A new energy output power uncertainty dynamic evaluation method

Abstract

The application discloses a new energy output power uncertainty dynamic evaluation method, and belongs to the technical field of power system performance evaluation, comprising the following steps: cleaning new energy actual power generation history data, predicted power history data and meteorological history data; constructing a feature data matrix through the new energy actual power generation history data set and reducing the dimension of the history data set; classifying the history data set after the dimension reduction, and constructing M+1 data subsets; constructing and training a classifier; calculating the probability density function and the prediction error percentage under a given significance level by independently applying the improved optimal bandwidth kernel density estimation to the M+1 data subsets; for each time section data, classifying by using the trained classifier, obtaining the corresponding prediction error percentage interval and the prediction interval under the given significance level according to the classification result, and correcting the prediction interval. The application can improve the accuracy of the output power uncertainty dynamic evaluation.

Description

Technical Field

[0001] This invention belongs to the field of power system performance evaluation technology, specifically relating to a dynamic evaluation method for uncertainty in the output power of new energy sources. Background Technology

[0002] Renewable energy generation, with its significant advantages such as being clean and sustainable, continues to increase its share in the power system. The rapid expansion of renewable energy installed capacity is of great significance for achieving a green and low-carbon energy transition, but it also brings many severe challenges to the stable operation and dispatch of the power system.

[0003] Renewable energy power generation is characterized by significant intermittency and volatility, with its output power strongly influenced by natural factors such as solar radiation, wind speed, and temperature. For example, rapid cloud movement can cause significant fluctuations in solar photovoltaic power generation within a short period. As the penetration rate of renewable energy generation continues to increase, the impact of such power uncertainty on the power system becomes increasingly apparent. When renewable energy generation accounts for a small proportion of the power system, traditional synchronous generators, with their strong inertia and stable output characteristics, can effectively maintain the system's frequency and voltage stability. However, with the high proportion of renewable energy integration, the proportion of traditional synchronous generators in the system decreases accordingly, significantly reducing the power system's inertia and weakening its ability to buffer and regulate fluctuations in renewable energy generation. This makes it difficult for the system to respond quickly and effectively to random fluctuations in renewable energy output power, easily triggering large fluctuations in system frequency and voltage, seriously threatening the safe and stable operation of the power system.

[0004] To address the uncertainties of renewable energy generation, power systems need to reserve sufficient capacity to ensure a reliable power supply even when renewable energy output is insufficient or load suddenly increases. Excessive reserve capacity can lead to reduced utilization of generating equipment and increased generation costs; insufficient reserve capacity can result in power shortages and affect power supply reliability. Accurately assessing the uncertainty of renewable energy output is crucial for rationally determining system reserve capacity. While interval forecasting can quantify the fluctuation range of renewable energy output to some extent, current technologies lack effective analytical methods for assessing the predicted output interval. For example, existing methods rely on a single mathematical distribution to fit the distribution pattern of renewable energy output, or use the Monte Carlo method to simulate the distribution pattern. With the development of distributed generation technology, the randomness of the combined output of different types of renewable energy no longer follows a single mathematical distribution, and the Monte Carlo simulation method is computationally intensive, facing many limitations in practical applications. Current technologies do not consider the differences in prediction intervals under different weather conditions and at different predicted power levels, thus failing to accurately predict the uncertainty of renewable energy output.

[0005] Existing bandwidth selection methods in kernel density estimation, when applied to the application scenario of this invention, result in low bandwidth accuracy. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A dynamic evaluation method for uncertainty in the output power of a new energy source includes:

[0008] Step 1: Clean historical data of actual power generation, historical data of predicted power, and historical meteorological data of new energy sources, including handling missing values, duplicate values, erroneous values, and data standardization;

[0009] Step 2: Construct a feature data matrix using historical datasets of actual power generation from new energy sources and reduce the dimensionality of the historical datasets;

[0010] Step 3: Use density-based clustering to classify the historical dataset after dimensionality reduction in Step 2, constructing M+1 data subsets; M is the total number of types after classification.

[0011] Step 4: Construct and train a classifier; the input data of the classifier is the data after dimensionality reduction in Step 2, and the output is the category obtained in Step 3. Then, train the classifier.

[0012] Step 5: For the M+1 data subsets from Step 3, independently apply the improved optimal bandwidth kernel density estimation to the data of each subset to calculate the probability density function, and then calculate the probability density function for a given significance level. The percentage of prediction error below;

[0013] Step 6: For each time-section data point, classify it using the trained classifier, obtain the corresponding prediction error percentage range based on the classification results, and then obtain the given significance level. The prediction range is set below, and the prediction range is corrected by taking into account the actual power generation boundary conditions.

[0014] The present invention has the following beneficial effects:

[0015] 1. The dynamic evaluation method for uncertainty of new energy output power of the present invention can evaluate the predicted output range of renewable energy.

[0016] 2. This invention distinguishes data with different prediction error characteristics by clustering data and training a classifier for each category, and optimizes the optimal bandwidth parameters for each category of data separately. This solves the problem of large differences in prediction error under different weather conditions (such as temperature, irradiance / wind speed) and different power prediction values, and improves the accuracy of dynamic assessment of output power uncertainty.

[0017] 3. In kernel density estimation, bandwidth is a key parameter that determines the smoothness of the estimation results. Too low a bandwidth may lead to overfitting, while too high a bandwidth may lead to underfitting. The selection of the optimal bandwidth directly affects the accuracy of the results. This invention transforms the traditional optimal bandwidth selection problem into an optimization problem, designs the objective function of the optimization algorithm, and constructs an optimization model based on prediction interval indices (PICP and PINAW). Through multi-objective optimization, the most suitable optimal bandwidth is obtained, thus leading to a suitable optimal bandwidth kernel density estimate.

[0018] 4. This invention reduces the dimensionality of data by using the correlation coefficient method and performs secondary dimensionality reduction on the training data by using principal component analysis. This reduces the input dimension of the data, accelerates the training and inference of the classifier, alleviates the overfitting problem, and makes the classifier model more lightweight. It is suitable for implementing the dynamic evaluation method for uncertainty of new energy output power in scenarios where edge computing resources are limited. Attached Figure Description

[0019] Figure 1 This is a flowchart of the dynamic evaluation method for uncertainty in the output power of new energy sources according to the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] like Figure 1 As shown, this invention provides a dynamic evaluation method for the uncertainty of new energy output power. Taking photovoltaic power generation as an example, the specific method is as follows:

[0022] Phase 1: Data preprocessing and feature selection; including:

[0023] Step 1: Clean historical data of actual power generation, historical data of predicted power, and historical meteorological data of new energy sources, including handling missing values, duplicate values, and erroneous values, and data standardization; historical data refers to the operational data collected by the system, including historical data of actual power generation, historical data of predicted power, and meteorological data of new energy sources at each time point.

[0024] Step 2 involves constructing a feature data matrix using historical datasets of actual power generation from new energy sources and reducing the dimensionality of the historical datasets. This includes:

[0025] Step 2.1: Construct a feature data matrix using a historical dataset of actual power generation from new energy sources.

[0026] Suppose that each sample in the historical dataset contains the following features: actual power generation data, predicted power generation value, meteorological data, and statistical values ​​of meteorological data from the previous N time periods;

[0027] For photovoltaic power generation, meteorological data includes irradiance, temperature, humidity, and may also include wind speed, PM2.5 (fine particulate matter, which refers to particulate matter in the atmosphere with a diameter of less than or equal to 2.5 micrometers), PM10 (inhalable particulate matter, which refers to particulate matter in the atmosphere with a diameter of less than or equal to 10 micrometers), cloud cover, precipitation, etc.

[0028] The statistical values ​​of the meteorological data for the first N time segments include the mean, median, and variance of the meteorological data for the first N time segments.

[0029] A feature data matrix is ​​constructed using the features of the samples. Each row of the feature data matrix belongs to the data of the same sample, and each column belongs to the data of the same feature.

[0030] Step 2.2, dimensionality reduction of the historical dataset of actual power generation of new energy: When selecting features, the Picerson correlation coefficient method can be used to select the features that have the greatest impact on the uncertainty of power generation of this type of renewable energy, and the PCA (Principal Component Analysis) method can be used for dimensionality reduction.

[0031] When the number of features is greater than 4, the correlation coefficient between each feature and the prediction error is first calculated using Picerson correlation analysis. Then, the elbow method is used to observe the elbow points where the slope of the curve changes significantly to determine the appropriate number of features. Finally, principal component analysis is used to reduce the dimensionality of the determined features, and the elbow method is used to select the principal components.

[0032] Phase Two: Data classification and training of the classifier model; including:

[0033] Step 3: Use density-based clustering to classify the dimensionality-reduced historical dataset from Step 2, ensuring that samples with similar characteristics are grouped into the same category.

[0034] Density-based clustering methods can be used to classify historical data. The main steps are as follows:

[0035] Step 3.1, determine the two core parameters of DBSCAN (Density-Based Spatial Clustering of Applications with Noise):

[0036] (1) The maximum radius of the neighborhood ε is based on the feature space distance representation, such as using Euclidean distance or setting it according to experience;

[0037] sample and The Euclidean distance is:

[0038] ;

[0039] in, The number of features and The first , The first sample Each feature value; sample and If the Euclidean distance exceeds the maximum neighborhood radius ε, the sample is considered to be... and They are not of the same type.

[0040] (2) Minimum number of samples in the neighborhood (MinPts).

[0041] Step 3.2: After clustering, construct a data subset for each data class (assuming there are M data classes, construct M data subsets, where M is the total number of types after classification). Place the noise data that does not belong to any class into a separate data subset, hereinafter referred to as the residual class, and construct a total of M+1 data subsets.

[0042] Step 4, construct and train the classifier:

[0043] The input data for the classifier is the feature data after dimensionality reduction by PCA in step 2.2, and the output is the category (or residual class) in step 3. The classifier can be trained using classification algorithms such as Support Vector Machine (SVM) and Random Forest, or using Convolutional Neural Network or other neural network training methods.

[0044] Third stage: Calculate the optimal bandwidth for the M+1 data subsets in step 3 and determine the prediction error range at a given confidence level.

[0045] Step 5: For the M+1 data subsets from Step 3, independently apply the improved optimal bandwidth kernel density estimation to the data of each subset to calculate the probability density function, and then calculate the probability density function for a given significance level. The percentage of prediction error.

[0046] Step 5.1, in the kernel density estimation process, the kernel density estimation formula using the Gaussian kernel function is as follows:

[0047] ;

[0048] in, The estimated probability density function value, The total number of samples, For bandwidth, As the independent variable, For the first One sample ( Take 1 to (integers);

[0049] The Gaussian kernel function has the following specific form: ;

[0050] in, As the independent variable, It is an exponential function;

[0051] Gaussian kernel function Substituting into the above formula for estimating the kernel density of the Gaussian kernel function, we obtain the formula for calculating the kernel density estimate (probability density function) of the Gaussian kernel function as follows:

[0052] .

[0053] Among them, bandwidth Excessive bandwidth can lead to over-smoothing of density estimates (large bias) and loss of data details; Too small a bandwidth will result in high noise (large variance) in the estimation results, failing to reflect the true distribution; therefore, a multi-objective optimization method is adopted with the objectives of minimizing the predicted interval coverage probability (PICP) to the closest to the confidence interval and minimizing the predicted interval average width (PINAW) to obtain the optimal bandwidth:

[0054] Predictive evaluation indicators include:

[0055] (1) Prediction Interval Coverage Probability (PICP) indicator:

[0056] ;

[0057] in, For the number of prediction points, the first... The actual value of each predicted point falls within the prediction interval. =1, otherwise It is 0.

[0058] (2) Mean Prediction Interval Width (MPIW) indicator:

[0059] ;

[0060] in, , Prediction points The upper and lower bounds of the prediction interval are given.

[0061] (3) Prediction Interval Normalized Average Width (PINAW) indicator:

[0062] ;

[0063] In the formula, For prediction points The width of the prediction interval at that point. It is a function with maximum value. It is a minimum value function; ;

[0064] Considering minimizing the predicted interval coverage probability (PICP) closest to the confidence interval and the predicted interval average width (PINAW) as objectives, a multi-objective optimization model is constructed, with the objective function being... for:

[0065] ;

[0066] in, Given a significance level, with Minimum as target, bandwidth For decision variables, bandwidth The initial values ​​can be calculated using the traditional maximum likelihood estimation method or cross-validation method; after calculating the evaluation index of the prediction interval, the objective function is obtained. Value, and for bandwidth Iterative corrections are performed, and the optimal bandwidth can be obtained by using optimization methods such as gradient descent and PSO (particle swarm optimization algorithm). This bandwidth is then used as the bandwidth in the probability density function calculation formula to calculate the estimated probability density function value.

[0067] coefficient and As the equilibrium coefficient, it is adjusted based on different forecasting scenarios (such as renewable energy power forecasting and load forecasting), considering whether to prioritize ensuring confidence level or accuracy. and Through multiple experiments, in the scenario of assessing the uncertainty of renewable energy output power, the coefficient... , The predicted interval coverage probability obtained after time optimization is closest to the actual probability, and the normalized average width of the predicted interval is the smallest.

[0068] Step 5.2: Calculate the percentage of prediction error at a given confidence level using the estimated probability density function value.

[0069] Based on the estimated probability density function value obtained in step 5.1 Find the cumulative distribution function. At a given significance level Next, find the quantiles. and And convert it into a prediction error percentage range. .

[0070] The specific method is as follows: Prediction error e = predicted value -The lower bound of the prediction error percentage is the actual value y. Upper limit of prediction error percentage .

[0071] Phase 4: Dynamically calculate the uncertainty interval:

[0072] Step 6: For each time segment of data, classify it using the trained classifier, and obtain the corresponding prediction error percentage range based on the classification results.

[0073] Step 6.1: Set the input of the classifier to the feature data after dimensionality reduction of the data for each time segment.

[0074] Step 6.2: Using the classifier trained in Step 4, obtain the category of the time-series data and the corresponding optimal bandwidth, and output the prediction error percentage range. .

[0075] Step 6.3: Obtain the percentage range of prediction error based on the category of the time-section data, and then overlay it with the actual value of the predicted power to obtain a given significance level. The predicted range below.

[0076] Step 6.4, adjust the prediction interval by considering the actual power generation boundary conditions: for example, for photovoltaic power generation, if the calculated lower limit of the prediction interval is less than 0, then the lower limit of the prediction interval is 0; if the calculated upper limit of the prediction interval is greater than the maximum power generation of the photovoltaic installation, then the upper limit of the prediction interval is the maximum power generation of the photovoltaic installation.

[0077] In the fifth stage, the model is continuously updated and parameters are adjusted:

[0078] Periodically perform second to fourth stage calculations (steps 3 to 6) on new rolling historical data to update the classifier and the optimal bandwidth for each category, so as to make the results of dynamic assessment of uncertainty of new energy output power (predicted uncertainty range) more accurate.

[0079] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The embodiments of the present invention can be implemented using various computer languages.

[0080] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0081] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0082] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0083] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0084] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0085] The above description is merely an embodiment of the present invention and does not limit the scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related system fields, are similarly included within the protection scope of the present invention.

[0086] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A method for dynamic evaluation of the uncertainty of new energy output power, characterized in that, include: Step 1: Clean historical data of actual power generation, historical data of predicted power, and historical meteorological data of new energy sources, including handling missing values, duplicate values, erroneous values, and data standardization; Step 2: Construct a feature data matrix using historical datasets of actual power generation from new energy sources and reduce the dimensionality of the historical datasets; Step 3: Use density-based clustering to classify the historical dataset after dimensionality reduction in Step 2, constructing M+1 data subsets; M is the total number of types after classification. Step 4: Construct and train a classifier; the input data of the classifier is the data after dimensionality reduction in Step 2, and the output is the category obtained in Step 3. Then, train the classifier. Step 5: For the M+1 data subsets in Step 3, independently apply the improved optimal bandwidth kernel density estimation to the data of each data subset to calculate the probability density function, and then calculate the probability density function for a given significance level. The percentage of prediction error below; include: Step 5.1, the probability density function is calculated as follows: ； in, The estimated probability density function value, The total number of samples, For bandwidth, As the independent variable, For the first One sample, Take 1 to Integers; The optimal bandwidth is obtained by using a multi-objective optimization method with the objectives of the prediction interval coverage probability being closest to the confidence interval and the average width of the prediction interval being minimized. This bandwidth is then used as the bandwidth in the probability density function calculation formula to calculate the estimated probability density function value. Predictive evaluation indicators include: (1) Predicted interval coverage probability index : ； in, For the number of prediction points, the first... The actual value of each predicted point falls within the prediction interval. =1, otherwise =0; (2) Average width index of the prediction interval : ； in, , These are the prediction points. The upper and lower bounds of the prediction interval at the given location; (3) Normalized average width index of prediction interval : ； In the formula, For prediction points The width of the prediction interval at that point. It is a function with maximum value. It is a minimum value function; ; Considering the objectives of maximizing the predicted interval coverage probability to be closest to the confidence interval and minimizing the average width of the predicted interval, a multi-objective optimization model is constructed, with the objective function being... for: ； in, Given a significance level, with Minimum as target, bandwidth For decision variables, bandwidth The initial values ​​are calculated using the traditional maximum likelihood estimation method or cross-validation method; after calculating the evaluation index of the prediction interval, the objective function is obtained. The value, and the bandwidth The optimal bandwidth is obtained by iterative correction. coefficient and As the equilibrium coefficient, it is adjusted based on whether to prioritize ensuring confidence level or accuracy, depending on the different prediction scenarios across different intervals. and This ensures that the predicted interval coverage probability obtained after optimization is closest to the actual probability, while minimizing the normalized average width of the predicted interval. Step 5.2: Calculate the given confidence level using the estimated probability density function value. The percentage of prediction error below; Step 6: For each time-section data point, classify it using the trained classifier, obtain the corresponding prediction error percentage range based on the classification results, and then obtain the given significance level. The prediction range is set below, and the prediction range is corrected by taking into account the actual power generation boundary conditions.

2. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, Step 2 includes: Step 2.1: Construct a feature data matrix using a historical dataset: Assume each sample in the historical dataset contains the following features: actual power generation data, predicted power generation value, meteorological data, and statistical values ​​of meteorological data from the previous N time periods; Construct a feature data matrix using the features of the samples, where each row of the feature data matrix belongs to the data of the same sample, and each column belongs to the data of the same feature. Step 2.2, Dimensionality reduction of the historical dataset: Principal component analysis is used for dimensionality reduction.

3. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, Step 3 includes: Step 3.1, determine the two core parameters of DBSCAN: (1) Maximum neighborhood radius ε; The maximum neighborhood radius ε is represented by Euclidean distance or set based on experience; No. , Sample and The Euclidean distance is: ； in, For the number of features, and The first , The first sample One eigenvalue; (2) Minimum number of samples within the neighborhood; Step 3.2: After clustering, construct a data subset for each data class, constructing M data subsets. For noise data that does not belong to any class, put it into a separate data subset as the residual class, constructing a total of M+1 data subsets.

4. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, In step 4, the classifier is trained using a convolutional neural network or other neural network training methods.

5. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, The optimal bandwidth is obtained by using gradient descent and particle swarm optimization algorithms.

6. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, Step 5.2 includes: the estimated probability density function value obtained in step 5.

1. Find the cumulative distribution function. At a given significance level Next, find the quantiles. and And convert it into a prediction error percentage range. ; Prediction error e = predicted value -True value y, lower bound of prediction error percentage Upper limit of prediction error percentage .

7. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, Step 6 includes: Step 6.1: Set the input of the classifier to the feature data after dimensionality reduction of the data for each time segment; Step 6.2: Using the classifier trained in Step 4, obtain the category of the time-series data and the corresponding optimal bandwidth, and output the prediction error percentage range. ; Step 6.3: Obtain the percentage range of prediction error based on the category of the time-section data, and then overlay it with the actual value of the predicted power to obtain a given significance level. The following prediction range; Step 6.4: Correct the prediction interval by considering the actual power generation boundary conditions; for photovoltaic power generation, if the calculated lower limit of the prediction interval is less than 0, then the lower limit of the prediction interval is 0; if the calculated upper limit of the prediction interval is greater than the maximum power generation of the photovoltaic installation, then the upper limit of the prediction interval is the maximum power generation of the photovoltaic installation.

8. The dynamic evaluation method for uncertainty of new energy output power according to claim 1, characterized in that, Also includes: Periodically perform steps 3 through 6 on new rolling historical data to update the classifier and the optimal bandwidth for each category.

Images