Island new energy power interval prediction method and device

Abstract

The application provides an island new energy power interval prediction method, comprising the following steps: obtaining island new energy data for processing, and taking the processed data as training data; inputting the data into a LASSO quantile regression model, performing factor compression through LASSO regression, and reducing characteristic quantities; performing cross-validation through generalized cross-validation, optimizing constraint coefficients, and obtaining optimal constraint coefficients; determining relevant parameters to obtain an island new energy power interval prediction model; obtaining island new energy data to be predicted; inputting the data into the island new energy power interval prediction model to construct an island new energy power prediction interval. The prediction method can take into account various characteristics and multiple environmental factors for interval prediction, provides more information for island power grid optimization scheduling, makes the prediction result accurate and reliable, and can obtain information from various external influencing factors, thereby effectively and accurately predicting the power of island new energy.

Description

Technical Field

[0001] This invention relates to the field of machine learning, and in particular to a method and apparatus for predicting the power range of new energy sources on islands. Background Technology

[0002] In recent years, environmental pollution has become increasingly serious, and the shortage of non-renewable energy sources has become increasingly prominent. With growing global attention to climate change and global warming, reducing carbon emissions has become a consensus. Countries around the world have begun exploring new energy power generation methods, and large-scale renewable energy grids, represented by photovoltaic and wind power, are an unstoppable trend and prominent feature of the next generation of power systems.

[0003] However, the randomness and volatility brought about by the large-scale grid connection of renewable energy also pose new challenges to the safe and stable operation of the power grid. New energy power generation is affected by various complex environmental factors, resulting in strong random fluctuations, intermittency, and instability in its power output. With a high proportion of new energy power generation in the power system, it, as an uncontrollable power source, will seriously threaten the safe and stable operation of the power system. Therefore, researching probabilistic prediction technology for new energy power segments is of great value for constructing a comprehensive power system security defense system and achieving risk control.

[0004] Power forecasting is fundamental to the safe, economical, and stable operation of the power grid. The maritime climate of islands is characterized by frequent cloud cover, rapid cloud movement, and susceptibility to typhoons, leading to more severe and transient fluctuations in the output of renewable energy sources such as wind, solar, and wave power, making accurate and reliable forecasting difficult. Currently widely used traditional power system forecasting relies primarily on deterministic forecasting, i.e., point-based forecasting, which struggles to quantitatively analyze and estimate the range of forecast error fluctuations, posing significant challenges for its application in renewable energy forecasting on islands.

[0005] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to solve the problem of accurate prediction of renewable energy power on islands, and to provide a method and apparatus for predicting renewable energy power ranges on islands.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a method for predicting the power range of new energy sources on islands, comprising the following steps:

[0009] S1: Acquire and process island new energy data, and use the processed data as training data. The island new energy data includes: wind power data, photovoltaic data and wave energy data.

[0010] S2: Input the data into the LASSO quantile regression model, and perform factor compression through LASSO regression to reduce the number of features;

[0011] S3: Perform cross-validation through generalized cross-validation to fine-tune the constraint coefficients and obtain the optimal constraint coefficients;

[0012] S4: Select variables based on the optimal constraint coefficients, determine relevant parameters, and obtain the island new energy power range prediction model;

[0013] S5: Obtain the island's new energy data to be predicted;

[0014] S6: Input the data into the island new energy power range prediction model to construct the island new energy power prediction range.

[0015] In some embodiments, between steps S4 and S5, there is an additional step: the test data is predicted by inputting the test data into the island new energy power range prediction model, and the prediction result is obtained. If the error requirement is not met, the constraint coefficient is reset and the prediction is performed again.

[0016] In some embodiments, step S6 includes: constructing a power prediction range for island renewable energy under a 90% confidence interval.

[0017] In some embodiments, the LASSO quantile regression model in step S2 is expressed by the following formula:

[0018]

[0019] st‖α(τ)‖1+‖β(τ)‖1≤S

[0020] Where S represents the constraint coefficient, ||α(τ)||1 and ||β(τ)||1 represent the L1 norm, α(τ) represents the regression coefficient of historical power at the τ quantile, β(τ) represents the regression coefficient of meteorological factors at the τ quantile, and y i Indicates power, x i Indicates meteorological factors, Let q represent the asymmetric loss function, n represent the total number of samples, p = 1, 2, ..., c, q = 1, 2, ..., d, d represent the maximum lag period, and c represent the power sequence y. i The maximum predicted lag period.

[0021] In some embodiments, the processing in step S1 includes: dividing the fluctuating segments of the offshore wind power power sequence into variable time windows, selecting factors affecting photovoltaic power generation, and selecting factors affecting wave power generation.

[0022] In some embodiments, a swing window algorithm is used to divide the fluctuating segments of the offshore wind power sequence into variable time windows.

[0023] In some embodiments, the factors influencing wave energy generation include effective wave height, effective wave period, and power generation.

[0024] In some embodiments, the verification formula for the generalized cross-validation in step S3 is as follows:

[0025]

[0026] Where RSS(S) represents the sum of squares of the residuals, S represents the constraint coefficient, N represents the total number of samples, and Y represents the response variable. i Let represent the i-th term in the response variable vector, and p(S) represent the number of effective parameters in the estimated regression coefficients.

[0027] The present invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0028] The present invention also provides a device for predicting the power range of new energy sources on islands, which stores a computer program that, when executed by a processor, implements the steps of the above-described method.

[0029] The present invention has the following beneficial effects:

[0030] The island renewable energy power range prediction method of this invention can take into account the power output characteristics of offshore wind power, wave energy, island photovoltaic power, and various environmental factors to make range predictions, providing more information for the optimized scheduling of island power grids. Compared with the point prediction in the prior art, range prediction can provide further information on the variability and uncertainty of future prediction values, making the prediction results accurate and reliable. By using LASSO regression to compress factors and reduce the number of features, valuable information can be quickly obtained from various external influencing factors, avoiding the collinearity problem caused by high-dimensional data, which leads to the loss of important elements when selecting variables, thus effectively and accurately predicting the power of island renewable energy. Attached Figure Description

[0031] Figure 1 This is a flowchart of the island new energy power range prediction method in an embodiment of the present invention;

[0032] Figure 2 This is a flowchart of the LASSO quantile regression model in an embodiment of the present invention;

[0033] Figure 3a This is a schematic diagram of the prediction results for 80% of the prediction interval in an embodiment of the present invention;

[0034] Figure 3b This is a schematic diagram of the prediction results for the 85% prediction interval in an embodiment of the present invention;

[0035] Figure 3c This is a schematic diagram of the prediction results for 90% of the prediction interval in an embodiment of the present invention;

[0036] Figure 3d This is a schematic diagram of the 95% prediction result of the prediction interval in an embodiment of the present invention. Detailed Implementation

[0037] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope and application of the present invention.

[0038] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0039] Considering extreme weather warnings and meteorological forecasts for islands, the probability prediction technology for renewable energy on islands is quite complex. However, existing research methods do not consider the prediction and assessment of offshore wind power operation scenarios, nor do they take into account the diverse information regarding new energy sources and the control needs of the upper-level power grid, thus failing to provide more information for the optimized dispatch of island power grids. Furthermore, existing prediction methods cannot simultaneously consider the output characteristics of offshore wind power, wave energy, island photovoltaic power, and various environmental factors.

[0040] Research on numerous current power forecasting methods reveals that most forecasts are deterministic point predictions, i.e., point estimates of future power values, failing to reflect the uncertainty of power. Point predictions cannot provide future predicted values ​​or power variation ranges within a certain confidence interval, nor can they indicate the probability of occurrence at each point within the interval, thus failing to provide further information on the variability and uncertainty of future predicted values.

[0041] Meanwhile, due to the complexity and uncertainty of the factors influencing new energy sources on islands, even considering only the probability density of new energy sources makes it difficult to explain their uncertainty. In this regard, the first major problem to be solved in order to achieve accurate power prediction in a big data environment is how to quickly extract valuable information from various external influencing factors. Methods for identifying important influencing factors in the model include the selection of statistical variables. Traditional variable selection methods often encounter collinearity problems when there are too many parameters (high-dimensional data). Therefore, important elements may be lost when selecting variables, greatly reducing the interpretability of the model.

[0042] This invention provides prediction and evaluation of operational scenarios for offshore renewable energy generation, encompassing various information such as wind power, photovoltaic power, wave energy, and upstream grid regulation requirements, thus offering more information for optimized scheduling of island power grids. Furthermore, the method proposed in this invention takes into account the internal power output characteristics of offshore wind power, island photovoltaic power, and wave energy, as well as various environmental factors, while also considering improving the robustness and generalization effect of the prediction algorithm, thereby enhancing its efficiency and accuracy.

[0043] The embodiments of the present invention are described as follows:

[0044] The fluctuation segment division of the offshore wind power sequence within a variable time window in this embodiment of the invention is as follows: Offshore wind power output fluctuations are random, and using a fixed time window division method is insufficient to effectively capture the fluctuation process. Considering both efficiency and accuracy, this embodiment of the invention employs the swinging door algorithm (SDA) to divide the fluctuation segment of the offshore wind power sequence within a variable time window. The calculation process of SDA is as follows:

[0045]

[0046] In the formula: W L W is the lower limit of the swing window. H ω is the upper limit of the swing window; P0 is the initial power; P(t) is the power at time t; and I is the power sequence at the last moment. The upper and lower limits of the swing window are calculated according to t = t + 1. The fluctuation segment terminates when WL ≥ WH, and the next fluctuation segment is defined until the entire sequence is divided. ω is taken as an empirical value, set at 5% of the wind farm's rated capacity.

[0047] The following analysis of the factors influencing photovoltaic power generation in this embodiment of the invention is provided: Photovoltaic power is generally affected by various meteorological factors such as irradiance, temperature, humidity, wind speed, and wind direction. Depending on the measurement method, irradiance also includes total horizontal irradiance, diffuse horizontal irradiance, total tilted irradiance, and diffuse tilted irradiance. First, the parameters affecting photovoltaic power generation were analyzed, and the main influencing factors were selected.

[0048] The Kendall rank correlation coefficient is a commonly used nonparametric method for measuring the degree of correlation between variables. It can quantitatively and accurately measure the correlation between photovoltaic power and many meteorological factors. The closer the Kendall rank correlation coefficient is to 1, the higher the correlation between this meteorological variable and the output power. This coefficient represents a positive number indicating a positive relationship, and a negative number indicating a negative correlation. The commonly used Pearson correlation coefficient requires that the variables must satisfy an approximately normal distribution, but photovoltaic power and meteorological factor variables do not satisfy this in most cases.

[0049] Neither photovoltaic power nor temperature variables follow a normal distribution in terms of probability density; therefore, the Kendall rank correlation coefficient is a better choice. The Kendall rank correlation coefficient is defined as follows:

[0050]

[0051] In the formula: P represents the number of consistent pairs; Q represents the number of non-consistent pairs. Let A represent the total number of observations, and N represent the total number of observations. When two pairs of observations of variables A and B are... i B i and A j B j Satisfying A i i And at this time A j j If the two pairs of observations are consistent, then they are considered to be consistent; otherwise, they are considered inconsistent.

[0052] After conducting correlation analysis, this invention selected total horizontal radiation, diffuse horizontal radiation, total tilt radiation, diffuse tilt radiation, wind speed, and temperature as the main meteorological factors affecting photovoltaic output.

[0053] The following analysis of the influencing factors for wave power generation in this embodiment of the invention is provided: Due to the characteristics of wave energy, such as volatility, intermittency, and high energy flow density, wave power generation is also volatile and intermittent. Therefore, the volatility and uncertainty of wave power should be fully considered when formulating development plans and scheduling decisions. To reduce the dimensionality of the data, the input data is dimensionality-reduced. After principal component analysis, the input parameters include significant wave height (m), significant wave period (s), and power generation (W).

[0054] The LASSO quantile regression model in this embodiment is conceptualized as follows:

[0055] LASSO (Least Absolute Shrinkage and Selection Operator) model: ​​

[0056] After considering the processing of different types of new energy power generation data, the expression of the multiple regression model based on classical assumptions is as follows:

[0057] Y=Xβ+ε (3)

[0058] In the formula, ε represents the residual vector, ε i ~iidN(0,σ 2 ), indicating that each residual term is independently and identically distributed, ε i This represents the residual term, which follows a normal distribution with a mean of 0 and a variance of σ. 2 ; Response variable Y ~ N(Xβ,σ 2 I), where I is the identity matrix, X is the explanatory variable, and β is the regression coefficient. The objective function is as follows:

[0059]

[0060] By adding an L1 penalty to the objective function, we obtain the following equation:

[0061]

[0062] In the formula, λ is the penalty parameter; the larger the value, the stronger the penalty. ‖β‖1 is the 1-norm, satisfying ‖β‖1=|β1|+|β2|+…+|β k The first term represents the loss function, indicating the fit of the regression model to the data; the second term represents the penalty function, which reduces the regression coefficients of insignificant factors to zero. The optimized formula (5) yields the estimated regression coefficients. Corresponding to the penalty form of formula (5), there is also an equivalent constraint form expression:

[0063]

[0064] st||β||1≤S (6)

[0065] In equation (6), S is the corresponding constraint parameter λ. The smaller the value of S, the stronger the ability of the constraint coefficient β to reach zero. In equation (5), the objective function is a combination of two convex functions; in equation (6), the objective function is convex and the constraint set is convex. Both formulas can yield the optimal solution. However, the expression for the penalty function is a first-order form with absolute values, and therefore cannot derive the angle of the constraint domain. Since LASSO regression is a nonlinear and nondifferentiable function, it is difficult to accurately estimate... and standard deviation.

[0066] Least Angle Regression (LARS):

[0067] The LARS algorithm can effectively solve the LASSO regression problem. The use of the LARS algorithm for LASSO regression has attracted widespread attention in the academic community. The approximate solution is as follows:

[0068]

[0069] Where X is the explanatory variable. It is a diagonal matrix. for The kth element, W - It is the generalized inverse of matrix W, and diag is a diagonal function.

[0070] In formula (6), the value of the constraint parameter S is crucial for solving the model, and it has a decisive impact on the estimation results of the regression coefficients (i.e., the number of regression coefficients with a constraint of 0). This embodiment of the invention uses generalized cross-validation to determine the regression parameters. The statistical data defining the generalized cross-validation class are shown below:

[0071]

[0072] in, Y represents the sum of squares of the residuals, and its value is closely related to S; N represents the total number of samples, and Y represents the response variable. i Representing the i-th term in the response variable vector, p(S) = tr{X(X′X+λW)} - ) -1 X′} approximates the estimated value of the regression coefficient. The number of valid parameters in the matrix, where tr is the trace matrix.

[0073] Quantile regression:

[0074] Suppose the distribution function of the random variable is as follows:

[0075] F(y)=P(Y≤y) (9)

[0076] The quantile τ of Y represents the probability that Y ≤ y. The minimum value of τ is defined as F(y) ≥ τ.

[0077] Right now

[0078] Q Y (τ∣X)=β0(τ)+β1(τ)X1+β2(τ)X2+...+β k (τ)X k =X′β(τ) (10)

[0079] For a set of random samples 1, Y1, Y2 22 , ..., Y n It is affected by k factors 1, X1, X2, ..., Xk The influence of sample quantile regression is to minimize the sum of the absolute values ​​of the weighted errors, i.e.

[0080] Q(τ)=inf{y:F(y)≥τ}; 0<τ<1 (11)

[0081] Q Y In the above formula, (τ|X) represents the response variable in the context of the explanatory variables X=[X1,X2,...,X]. k ] T Conditional quantiles of Y under the given conditions; β(τ)=(β1(τ),β2(τ),...,β k (τ)) T Let f represent the regression coefficient vector at that quantile τ, where inf is the lower bound.

[0082] The expansion formula for sample quantile regression is as follows:

[0083]

[0084] Where n represents the total number of samples, ρ τ (u) is the check function, defined as follows:

[0085] ρ τ (u)=(τ-I(u<0))u (13)

[0086] In equation (11), τ is the quantile, u is the independent variable of the check function, I(m) is the indicator function, and m is the conditional relation. When m is true, I(m) = 1; when m is false, I(m) = 0.

[0087] For the conditional quantile function, solve equation (12) to obtain the parameter estimates, which are expressed as follows:

[0088]

[0089] In formula (14), X i Let represent the i-th value of the explanatory variable vector, and let argmin{...} represent the value of β when the minimum value of this function is taken.

[0090] LASSO quantile regression model:

[0091] The impact of historical power and meteorological factors on short-term renewable energy generation was studied using a LASSO quantile regression model. It was assumed that the power sequence to be predicted was (y1, y2, ..., y...). n ), where n represents the predicted sample size. Meteorological factors are coefficients representing meteorological factors, and the meteorological factor sequence is represented as X. i =(x i-1 ,x i-2,...,x i-d ), where d represents the maximum lag period. Y i =(y i-1 ,y i-2 ,...,y i-c Let represent the historical power with lag period c, where c represents the maximum predicted lag period of the power sequence yi. The established quantile regression model is as follows:

[0092]

[0093] In the formula, α(τ)=(α p (τ)), p=1,2,…,c are the regression coefficients of historical power at the τ quantile, β(τ)=(β q (τ)), q=1,2,…,d are the regression coefficients of the meteorological factors at the quantiles of τ. represents the quantiles from 0 to 1. z(τ) refers to the intercept term, δ i This refers to the error term.

[0094] According to formula (15), the value of new energy power generation y i The expression predicts the following for the τ quantile:

[0095]

[0096] According to formulas (15) and (16), z(τ), α p (τ) and β q The heterogeneity of (τ) further reveals the impact of the selection of different quantiles on the prediction results of new energy power generation. p (τ) and β q The parameter estimate of (τ) can be obtained by the following equation (17).

[0097]

[0098] in, The asymmetric loss function is defined as follows:

[0099]

[0100] Where v represents the independent variable of the asymmetric loss function. In order to solve equation (17), the interior point algorithm can be used to transform the problem into a linear programming problem, which essentially improves the running speed.

[0101] If the influence of meteorological conditions is taken into account, the number of independent variables in the model is relatively large. This embodiment of the invention uses the LASSO quantile regression (LASSO-QR) model to implement the variable selection process. The LASSO quantile regression model adds an L1 penalty to the initial quantile regression objective function, as shown below:

[0102]

[0103] In formula (19), λ is the penalty parameter, whose value is proportional to the penalty intensity. ‖α(τ)‖1and‖β(τ)‖1 are the 1-norm. The first term represents the loss function, indicating the fitting effect of the regression model on the data; the second term represents the penalty function, which reduces the regression coefficients of insignificant factors to zero. To solve the LASSO-QR model, equation (19) is equivalent to the constraint form of the following equation (20).

[0104]

[0105] st‖α(τ)‖1+‖β(τ)‖1≤S (20)

[0106] In formula (20), S represents the constraint coefficient corresponding to λ. The smaller the value of S, the stronger the effect of compressing the regression coefficient to 0. The optimal constraint coefficient S was selected using the generalized cross-validation method. The planning problem in equation (20) can be solved using the LARS algorithm. Through this algorithm, the parameter estimates and conditional quantiles of the new energy power generation to be predicted can be obtained. The LASSO-QR method not only describes the distribution of new energy power generation under different quantiles, but also further selects factors that have important explanatory significance for new energy power generation, reduces the data dimension, optimizes the prediction model, and lays the corresponding foundation for subsequent prediction work.

[0107] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for predicting the power range of new energy sources on islands, which includes the following steps:

[0108] S1: Acquire and process island renewable energy data, and use the processed data as training data. The island renewable energy data includes: wind power data, photovoltaic data, and wave energy data. ； The processing includes: dividing the offshore wind power power sequence into fluctuating segments with varying time windows, selecting factors influencing photovoltaic power generation, and selecting factors influencing wave energy generation. The factors influencing wave energy generation include significant wave height, significant wave period, and power output.

[0109] S2: Input the data into the LASSO quantile regression model, and perform factor compression through LASSO regression to reduce the number of features; the LASSO quantile regression model is expressed by the following formula:

[0110]

[0111] st‖α(τ)‖1+‖β(τ)‖1≤S

[0112] Where S represents the constraint coefficient, ||α(τ)||1 and ||β(τ)||1 represent the L1 norm, α(τ) represents the regression coefficient of historical power at the τ quantile, β(τ) represents the regression coefficient of meteorological factors at the τ quantile, and y i Indicates power, x i Indicates meteorological factors, Let q represent the asymmetric loss function, n represent the total number of samples, p = 1, 2, ..., c, q = 1, 2, ..., d, d represent the maximum lag period, and c represent the power sequence y. i The maximum predicted lag period.

[0113] S3: Perform cross-validation using generalized cross-validation to fine-tune the constraint coefficients and obtain the optimal constraint coefficients; the validation formula for the generalized cross-validation is as follows:

[0114]

[0115] Where RSS(S) represents the sum of squares of the residuals, S represents the constraint coefficient, N represents the total number of samples, and Y represents the response variable. i Let represent the i-th term in the response variable vector, and p(S) represent the number of effective parameters in the estimated regression coefficients.

[0116] S4: Select variables based on the optimal constraint coefficients, determine relevant parameters, and obtain the island new energy power range prediction model;

[0117] S5: Obtain the island's new energy data to be predicted;

[0118] S6: Input the data into the island renewable energy power range prediction model to construct the island renewable energy power prediction range. Preferably, the island renewable energy power prediction range is constructed under a 90% confidence interval.

[0119] Preferably, between steps S4 and S5, the following steps are also included: the test data is input into the island new energy power range prediction model to predict the test data and obtain the prediction result. If the error requirement is not met, the constraint coefficient is reset and prediction is performed again.

[0120] This invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0121] This invention also provides a device for predicting the power range of new energy sources on islands, which stores a computer program that, when executed by a processor, implements the steps of the method described above.

[0122] The flowchart of the new energy power generation prediction based on LASSO quantile regression proposed in this embodiment of the invention is as follows: Figure 2As shown. After obtaining the data, the fluctuation segments of wind power, photovoltaic, and wave energy were divided into variable time windows, and correlation and principal component analyses were performed. Then, the segmented data were input into the LASSO quantile regression model, and factor compression was performed using LASSO. At this point, the number of features in the input data could be filtered to reduce the number of features. The constraint coefficients were then optimized through generalized cross-validation to obtain the optimal constraint coefficients. After setting the optimal constraint coefficients, the prediction results were obtained. That is, through training data, relevant parameters were obtained, and a prediction model for the power range of new energy on the island was established based on the required conditional quantiles. The model was then used to predict the test data and obtain the prediction results. If the error requirements were not met, the optimal constraint coefficients were reset and the prediction was performed again. If the requirements were met, the prediction range was constructed to evaluate the prediction results.

[0123] The assessment and prediction results for the predicted power range of new energy sources on islands include: obtaining the new energy data of the islands to be predicted; inputting the new energy data of the islands into the prediction model of the power range of new energy sources on islands to construct the power prediction range of new energy sources on islands.

[0124] Experimental example:

[0125] The probability interval prediction method proposed in the experimental examples of this invention requires constructing a prediction interval to quantify the uncertainty in the prediction process. Typically, prediction interval coverage probability (PICP) and prediction interval normalized average bandwidth (PINAW) are chosen as reliability evaluation indicators for the prediction interval. The reliability of the prediction interval was evaluated using the PICP method. The prediction interval represents the probability that the actual observed value drops to the upper or lower bound of that interval. The larger the PICP value, the more observations are within the prediction interval. The formula for calculating PICP is as follows:

[0126]

[0127] Where n is the number of predicted samples, μ i It is a Boolean variable. μ i The definition is as follows:

[0128]

[0129] In formula (22), if the power load value is y i The prediction is that the value falls within the interval M. i U i Above, μ i =1 Otherwise, μ i=0. The optimal value of PICP is 100%, indicating that all targets are within the prediction range. However, an excessive pursuit of reliability can lead to an overly large prediction interval, failing to provide effective predictions and thus negating the importance of decision-making. The normalized average bandwidth (PINAW) of the prediction interval is an important criterion for evaluating its quality, reflecting its width. In practical applications, PINAW is usually used as a relatively small value to ensure a narrow prediction interval. Furthermore, the value of PICP generally satisfies the condition of being no less than the confidence interval of the prediction interval. The PINAW formula is defined as follows:

[0130]

[0131] Where N represents the total number of samples, Q is the range of variation of the predicted target value, and the average bandwidth of the prediction interval is normalized as a percentage.

[0132] The input variables in the experimental examples of this invention are the historical renewable energy power and meteorological values ​​before node prediction, used to verify the method proposed in the experimental examples of this invention. A rolling prediction method is used to predict the power values ​​of 720 nodes, including candidate independent variables. In specific applications, the LASSO regression method is used to extract important variables from the training sample data. Through the generalized cross-validation algorithm, the optimal constraint parameter S is selected as 0.3, and important variables are selected for renewable energy power generation prediction. Combined with the quantile regression method, a LASSO-QR model is established, predicting power output at 15-minute intervals, and the results are analyzed.

[0133] The prediction results of the prediction interval based on the LASSO-QR method in the experimental examples of this invention are as follows: Figure 3a , 3b As shown in Figures 3c and 3d, it can be seen that the probability intervals obtained by the LASSO-QR method match the real values ​​well, with almost all real values ​​falling within the prediction range obtained by the LASSO-QR method.

[0134] Table 1 shows the PICP and PINAW values ​​at different confidence levels in the experimental examples of this invention. Table 1 displays the results of PICP and PINAW at different confidence levels. It can be seen that the higher the confidence level, the larger the values ​​of PICP and PINAW. That is, the reliability and width of the interval increase. Overall, it can be seen that the PICP and PINAW values ​​of the LASSO-QR method are excellent.

[0135] Table 1

[0136]

[0137] In the experimental examples of this invention, PICP and PINAW were performed using different methods, as shown in Table 2. Table 2 presents the prediction interval coverage probability and normalized average width of LASSO-QR and QR. As can be seen from Table 2, the constructed prediction interval largely covers the actual values. The PICP value without variable selection is lower than the PICP value without variable selection. Meanwhile, the PINAW value should be exactly the opposite. Therefore, we can conclude that compared with quantile regression, the probability density prediction method based on LASSO-QR has a stronger ability to reduce prediction uncertainty and a narrower prediction interval, which can meet the preset confidence level.

[0138] Table 2 shows that LASSO-QR outperforms QR in both PICP and PINAW. This means it has a wider coverage and a narrower time interval. The above analysis results indicate that the probability density prediction method based on LASSO-QR achieves a more accurate prediction interval than the QR method.

[0139] Table 2

[0140]

[0141] This invention proposes a probabilistic prediction method for renewable energy output on islands based on LASSO quantile regression, addressing the challenges of quantile regression requiring extensive domain knowledge and complex parameters. Compared to existing probabilistic prediction methods, the proposed method considers the output characteristics of offshore wind power, wave energy, and island photovoltaic power, as well as various environmental factors. It utilizes accurate confidence interval grading for probabilistic prediction while simultaneously enhancing the robustness and generalization effect of the prediction algorithm, thereby improving its efficiency and accuracy.

[0142] Probabilistic forecasting technology, by estimating the probability distribution of forecast uncertainty, enables quantitative analysis of forecast uncertainty, providing richer and more accurate uncertainty information for new energy forecasting on islands. It can help grid dispatch operators better understand the range of future power fluctuations and improve the flexibility of grid operation.

[0143] The new energy probability interval prediction method overcomes the shortcomings of traditional point prediction methods. It can obtain the future power value and power variation range under the given confidence interval. By constructing the prediction range of the predicted power, more detailed information can be obtained than existing point prediction methods.

[0144] Therefore, obtaining the predicted probability range of new energy sources that takes into account external factors, and using this method, can yield more comprehensive and effective information on new energy sources, which can help power system decision-makers make scientific decisions and avoid huge errors and economic losses.

[0145] The method proposed in this invention is based on historical data and can effectively predict the power value of new energy sources on islands by making full use of the characteristics of multiple feature variables.

[0146] The method proposed in this invention has higher accuracy than existing similar methods and is a promising method for predicting renewable energy power on islands.

[0147] The method proposed in this invention solves the problems of insufficient algorithm accuracy, lack of relevant prediction methods, and excessive algorithm time in the prior art.

[0148] The present invention presents a method for predicting the probability of new energy sources for island development that can be applied on a large scale.

[0149] The embodiments of the present invention can select feature variables through variable screening, and at the same time improve the interpretability of probability prediction, which is of great significance for applications such as probability prediction of new energy.

[0150] This invention provides a foundation for large-scale probability prediction of new energy sources on islands by systematically studying the probability prediction of new energy sources on islands, analyzing and summarizing the mechanism of probability prediction.

[0151] The embodiments of this invention employ quantile regression, which simplifies the prediction process. Furthermore, the use of LASSO variable screening shortens the prediction time and effectively saves costs.

[0152] The specific application scenarios of this invention are as follows:

[0153] 1. Operations such as probability prediction of new energy sources on islands.

[0154] 2. It can realize the probability prediction of wind power on islands and can be provided to scientific research projects such as scientific research institutes, R&D institutions, and commercial companies.

[0155] 3. It can realize the probability prediction of photovoltaic power on islands.

[0156] 4. It can realize the probability prediction of wave energy of islands.

[0157] 5. Use variable filtering to select characteristic variables.

[0158] The beneficial effects of the embodiments of the present invention are as follows:

[0159] 1) This method can be used for power probability prediction in offshore photovoltaic power generation, greatly increasing the reliability of the prediction.

[0160] 2) This method can perform power probability prediction in offshore wind power generation, greatly increasing the reliability of the prediction. Its variable screening property reduces costs. Compared with existing point prediction products, the probability prediction formed by the technology of this invention is more stable.

[0161] 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 embodied 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.

[0162] 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.

[0163] 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.

[0164] 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.

[0165] The above description provides a further detailed explanation of the present invention in conjunction with specific / preferred embodiments, and it should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various substitutions or modifications can be made to these described embodiments without departing from the concept of the present invention, and all such substitutions or modifications should be considered within the scope of protection of the present invention. In the description of this specification, the reference to terms such as "an embodiment," "some embodiments," "preferred embodiment," "example," "specific example," or "some examples," etc., indicates that the specific features, structures, materials, or characteristics described in connection with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples. Although the embodiments of the present invention and their advantages have been described in detail, it should be understood that various changes, substitutions, and modifications can be made herein without departing from the scope of protection of the patent application.

Claims

1. A method for predicting the power range of new energy sources on islands, characterized in that, Includes the following steps: S1: Acquire and process island new energy data, and use the processed data as training data. The island new energy data includes: wind power data, photovoltaic data and wave energy data. The processing includes: using a swing window algorithm to divide the fluctuation segment of the offshore wind power sequence into variable time windows; using Kendall's rank correlation coefficient to perform correlation analysis on the influence parameters of photovoltaic power generation; selecting the influencing factors of photovoltaic power generation; and selecting the influencing factors of wave energy power generation. The factors influencing wave energy generation include significant wave height, significant wave period, and power generation. The Kendall rank correlation coefficient is defined as follows: In the formula: P represents the number of consistent pairs; Q represents the number of non-consistent pairs. Let N represent the total number of observations, where A is the total number of observations and B is the total number of observations. i B i and A j B j Satisfying A i < B i And at this time A j < B j If the two pairs of observations are consistent, then they are considered to be consistent; otherwise, they are considered inconsistent. S2: Construct a LASSO quantile regression model. Input the data into the LASSO quantile regression model and perform factor compression through LASSO regression to reduce the number of features. S3: Cross-validation is performed using generalized cross-validation to fine-tune the constraint coefficients of the LASSO quantile regression model and obtain the optimal constraint coefficients. S4: Select variables based on the optimal constraint coefficients, determine relevant parameters, and obtain a trained island new energy power range prediction model; S5: Obtain the island's new energy data to be predicted; S6: Input the island new energy data to be predicted into the island new energy power range prediction model to construct the island new energy power prediction range.

2. The method as described in claim 1, characterized in that, Between steps S4 and S5, there are the following steps: the test data is input into the island's new energy power range prediction model to predict the test data and obtain the prediction result. If the error requirement is not met, the constraint coefficients are reset and the prediction is performed again.

3. The method as described in claim 1, characterized in that, Step S6 includes: constructing a power prediction range for new energy sources on the island under a 90% confidence interval.

4. The method as described in claim 1, characterized in that, The LASSO quantile regression model described in step S2 is expressed by the following formula: in, S Represents the constraint coefficient. Denotes the 1-norm. Indicates historical power at The regression coefficient at the quantile, Indicating meteorological factors in The regression coefficient at the quantile, y i Indicates power, Indicates meteorological factors, Represents the asymmetric loss function. n Represents the total number of samples. , , d Indicates the maximum lag period. c Represents power sequence y i The maximum predicted lag period.

5. The method as described in claim 1, characterized in that, The verification formula for the generalized cross-validation described in step S3 is as follows: ； in, This represents the sum of squares of the residuals. S Represents the constraint coefficient. N Represents the total number of samples. Y Represents the response variable. Y i Represents the first in the response variable vector i item, This indicates the number of effective parameters in the estimated regression coefficients.

6. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-5.

7. A device for predicting the power range of new energy sources on an island, wherein a computer program is stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-5.

Images