel function. The KELM model includes an input layer, a hidden layer, and an output layer.

6. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 5, characterized in that: The input layer is used to receive raw data, which is composed of... n It consists of several features, which are represented as an input vector. .

7. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 5, characterized in that: The hidden layer is implemented through a kernel function. Calculate input data x The mapping values ​​of each hidden layer neuron, where These represent the center or sample point of the hidden layer neurons. These mapping values ​​constitute the output of the hidden layer neurons. The kernel function is shown in the following formula: , in, g For kernel parameters, each neuron in the hidden layer has a weight vector. w and a deviation term b These weights and biases are typically initialized randomly and then determined during training by solving linear equations. For different sets of samples, the output of the hidden layer neurons is represented as follows: , , Where H is the output of the hidden layer; This is the weight vector; Output the target matrix; It is the number of neurons in the hidden layer.

8. The method for predicting photovoltaic power generation under severe extreme weather conditions based on VMD-KELM according to claim 5, characterized in that: The output layer receives the output of the hidden layer, which is a The output of the output layer is the predicted output, and its function is as follows: , in, I It is the identity matrix. This is the final output of KELM.