Photovoltaic power generation power prediction method based on fuzzy C-means mixed density network

Through the fuzzy C-mean mixed density network method, the problem of insufficient discussion of uncertainty in photovoltaic power generation prediction is solved, the reliability and accuracy of the prediction are improved, and uncertainty information is provided to help manage photovoltaic power generation systems.

CN119009958APending Publication Date: 2024-11-22STATE GRID JIANGSU ELECTRIC POWER CO XUZHOU POWER SUPPLY CO
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411037824.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing photovoltaic power prediction methods mainly focus on the expected value of the predicted value, and the degree of uncertainty is not fully discussed, which leads to difficulties in coping with uncertainty fluctuations and increases backup power supply and costs.

Method used

Using a method based on fuzzy C-mean mixed density network, data clusters are divided by fuzzy C-mean clustering method, and a mixed density network is trained on each cluster. Combining the negative log likelihood function and the stochastic gradient descent algorithm, the predicted expected value and probability are output. Density function and weighted average by membership weights to obtain the confidence interval of the predicted value.

Benefits of technology

It improves the reliability and accuracy of photovoltaic power prediction, provides uncertainty information on the predicted value, and helps better plan and manage photovoltaic power generation systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119009958A_ABST
    Figure CN119009958A_ABST
Patent Text Reader

Abstract

The invention discloses a photovoltaic power generation power prediction method based on a fuzzy C-means mixed density network, and the method comprises the following steps: S1, collecting the output data and weather data of a photovoltaic panel, and carrying out the feature engineering; s2, dividing the data into a plurality of fuzzy subsets based on a fuzzy C-means clustering method, training a mixed density network in each fuzzy subset, adopting a negative logarithm likelihood function and a stochastic gradient descent algorithm, and using a root-mean-square error to evaluate; and S3, outputting a predicted expected value of the photovoltaic power generation power on each cluster in the FCM stage, outputting a probability density function of the photovoltaic power generation power on each cluster in the MDN stage, then performing weighted average on the predicted expected value and the probability density function in combination with the membership weight to obtain a final prediction result, and obtaining a confidence interval of the predicted value. According to the invention, the reliability and accuracy of the prediction result can be obtained more scientifically, so that the photovoltaic power generation system can be planned and managed better.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a photovoltaic power prediction method based on a fuzzy C-mean mixed density network, and belongs to the technical field of photovoltaic power prediction. Background Art

[0002] Compared with fossil energy, solar energy is very abundant in nature and is an important focus for achieving low-carbon power systems. A notable feature of photovoltaics is the uncertainty of their power generation, which poses a challenge to the absorption of photovoltaics and the stability of the power system. Therefore, it is necessary to predict photovoltaic power generation as accurately as possible.

[0003] At present, most of the predictions of photovoltaic power generation only focus on the expected value of the prediction, but the discussion on the uncertainty of the prediction is not sufficient. In order to cope with the fluctuation of uncertainty, additional backup power and its additional cost may be required. In order to help better understand the reliability of the prediction results, it is necessary to obtain relevant information about the uncertainty of the prediction value while predicting the photovoltaic power generation. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a photovoltaic power generation prediction method based on fuzzy C-mean hybrid density network, which can more scientifically obtain the reliability and accuracy of the prediction results, so as to better plan and manage the photovoltaic power generation system.

[0005] In order to achieve the above object, the present invention adopts a photovoltaic power prediction method based on fuzzy C-mean mixed density network, which includes the following steps:

[0006] S1, collect photovoltaic panel output data and weather data, and perform feature engineering;

[0007] S2, based on the fuzzy C-means clustering method, the data is divided into several fuzzy subsets, and a mixed density network is trained in each fuzzy subset, using the negative log-likelihood function and the stochastic gradient descent algorithm, and the root mean square error is used for evaluation;

[0008] S3. In the FCM stage, the predicted expected value of the photovoltaic power generation power on each cluster is output, and in the MDN stage, the probability density function of the photovoltaic power generation power on each cluster is output. Then, the predicted expected value and the probability density function are weighted averaged in combination with the membership weight to obtain the final prediction result, and the confidence interval of the predicted value is obtained.

[0009] In some embodiments, the step S1 collects power generation data of photovoltaic panels, regional weather data, grid load data, and historical maintenance record data of the photovoltaic power plant in the past two years.

[0010] In some embodiments, feature engineering is performed on the collected data in combination with a scientific computing library, including:

[0011] For the power generation data of photovoltaic panels, regional weather data, and power grid load data, extreme outliers are removed, and missing values ​​are filled by interpolation or mean method, and normalization is performed at the same time.

[0012] In some embodiments, the historical maintenance record data is converted into a 0 / 1 variable, indicating that a fault has occurred or not occurred in the equipment.

[0013] In some embodiments, the implementation process of step S2 is as follows:

[0014] (1) Initializing the initial value of the fuzzy membership, which is given randomly or based on prior information about the acquired data;

[0015] (2) Taking the current membership value as the weight of each sample, the center point of each cluster is calculated by taking the weighted average of each sample;

[0016] (3) Update the membership value of each sample by the distance between each sample and the cluster center of the previous step;

[0017] (4) Determine whether the difference between the new membership value and the previous membership value is less than a preset threshold. If so, it is determined to be converged; otherwise, return to iterative calculation.

[0018] In some embodiments, the implementation process of step S3 is as follows:

[0019] (1) A mixed density network model is trained on each cluster. The stochastic gradient descent algorithm is used for training, and the root mean square error is used as the evaluation index of the model. In the prediction stage, the FCM algorithm is first used to classify the feature vector xi input at the i-th moment into a certain cluster, and then the probability density function of the power generation at that moment is predicted according to the MDN model corresponding to the cluster;

[0020] (2) The probability density function predicted by each cluster is weighted according to its weight to obtain the final predicted probability density function, and then the confidence level is determined. The confidence interval is calculated based on the sample size and is used as the final prediction together with the predicted expected value.

[0021] Compared with the prior art, the photovoltaic power generation prediction method based on the fuzzy C-means mixed density network of the present invention can divide the data into multiple clusters based on the mixed density function of fuzzy clustering, and then train the mixed density network model on each cluster. This method can adaptively adjust the weight of the kernel function according to the fuzzy nature of the data, thereby improving the prediction accuracy and robustness, which is of great significance to the planning, operation and management of low-carbon distribution systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 It is a schematic diagram of the process of the present invention;

[0024] Figure 2 It is a schematic diagram of the feature engineering process in the present invention;

[0025] Figure 3 This is a schematic diagram of the data clustering process based on the fuzzy C-means clustering method;

[0026] Figure 4 The figure is a schematic diagram of the training and prediction process based on the mixed density network model. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present application are described in detail below with the help of accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solutions of the present application, rather than limitations on the technical solutions of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0028] like Figure 1-Figure 4 As shown, a photovoltaic power generation power prediction method based on fuzzy C-means mixed density network includes the following steps:

[0029] S1, collect photovoltaic panel output data and weather data, and perform feature engineering;

[0030] The implementation process is as follows:

[0031] (1) Collect data on the power generation of photovoltaic panels, regional weather data, grid load data, and historical maintenance records of photovoltaic power plants in the past two years;

[0032] (2) Combine scientific computing libraries such as numpy and pandas to perform feature engineering on the collected data;

[0033] The preprocessing of the first three types of data (data on power generation of photovoltaic panels, regional weather data, and grid load data) mainly involves removing extreme outliers and filling missing values ​​through interpolation or mean method. At the same time, normalization is performed to prevent the model from being more sensitive to features with larger values ​​under different dimensions due to the different dimensions of each type of data.

[0034] For historical maintenance record data, it can be converted into a 0 / 1 variable, indicating that the equipment has or has not failed, such as Figure 2 As shown;

[0035] S2, based on the fuzzy C-means clustering method, the data is divided into several fuzzy subsets, and a mixed density network is trained in each fuzzy subset, using the negative log-likelihood function and the stochastic gradient descent algorithm, and the root mean square error is used for evaluation;

[0036] The implementation process is as follows:

[0037] First, an initialization operation is performed, that is, a data sample set after feature engineering in step S1 and an initial value of the fuzzy membership are given. The initial value is given randomly or based on prior information about the acquired data.

[0038] Secondly, the current membership value is used as the weight of each sample, and the i The weighted average method is used to calculate the center point of each cluster V = [v1, v2, ..., v c ], and then calculate the distance d between each sample and the cluster center of the previous step ij To update the membership value of each sample, the calculation formula of Euclidean distance is as follows:

[0039] d ij =|x i -v j |;

[0040] Where, d ij For sample x i To the sample center point v j The Euclidean distance of

[0041] Then, it is determined whether the difference between the new membership value and the previous membership value is less than the preset threshold. If so, it is determined to be converged, otherwise it returns to the second step of iterative calculation;

[0042] Finally, the Figure 3 The schematic diagram of the data clustering process based on the fuzzy C-means clustering method shown;

[0043] S3. Output the predicted expected value of photovoltaic power generation on each cluster in the FCM stage, and output the probability density function of photovoltaic power generation on each cluster in the MDN stage. The specific calculation method is as follows:

[0044]

[0045] In the formula, π k (x) is the mixing coefficient, which represents the weight of each Gaussian distribution; μ k (x) represents the mean of the kth Gaussian distribution; σ k (x) represents the standard deviation of the kth Gaussian distribution; N(y|μ k ,σ k ) indicates that the mean is μ k , standard deviation is σ k Normal distribution of

[0046] Then, the predicted expected value and probability density function are weighted averaged in combination with the membership weights to obtain the final prediction result, and the confidence interval of the predicted value is obtained;

[0047] The implementation process is as follows:

[0048] First, for each cluster formed in the previous step, a mixed density network model is trained. The optimization method during training can be the stochastic gradient descent algorithm, and the root mean square error is used as the evaluation index of the model.

[0049] Secondly, in the prediction stage, the FCM algorithm is first used to classify the feature vector xi input at the i-th moment into a certain cluster, and then the probability density function of the power generation at that moment is predicted according to the MDN model corresponding to the cluster;

[0050] Finally, the probability density function predicted by each cluster is weighted according to its weight to obtain the final predicted probability density function, and then the confidence level (such as 95%) is determined, and the confidence interval is calculated based on the sample size, and the confidence interval is used together with the predicted expected value as the final prediction result. Figure 4 Schematic diagram of the training and prediction process based on the mixed density network model shown.

[0051] The photovoltaic power generation power prediction method based on the fuzzy C-means mixed density network of the present invention can divide the data into multiple clusters based on the mixed density function of fuzzy clustering, and then train the mixed density network model on each cluster. This method can adaptively adjust the weight of the kernel function according to the fuzzy nature of the data, thereby improving the prediction accuracy and robustness, which is of great significance to the planning, operation and management of low-carbon distribution systems.

[0052] In addition, those skilled in the art will understand that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments also means being within the scope of protection of the present invention and forming different embodiments. For example, in the above embodiments, those skilled in the art can use them in combination according to the known technical solutions and the technical problems to be solved by this application.

Claims

1. A photovoltaic power prediction method based on fuzzy C-means mixed density network, characterized in that: The following steps are involved: S1, collect photovoltaic panel output data and weather data, and perform feature engineering; S2, based on the fuzzy C-means clustering method, the data is divided into several fuzzy subsets, and a mixed density network is trained in each fuzzy subset, using the negative log-likelihood function and the stochastic gradient descent algorithm, and the root mean square error is used for evaluation; S3. In the FCM stage, the predicted expected value of the photovoltaic power generation power on each cluster is output, and in the MDN stage, the probability density function of the photovoltaic power generation power on each cluster is output. Then, the predicted expected value and the probability density function are weighted averaged in combination with the membership weight to obtain the final prediction result, and the confidence interval of the predicted value is obtained.

2. The photovoltaic power prediction method based on fuzzy C-means mixed density network according to claim 1 is characterized in that: In the step S1, the power generation data of the photovoltaic panels of the photovoltaic power plant in the past two years, the regional weather data, the grid load data, and the historical maintenance record data are collected.

3. The photovoltaic power prediction method based on fuzzy C-means mixed density network according to claim 2 is characterized in that: Combine scientific computing libraries to perform feature engineering on the collected data, including: For the power generation data of photovoltaic panels, regional weather data, and power grid load data, extreme outliers are removed, and missing values ​​are filled by interpolation or mean method, and normalization is performed at the same time.

4. The photovoltaic power generation prediction method based on fuzzy C-means mixed density network according to claim 3 is characterized in that: For historical maintenance record data, they are converted into 0 / 1 variables, indicating that the equipment has or has not failed.

5. The photovoltaic power generation prediction method based on fuzzy C-means mixed density network according to claim 1 is characterized in that: The implementation process of step S2 is as follows: (1) Initializing the initial value of the fuzzy membership, which is given randomly or based on prior information about the acquired data; (2) Taking the current membership value as the weight of each sample, the center point of each cluster is calculated by taking the weighted average of each sample; (3) Update the membership value of each sample by the distance between each sample and the cluster center of the previous step; (4) Determine whether the difference between the new membership value and the previous membership value is less than a preset threshold. If so, it is determined to be converged; otherwise, return to iterative calculation.

6. The photovoltaic power generation prediction method based on fuzzy C-means mixed density network according to claim 1 is characterized in that: The implementation process of step S3 is as follows: (1) A mixed density network model is trained on each cluster. The stochastic gradient descent algorithm is used for training, and the root mean square error is used as the evaluation index of the model. In the prediction stage, the FCM algorithm is first used to classify the feature vector xi input at the i-th moment into a certain cluster, and then the probability density function of the power generation at that moment is predicted according to the MDN model corresponding to the cluster; (2) The probability density function predicted by each cluster is weighted according to its weight to obtain the final predicted probability density function, and then the confidence level is determined. The confidence interval is calculated based on the sample size and is used as the final prediction together with the predicted expected value.

Citation Information

Cited By

  • Environment data acquisition method and system for vehicle-mounted refrigerator

    CN119917790A

  • Section transmission power probability prediction method, apparatus and device, and storage medium

    CN120165366A

  • A section transmission power probability prediction method, device, equipment and storage medium

    CN120165366B

  • Large-screen data visualization method, system and equipment and storage medium

    CN120337860A