Converter station power supply side wind power and photovoltaic novel power supply output prediction model and method

Abstract

The invention provides a converter station power supply side wind power and photovoltaic novel power supply output prediction model and method, and belongs to the field of electric power prediction. Data are processed and analyzed through various technologies such as an isolation forest algorithm, autocorrelation, partial autocorrelation analysis, Pearson correlation coefficients and maximum information coefficients. And then, constructing a power generation output prediction model of wind, light and water renewable resources by adopting a bidirectional long-short-term memory network based on an Attention mechanism, and realizing accurate prediction of future new energy power generation output. According to the method, various data processing and prediction technologies are integrated, various factors influencing new energy power generation output are comprehensively analyzed, the prediction accuracy and reliability are improved, and the method plays an important role in promoting optimal scheduling and operation safety of a power grid.

Description

Technical Field

[0001] The present invention belongs to the field of power prediction, and more specifically relates to a prediction model and method for the output of new energy sources such as wind power and photovoltaic power on the power supply side of a converter station. Background Art

[0002] As a key link for new energy to access the power grid, the power supply end of the converter station is mainly composed of new renewable energy sources such as wind power and photovoltaic power. The output of these new energy sources varies greatly due to environmental factors such as weather and season, bringing problems to the operation and dispatching of the power grid. In particular, the output of new energy sources such as wind power and photovoltaic power is greatly affected by climate change. Factors such as wind speed and sunlight will affect the power generation of new energy. Therefore, accurately and quickly predicting the output of new energy generation on the power supply side of the converter station is of great significance for improving the dispatching level and operation safety of the power grid.

[0003] To solve the above problems, researchers have developed various prediction models for wind power and photovoltaic power output prediction, including prediction models based on mathematical statistics, prediction models based on machine learning, etc. However, most of the above prediction models only consider some of the influencing factors in wind power and photovoltaic power, and the prediction effect on the overall power generation performance is not ideal.

[0004] In addition, the existing prediction models usually predict based on historical power generation data, fail to fully utilize environmental factor data such as meteorology, and also fail to consider the seasonal variation characteristics of wind power and photovoltaic power, which also limits the improvement of prediction performance.

[0005] Therefore, developing a prediction model that can comprehensively consider various influencing factors and effectively predict the output of new energy on the power supply side of the converter station has important practical application value for improving the dispatching level and operation safety of the power grid. Summary of the Invention

[0006] The present invention is committed to solving the problems in the prior art such as the low prediction accuracy of the output of new energy sources such as wind power and photovoltaic power, incomplete consideration of factors, and failure to fully explore and utilize environmental factor data such as meteorology and historical power generation data. The objective of the present invention is to develop a prediction model and method for the output of new energy sources such as wind power and photovoltaic power that comprehensively considers various influencing factors based on multiple methods such as isolated forest, autocorrelation, partial autocorrelation analysis, Pearson correlation coefficient, and maximum information coefficient. It can comprehensively analyze the correlation between various influencing factors and the power generation power, in order to improve the prediction accuracy of the output of new energy on the power supply side of the converter station, and has a positive promoting effect on improving the dispatching level and operation safety of the power grid.

[0007] To achieve the above objective, the present invention is implemented by adopting the following technical solutions: The prediction model and method include:

[0008] Obtain the historical monitoring data and real-time operation data of power sources such as wind power, photovoltaic power, and hydropower, and use the Isolation Forest algorithm to clean the monitoring data, detect and remove abnormal data;

[0009] Autocorrelation and partial autocorrelation analysis: Conduct autocorrelation analysis and partial autocorrelation analysis on wind and light power data to judge the stationarity of time series data, and simultaneously obtain the long-term trend, seasonality, and residuals of the data;

[0010] Correlation analysis between power generation and meteorological factors: Analyze renewable new energy sources such as wind power, photovoltaic power, and hydropower based on Pearson correlation coefficient and maximum information coefficient correlation analysis methods to comprehensively analyze the correlation between various influencing factors and power generation;

[0011] Construct a power generation output prediction model for wind, light, and water renewable resources.

[0012] In one solution, the data cleaning includes:

[0013] Randomly select features from the monitoring dataset to construct a feature space;

[0014] Randomly select a splitting value within the value range of the selected features to construct an isolation tree, perform binary partitioning on the sample set, and recursively execute until each data is completely isolated or the set tree height is reached;

[0015] Construct an isolation forest composed of multiple isolation trees, calculate the path length of each sample point in the forest, and then calculate its anomaly score to determine abnormal data.

[0016] In one solution, the autocorrelation analysis discriminates the stationarity and time series dependence of the sequence by calculating the degree of correlation between the values at each time series in the sequence and its historical values; when the sequence has strong autocorrelation, the future values of the sequence can be effectively predicted through its historical values.

[0017] In one solution, the partial autocorrelation analysis determines the true correlation between two non-directly adjacent time series after removing the influence of these intermediate orders by identifying whether the correlation between any two time series is obtained only through other intermediate time series.

[0018] In one solution, the correlation analysis between power generation and meteorological factors is analyzed using Pearson correlation coefficient and maximum information coefficient to comprehensively consider the influence of various influencing factors on power generation.

[0019] In one solution, the Pearson correlation coefficient is used to measure the strength and direction of the linear correlation between variables, and the maximum information coefficient is based on mutual information and the complex relationship between variables and is used to evaluate the association strength between variables.

[0020] In one solution, to build a prediction model for the power generation output of wind, light, and water renewable resources, first preprocess and perform wavelet decomposition on the time series data, then use a bidirectional long short-term memory network based on the Attention mechanism to predict and reconstruct each decomposed subsequence respectively, and finally obtain the prediction result by using a method to verify the prediction accuracy.

[0021] In one solution, the method for building a prediction model for the power generation output of wind, light, and water renewable resources uses a wavelet decomposition - bidirectional long short-term memory network based on the Attention mechanism as the prediction model to predict the power generation of new energy such as wind power, photovoltaic power, and hydropower. By introducing wavelet decomposition, it comprehensively considers the volatility, non-linearity, and time correlation of the data, and splits the LSTM into two directions to fully learn the forward and backward characteristics of the time series data. The Attention mechanism is adopted to assign different weights to the hidden layer units of the network to strengthen the influence of the important information of the sample data, highlighting the different attention distributions of wind and light power at different time point positions due to the differences in data characteristics, and selectively obtaining more effective information, so as to capture the time correlation between the data and select the corresponding driving data for prediction.

[0022] Advantages of the present invention:

[0023] Based on multiple methods such as the isolated forest, autocorrelation, partial autocorrelation analysis, Pearson correlation coefficient, and maximum information coefficient, the present invention comprehensively analyzes the correlation between various influencing factors and the power generation, making the prediction model more adaptable to the complex characteristics of new energy such as wind power and photovoltaic power, and thus improving the prediction accuracy.

[0024] The present invention uses a wavelet decomposition - bidirectional long short-term memory network based on the Attention mechanism to predict the power generation of new energy such as wind power, photovoltaic power, and hydropower, considering the volatility, non-linearity, and time correlation of the data, and can more accurately predict the future power generation output of new energy.

[0025] Accurate prediction of the power generation output of new energy is extremely important for the optimal dispatching and safe and stable operation of the power grid. The present invention can effectively improve the prediction accuracy of the power generation output of new energy such as wind power and photovoltaic power, and contribute to the optimal operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of the method of the present invention;

[0027] Figure 2 It is a schematic diagram of the wavelet transform process;

[0028] Figure 3 It is a schematic diagram of the Attention mechanism;

[0029] Figure 4 Distributed renewable energy power generation prediction framework diagram. Detailed implementation manners

[0030] To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0031] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as understood by those skilled in the technical field to which the present invention belongs. The terms used in the description of the present invention in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0032] As Figure 1 shown, the specific steps of a new type of power output prediction model and method for wind power and photovoltaic power sources on the power supply side of a converter station are as follows:

[0033] Step 1: Clean the wind, light, and water monitoring data. Similar to the random forest, the isolation forest is a collection of isolation trees (IsolationTree, iTree), and its construction is a completely random process.

[0034] Given a data set X = {x1, x2,..., x n}, the number of its data is n, and the feature dimension is d. Randomly select features from the data set X to form a feature space. Randomly select a value within the value range of the selected features as a random split value p, construct an isolation tree, and perform binary partitioning on the sample set, that is, the points less than the p value enter the left branch, and the points greater than or equal to the p value enter the right branch. Continuously repeat the above recursive partitioning operation until each data is completely separated from other data or reaches the maximum height of the set tree, then stop the partitioning. The relationship between the tree height limit l and the number of subsamples ψ is:

[0035] h = ceiling(log2ψ)

[0036] where ceiling is the CEILING function, and the tree height l is the average tree height. Construct multiple isolation trees in the same way, then the isolation forest is a collection of multiple isolation trees.

[0037] Anomaly score calculation: Compared with normal data, anomaly points often have obvious differences in one or more dimensions, and at the same time, the amount of their data is much smaller than that of normal data. Therefore, abnormal data is closer to the root node of the number. Define the path length h(x), and it is considered that the smaller the path length of the sample, the higher the degree of "isolation". The calculation formula is as follows:

[0038] h(x) = e∣C(ψ)

[0039] where ψ is the number of training samples extracted from the dataset to construct a single isolation tree, and e is the number of edges passed by the sample x i from the root node of the tree to the leaf node. C(ψ) is the average path length of this tree, and its calculation formula is as follows:

[0040]

[0041] where H(i) is the harmonic function, approximately equal to ln(i) + 0.5772156. Calculate the anomaly score s(x, ψ) of each point to evaluate the degree of "isolation" of the sample point. The formula is as shown:

[0042]

[0043] where E(h(x)) is the data point x i The average value of the path lengths in all isolation trees of the forest. For the data in the dataset X, if its anomaly score s is close to 1, it is considered that this data point is abnormal data; if its anomaly score is much less than 0.5, it can be considered that this data point is a normal instance; if the anomaly scores of all data points are around 0.5, it is considered that there is no obvious anomaly in the entire sample dataset.

[0044] Missing data filling

[0045] Aiming at the problems of data distortion and missing in the monitoring, statistics, and transmission of renewable energy data such as wind power, photovoltaic, and hydropower, a multiple imputation method based on the chained method (Multiple Imputation by Chained Equations, MICE) is used to fill in the missing data.

[0046] Suppose Y = (Y1, Y2,..., Yk) is a k-dimensional random variable, following the k-variable distribution P(Y|θ). The sample y = (y1, y2,..., yk) of Y is an independent and uniformly distributed data missing matrix, where y = (yi1, yi2,..., yik) is the i-th sample. Let yobs = (yobs1, yobs2,..., yobsk) and ymis = (ymis1, ymis2,..., ymisk) represent all test sets and missing sets respectively. The standard process of creating multiple imputations y* for ymis is as follows:

[0047] Step1: Calculate the posterior distribution P(θ|y) of the missing variable θ according to the test set y obs obs )

[0048] Step2: Generate θ according to the distribution P(θ|y obs ) *

[0049] Step3: Given θ * , calculate the conditional posterior distribution P(y mis |y mis , θ = θ obs * )

[0050] Step4: Repeat steps Step2 and Step3 m times to generate m imputation results.

[0051] The multiple imputation method in a chained manner performs cyclic sampling through conditional distributions in the following form to obtain the posterior distribution of θ:

[0052]

[0053] where Y -i =(Y1,......,Y i-1 ,Y i+1 ,......,Y k ) represents the new variable after deleting h columns in Y. The parameters θ1, ……, θ k are generated according to the corresponding conditional density distributions. The following consecutive Gibbs samplings are included in the t (t < m) cycle periods of the whole process:

[0054]

[0055] Step 2, Autocorrelation and Partial Autocorrelation Analysis

[0056] Since the prediction of wind power and photovoltaic power is based on the power data information of historical wind, light, etc., it is necessary to perform autocorrelation analysis and partial autocorrelation analysis on the wind and light power data to judge the stationarity of the time series data, and at the same time obtain the long-term trend (Trend), seasonality (Seasonality), and residuals (Residuals) of the data.

[0057] The Autocorrelation Function (ACF) reflects the correlation between the values of the same sequence at different time series, that is, the correlation degree between the current value of the sequence and its historical values. For the time series X = {x t ​​, where \(t = 1, 2,\cdots, n\), and its autocovariance function of lag \(k\) is:

[0058]

[0059] where is the average value of the elements in the time series \(\{x t \}\). Let \(k = 0\), we get:

[0060]

[0061] Therefore, the autocorrelation coefficient of the time series \(\{x t \}\) is:

[0062]

[0063] The partial autocorrelation function (PACF) is a correlation measure that describes the influence of the time series \(\{x t-k \}\) on \(\{x t \}\), while excluding the interference of the middle \(k - 1\) random variables, that is, the interference of \(\{x t-1 , x t-2 ,\cdots, x t-k-1 \}\). Its expression is:

[0064]

[0065] Step 3: Correlation analysis between power generation and meteorological factors

[0066] Since there is a large correlation between renewable new energy power generation such as wind power, photovoltaic power, and hydropower and meteorological factors, it is necessary to conduct a correlation analysis between power generation and meteorological factors. Based on the Pearson correlation coefficient and the maximum information coefficient correlation analysis method, the renewable new energy such as wind power, photovoltaic power, and hydropower is analyzed to comprehensively analyze the correlation between each influencing factor and power generation.

[0067] The calculation formula for the Pearson correlation coefficient of the variable sequences \(X=\{x i , i = 1, 2,\cdots, n\}\) and \(Y = \{y i , i = 1, 2,\cdots, n\}\) is:

[0068]

[0069] where \(cov(X, Y)\) is the covariance between variables \(X\) and \(Y\), \(\sigma X \) and \(\sigma Y \) are the standard deviations between variables \(X\) and \(Y\). The Pearson coefficient \(r\) of variables \(X\) and \(Y\) XYIf it is greater than 0, the two variables are positively correlated; if the Pearson coefficient r of variables X and Y is obtained XY is less than 0, the two variables are negatively correlated. |r XY | The larger it is, the higher the degree of correlation between variables X and Y.

[0070] The calculation of the maximum information coefficient is based on mutual information and the grid partition method. For variable sequences X = {x i , i = 1, 2, ……, n} and Y = {y i , i = 1, 2,......, n}, the formula for calculating their mutual information MI(X,Y) is:

[0071]

[0072] where p(x) and p(y) are the marginal probability densities of variables X and Y, and p(x,y) is the density of the joint probability of variables X and Y.

[0073] Let the variable X and Y form a finite set of ordered pairs D = {(x i , y i ), i = 1, 2,......, n}, define a grid G of size a×b, divide the number of samples of variables X and Y into a and b parts, calculate the mutual information MI(X,Y) in each cell of the grid G, and take the maximum value of the mutual information calculated in different grid partition methods as the mutual information value of the partitioned grid G. The maximum mutual information of D under the grid G partition is:

[0074] MI*(D,a,b) = maxMI(D∣G)

[0075] where D∣G represents using the grid G to partition D, and form the feature matrix M(D) with the maximum normalized mutual information values under different partitions a,b

[0076]

[0077] Since different grids G will result in different D∣G, therefore, by performing an exhaustive search on the feature matrix M(D) a,b to obtain the optimal a0×b0 grid G0, the MIC of variables X and Y is defined as:

[0078]

[0079] where B(n) is the maximum grid area for the exhaustive search, and it is known from the literature that usually B(n) = n 0.6 gives the best effect.

[0080] The value range of MIC is [0,1], and the larger the value, the stronger the association strength between variables X and Y.

[0081] Cluster analysis: The variation law of renewable energy power generation is somewhat related to the corresponding weather types. Under different weather conditions, there are also certain differences in its prediction accuracy. Therefore, by classifying different weather types based on wind and light power curves, it is possible to effectively determine the similar day type to which the period to be predicted belongs, retaining the volatility and diversity of historical data within a certain period of time, thereby improving the prediction accuracy. Based on the relatively common K-means clustering algorithm, data clustering analysis is carried out to achieve the classification of similar days under different weather types.

[0082] The K-means clustering algorithm is more suitable for large-scale data sets compared to other clustering algorithms because of its high computational efficiency. Its algorithm process is as follows:

[0083] Step1: For a given number of clusters k, the K-means algorithm randomly selects k objects in the original data set for initialization, which are the center points of the clusters;

[0084] Step2: Calculate the distances from the remaining samples to each cluster and assign them to the cluster with the closest distance;

[0085] Step3: After the assignment is completed, calculate the updated cluster centers of each cluster, that is, the mean values of all samples within the cluster;

[0086] Step4: Repeat Step2 and Step3 until the center points of the clusters no longer change and the clustering criterion function

[0087] converges to obtain the final clustering result.

[0088] The K-means clustering algorithm usually uses the squared error criterion as the criterion function, and its calculation formula is as follows:

[0089]

[0090] where k is the number of clusters, m is the total number of samples in each cluster, p ij is the j-th sample point in the i-th cluster, and C i is the centroid of the i-th cluster.

[0091] Step 4: Construct a prediction model for the power generation output of wind, light, and water renewable resources.

[0092] Considering that existing prediction methods only focus on one-way data information flow, ignore the influence of reverse data sequence transformation rules on short-term prediction, and insufficiently consider the time correlation and periodicity of data. When the input time series is long, sequence information is easily lost and the prediction accuracy of the model is not high. Therefore, a wavelet decomposition-bi-directional long short-term memory network (WBiLSTM) based on the Attention mechanism is used as a prediction model to predict the power generation of new energy sources such as wind power, photovoltaic power, and hydropower. By introducing wavelet decomposition, the volatility, nonlinear characteristics, and time correlation of data are comprehensively considered, and the LSTM is split into two directions to fully learn the forward and reverse characteristics of time series data. The Attention mechanism is adopted to assign different weights to the hidden layer units of the network to strengthen the influence of important information in sample data, highlight the different attention distributions of wind and light power at different time point positions due to data characteristic differences, selectively obtain more effective information, thereby capturing the time correlation between data and selecting corresponding driving data for prediction.

[0093] The renewable energy power prediction model uses historical power data and factors affecting the change of power values as input parameters, and the power value to be predicted as the output. Then, the power prediction problem can be transformed into a mathematical problem of finding the mapping relationship between input variable X and output variable Y. The multi-hidden layer structure of the deep learning network can better extract effective features from a large amount of input data information and has its unique advantages in dealing with the time series problem of power data itself and the nonlinear relationship between various influencing factors and it. Therefore, for the power prediction requirements of distributed new energy sources such as wind power, photovoltaic power, and hydropower, the long short-term and extremely neural network (LSTM) is selected as the basic prediction model to conduct research.

[0094] The specific steps are as follows:

[0095] Step1: Preprocess the obtained input data. To eliminate the influence of dimensions, eliminate dimension errors, and accelerate the training process, normalize the input training data and convert it into scalar values, and output the power values.

[0096] Step2: Initialize the deep learning network, input the data set into the network respectively, and use the training samples to train the network.

[0097] Step3: Use the trained neural network model to predict the period to be predicted, verify the prediction accuracy, and obtain the finally predicted power data.

[0098] Wavelet decomposition

[0099] Regarding the strong randomness and volatility of new energy output, and at the same time having a certain periodicity and relatively high time correlation. Wavelet decomposition is used to extract feature data, and the principle of wavelet decomposition is as follows:

[0100] Let any functions \(g(t)\) and \(\psi(t)\) be square-integrable functions, that is, \(g(t)\in L\) 2 (\mathbb{R})\), \(\psi(t)\in L\) 2 (\mathbb{R})\), and \(\psi(t)\) satisfies the following conditions:

[0101]

[0102] Then the calculation formula for the continuous wavelet transform is:

[0103]

[0104] where \(\psi\) p,q (t) is the continuous wavelet generated by transforming the mother wavelet through the scaling variable \(p\) and the translation variable \(q\), and \(t\) is the time index.

[0105] The reconstruction formula for the wavelet transform is:

[0106]

[0107] The discrete wavelet transform (Discrete Wavelet Transform, DWT) is often used to process discrete signals. The discretization formulas for \(p\) and \(q\) are taken as and Then the calculation formula for the discrete wavelet is:

[0108]

[0109] where \(p_0 = 2\), \(q_0 = 1\); the calculation formula for DWT is:

[0110]

[0111] where the input signal passes through the low-pass filter \(L(e)\) and the high-pass filter \(H(e)\), and further uses the Mallat algorithm, and its calculation formula is as follows:

[0112] a j = a j+1 h_1; d j = d j+1 l_1\ j = 0,1,\cdots,n - 1

[0113] where \(h_1\) and \(l_1\) are the low-pass filtering coefficient and the high-pass filtering coefficient respectively, \(a\) j and \(d\) j are the low-frequency signal sequence and the high-frequency signal sequence after decomposition respectively, and \(n\) is the number of wavelet decomposition layers.

[0114] Its fusion reconstruction formula is:

[0115] a j = a j+1 h_2 + dj+1 l2 j = n - 1, L, 1, 0

[0116] where h2 and l2 are the dual operators of h1 and l1 respectively, and the wavelet transform process is as Figure 2 shown:

[0117] The Attention mechanism simulates the situation where the human brain's attention focuses on a specific area at a specific moment, thereby selectively obtaining more effective information and ignoring useless information. By assigning different probability weights to the units in the hidden layer of the neural network, the influence of key information is highlighted, and the accuracy of the model's judgment is enhanced. The Attention mechanism is used to effectively solve the situation where it is difficult for the model to learn a reasonable vector representation due to the long input time series; at the same time, aiming at the power characteristics of wind power and photovoltaic power, the differences in the attention distributions at different time points caused by the differences in data characteristics between the two are highlighted, the temporal correlation between data is captured, and the corresponding driving data is selected for prediction. The structural schematic of the Attention mechanism is Figure 3 as follows, where a t is the attention probability distribution value of the Attention mechanism for the output of the hidden layer of the neural network.

[0118] The calculation formula for the weight coefficient of the Attention mechanism layer can be expressed as:

[0119] e t = utanh(wh t + b)

[0120]

[0121] where e t represents the attention probability distribution value determined by the output vector h t of the hidden layer of the neural network at time t, u and w are weight coefficients, b is the bias coefficient, and s t is the output of the Attention mechanism at time t.

[0122] (4) The power prediction model for distributed renewable energy, such as Figure 4 :

[0123] The specific process of the prediction model for distributed renewable energy is as follows:

[0124] Step1: Preprocess the obtained input data. To eliminate the influence of dimensions, eliminate the dimensional error, and accelerate the training process, the input training data is normalized and converted into scalar values, and the output power values are obtained.

[0125] Step 2: Use the wavelet decomposition method to perform wavelet decomposition and single-branch reconstruction on the input time series data, achieve the optimized extraction of trend information, filter out the fluctuation information, separate the noise, and obtain the final input sequence of the model.

[0126] Step 3: Initialize the BiLSTM network based on the Attention mechanism, input the decomposed dataset training data into the network respectively, and use the training samples to train the network. Use the mean square error as the loss function, and adopt the Adam optimization algorithm for network training to update the weights and obtain the prediction model.

[0127] Use the trained neural network model to predict each sub-sequence decomposed during the period to be predicted respectively, add the predicted sub-sequences, verify the prediction accuracy, and obtain the finally predicted wind power and photovoltaic power data.

[0128] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The described program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above various methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), etc.

[0129] It should be understood that the detailed description of the technical solutions of the present invention with the aid of the preferred embodiments is illustrative rather than restrictive. Those of ordinary skill in the art can modify the technical solutions recorded in each embodiment on the basis of reading the specification of the present invention, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A novel power output prediction model and method for wind power and photovoltaic new power sources on the power supply side of a converter station, characterized in that: The described prediction model and method include: Obtain historical monitoring data and real-time operation data of wind power, photovoltaic power, and hydropower power sources, and use the Isolation Forest algorithm to clean the monitoring data, detect and remove abnormal data; Autocorrelation and partial autocorrelation analysis, perform autocorrelation analysis and partial autocorrelation analysis on wind and light power data to judge the stationarity of time series data, and at the same time obtain the long-term trend, seasonality, and residuals of the data; Correlation analysis between power generation and meteorological factors, analyze wind power, photovoltaic power, and hydropower renewable new energy based on Pearson correlation coefficient and maximum information coefficient correlation analysis methods to comprehensively analyze the correlation between various influencing factors and power generation; Construct a prediction model for power generation output of wind, light, and water renewable resources.

2. A new power output prediction model and method for wind power and photovoltaic new power sources on the power supply side of a converter station according to claim 1, characterized in that: The described data cleaning includes: Randomly select features from the monitoring data set to construct a feature space; Randomly select a splitting value within the value range of the selected features, construct an isolation tree, perform binary partitioning on the sample set, and recursively execute until each data is completely isolated or reaches the set tree height; Construct an isolation forest composed of multiple isolation trees, calculate the path length of each sample point in the forest, and then calculate its anomaly score to determine abnormal data.

3. A novel power output prediction model and method for wind power and photovoltaic new power sources on the power supply side of a converter station according to claim 1, characterized in that: The described autocorrelation analysis discriminates the stationarity and time series dependence of the sequence by calculating the degree of correlation between the values at each time series in the sequence and its historical values; When the sequence has autocorrelation, the future value of the sequence can be effectively predicted through its historical values.

4. A new power output prediction model and method for wind power and photovoltaic power sources on the power supply side of a converter station according to claim 1, characterized in that: The described partial autocorrelation analysis determines the true correlation between two non-directly adjacent time series after removing the influence of these intermediate orders by identifying whether the correlation between any two time series is only obtained through other intermediate time series.

5. A prediction model and method for the output of a new type of power source of wind power and photovoltaic power on the power supply side of a converter station according to claim 1, characterized in that: The described correlation analysis between power generation and meteorological factors uses Pearson correlation coefficient and maximum information coefficient for analysis to comprehensively consider the influence of various influencing factors on power generation.

6. The novel power output prediction model and method for wind power and photovoltaic new power sources on the power supply side of a converter station according to claim 5, wherein: Pearson correlation coefficient is used to measure the strength and direction of linear correlation between variables, and the maximum information coefficient is based on mutual information and complex relationships between variables and is used to evaluate the association strength between variables.

7. A new power output prediction model and method for wind power and photovoltaic new power sources on the power supply side of a converter station according to claim 1, characterized in that: For the described construction of a prediction model for power generation output of wind, light, and water renewable resources, first preprocess and wavelet decompose the time series data, then use a bidirectional long short-term memory network based on the Attention mechanism to predict and reconstruct each decomposed subsequence respectively, and finally obtain the prediction result using a method to verify the prediction accuracy.

8. The novel power output prediction model and method for wind power and photovoltaic new power sources on the power supply side of a converter station according to claim 1, characterized in that: The described construction of the prediction model for the power output of wind, light, and water renewable resources uses a wavelet decomposition-bi-directional long short-term memory network based on the Attention mechanism as the prediction model to predict the power generation of new energy sources such as wind power, photovoltaic power, and hydropower. By introducing wavelet decomposition, the volatility, non-linear characteristics, and time correlation of the data are comprehensively considered, and the LSTM is split into two directions to fully learn the forward and backward characteristics of the time series data. The Attention mechanism is adopted to assign different weights to the hidden layer units of the network to strengthen the influence of the important information of the sample data, highlighting the different attention distributions of wind and light power at different time point positions due to the differences in data characteristics, selectively obtaining more effective information, thereby capturing the time correlation between the data and selecting the corresponding driving data for prediction.

Images