A probabilistic forecasting method for offshore wind power based on composite quantile regression
The composite quantile regression model addresses the variability and uncertainty in offshore wind power by optimizing sensitivity and reliability, resulting in enhanced prediction interval performance and improved forecasting accuracy.
Patent Information
- Application Number
- CN202111059147.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-10
- Publication Date
- 2024-12-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing wind power prediction methods struggle with variability and uncertainty, particularly in offshore wind power, due to the difficulty in determining the actual error distribution and lack of sensitivity optimization in traditional quantile regression models, leading to inefficient and unreliable prediction intervals.
A composite quantile regression model is developed, incorporating sensitivity optimization into the cost function, balancing reliability and sensitivity through an adjustment coefficient, and optimizing the prediction interval performance using linear programming.
The method achieves improved reliability and comprehensive performance of prediction intervals, enhancing the accuracy and efficiency of offshore wind power forecasting.
Smart Images
Figure CN113902170B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to renewable energy power generation and smart grid technology, and in particular to an offshore wind power probability prediction method based on composite quantile regression. Background Art
[0002] Compared with traditional thermal power generation, wind power is clean, renewable and sustainable. As an environmentally friendly form of electricity, wind power has relatively low operation and maintenance costs and power generation costs, occupies less land, has flexible development scale, and has good large-scale development conditions and commercial development prospects. At present, wind power technology has become one of the most recognized, fastest-growing and most mature renewable energy power generation technologies.
[0003] Under the "dual carbon" goal, wind power has become the core path for low-carbon or even zero-carbon energy transformation in many countries, including China. my country has sufficient offshore wind energy resources, with an available sea area of more than 3 million square kilometers. In the future, with technological progress and the decline in the cost of electricity, there is huge room for the development of offshore wind power. Compared with the instability of onshore wind power, offshore wind power has obvious advantages over onshore wind power. On the one hand, offshore wind power does not occupy onshore resources; on the other hand, in the same geographical location, the effective power generation hours of offshore wind turbines are 20% to 70% higher than those of onshore wind turbines.
[0004] Although wind energy has high development value due to its significant advantages such as renewable, pollution-free and low cost, it is difficult to achieve efficient, economical and safe use. The size of traditional thermal power generation depends entirely on fuel input. Its output power can be quickly and effectively controlled within a large range according to the dispatching instructions of the power system to cope with real-time changes in load. However, the output of wind power has strong randomness and uncontrollability. The output of wind power cannot be controlled at will, and it changes according to different time periods, meteorological conditions, etc. As the proportion of wind power installed capacity in the power system increases, the challenges brought by wind power grid connection to the power system are mainly due to its variability and uncertainty. In view of the uncertainty of wind power output, studying wind power probability prediction technology is an important solution.
[0005] At present, probability interval prediction mainly adopts probability interval prediction based on parameterized model. The construction of prediction interval is based on two parts: point prediction and uncertainty analysis. After the point prediction is completed, it is assumed that the wind power prediction error satisfies a certain distribution, such as beta distribution, normal distribution, etc., and then the prediction error is calculated according to the pre-assumed distribution, and the point prediction value is added to form the calculation of the upper and lower limits of the interval. However, the actual wind power fluctuates greatly and is highly complex, so it is difficult to determine the actual error distribution.
[0006] Non-parametric modeling is an effective method to solve the above problems, among which the most efficient and widely used method is the quantile regression model. Therefore, the probability interval prediction method based on the quantile regression model is also increasingly valued by technical personnel. The traditional linear quantile regression method is often used for regression analysis and prediction in statistical regression analysis. In order to improve the traditional quantile regression model, the extreme learning machine model is currently used to improve the quantile regression method and improve the linear model to a nonlinear model. The comprehensive performance evaluation index of the prediction interval is the comprehensive score of the prediction interval, which comprehensively considers the reliability and sensitivity of the prediction interval. However, the traditional quantile regression model only considers the reliability optimization of the prediction interval and lacks attention to sensitivity. Therefore, it is of great significance to introduce the sensitivity optimization strategy into the quantile regression model to improve the performance of the prediction interval. Summary of the invention
[0007] Purpose of the invention: In order to further improve the quality of wind power prediction interval, the present invention provides an offshore wind power probability prediction method based on composite quantile regression.
[0008] In order to solve the above technical problems, the offshore wind power probability prediction method based on composite quantile regression described in the present invention comprises the following steps:
[0009] (1) Processing the data, initializing the coefficients from the input layer to the hidden layer and the threshold of the hidden layer of the extreme learning machine model, predicting the rated confidence interval, importing the normalized historical wind power time series, and constructing the sample;
[0010] (2) Construct a composite quantile regression model, introduce the acuity index optimization into the cost function of traditional quantile regression, and combine it with reliability optimization to form a composite index;
[0011] The cost function of the composite quantile regression model is as follows:
[0012]
[0013] Among them, α is the confidence level, and is the transition variable, λ is the adjustment coefficient, and the cost function is optimized based on the linear programming method;
[0014] (3) balancing the weights of reliability and sensitivity in the composite indicator through an adjustment coefficient, adjusting the prediction interval performance, and optimizing the adjustment coefficient based on the prediction interval performance of the training set;
[0015] (4) Based on the optimal adjustment coefficient, the output coefficient of the composite quantile regression model is optimized, the training is completed, and the prediction interval is output through the wind power time series.
[0016] Furthermore, in step (1), the data is processed, that is, the coefficients from the input layer to the hidden layer and the threshold of the hidden layer of the extreme learning machine model are initialized, the rated confidence interval is predicted, and the normalized historical wind power time series is imported. The sample construction specifically includes the following steps:
[0017] (1.1) Initialize the hidden layer coefficients and thresholds of the extreme learning machine in sequence;
[0018] (1.2) Set the boundary quantile percentages of the quantile regression model so that the upper quantile percentage minus the lower quantile percentage is the confidence interval percentage;
[0019] (1.3) Import the historical wind power time series and normalize the series;
[0020] (1.4) Construct the input and output sample set of the time series, which is expressed as follows:
[0021]
[0022] Among them, x i and i are wind power values, are the input and output of the i-th group of samples, and S is the number of samples.
[0023] Furthermore, in step (2) of constructing the composite quantile regression model, the quantile regression sensitivity introduced is as follows:
[0024]
[0025] Among them, G() is the sensitivity optimization term, T is the total number of samples, g() is the output of the extreme learning machine, and w α The upper and lower quantile percentages are and α The output coefficient of
[0026] The cost function of the composite quantile regression model satisfies the following constraints:
[0027]
[0028] 0≤g(x t ,w α )≤1
[0029] -|β α,j |≤w α,j ≤|β α,j |
[0030]
[0031]
[0032] Among them, y t is the measured value.
[0033] Furthermore, in step (3), the weights of reliability and sensitivity in the composite index are balanced by adjusting the adjustment coefficient, the prediction interval performance is adjusted, and the adjustment coefficient is optimized according to the training set prediction interval performance, which specifically includes the following steps:
[0034] (3.1) Optimize the composite quantile regression model based on the adjustment coefficient value and calculate the prediction interval of the training set;
[0035] (3.2) Calculate the reliability of the prediction interval as follows:
[0036] |ACE|=|PICP-PINC|
[0037] Among them, |ACE| is the coverage deviation of the prediction interval, which is an indicator to measure the reliability of the prediction interval, PICP is the percentage of the prediction interval covering the measured value, and PINC is the rated confidence level.
[0038] (3.3) Calculate the comprehensive performance of the prediction interval as follows:
[0039]
[0040]
[0041]
[0042] in, Provide a comprehensive score for the prediction interval to comprehensively evaluate the performance of the prediction interval; Rate each point; is the prediction interval width; and are the upper and lower quantiles of the prediction interval, respectively.
[0043] Furthermore, in step (4), the output coefficient of the composite quantile regression model is optimized according to the optimal adjustment coefficient, the training is completed, and the prediction interval is output through the wind power time series, which specifically includes the following steps:
[0044] (4.1) According to the optimization results of the prediction interval adjustment coefficient, based on the training set input sample set, the linear programming method is used to optimize the composite quantile regression model, calculate the optimal output coefficient, and complete the model training;
[0045] (4.2) Import the input of the test sample and calculate the prediction interval based on the trained model.
[0046] Beneficial effects: Compared with the prior art, the prediction interval performance of the present invention is significantly better than the traditional prediction model based on the quantile regression model, and has better reliability and comprehensive performance. For example, the conditional quantile is based on clustering theory. The present invention can be applied to offshore wind power data, and the output ratio of offshore wind power is higher than that of land. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a schematic diagram of the method flow of the present invention;
[0048] Figure 2 It is the relationship diagram between the adjustment coefficient and the prediction interval performance;
[0049] Figure 3 This is a one-hour-ahead forecast interval diagram of spring wind power data at 90% confidence level of the present invention;
[0050] Figure 4 This is a one-hour-ahead prediction interval diagram of summer wind power data at 90% confidence level of the present invention;
[0051] Figure 5 This is a one-hour-ahead forecast interval diagram of autumn wind power data at 90% confidence level of the present invention;
[0052] Figure 6 This is a one-hour-ahead prediction interval diagram of winter wind power data at 90% confidence level of the present invention. DETAILED DESCRIPTION
[0053] In order to explain the technical solution disclosed in the present invention in detail, further elaboration is made below in conjunction with the accompanying drawings and specific embodiments of the specification.
[0054] The present invention provides a method for probabilistic forecasting of offshore wind power based on composite quantile regression, which can be applied to other ranges and fields such as load, wind power / photovoltaic output, etc.
[0055] The prediction model flow chart of the present invention is as follows Figure 1 As shown, the steps of the embodiment are mainly as follows:
[0056] (1) Processing the data, i.e. initializing the coefficients from the input layer to the hidden layer and the threshold of the hidden layer of the extreme learning machine model, predicting the rated confidence interval, importing the normalized historical wind power time series, and constructing the sample;
[0057] (2) Construct a composite quantile regression model, introduce the acuity index optimization into the cost function of traditional quantile regression, and combine it with reliability optimization to form a composite index;
[0058] (3) balancing the weights of reliability and sensitivity in the composite indicator through an adjustment coefficient, adjusting the prediction interval performance, and optimizing the adjustment coefficient based on the prediction interval performance of the training set;
[0059] (4) Based on the optimal adjustment coefficient, the output coefficient of the composite quantile regression model is optimized, the training is completed, and the prediction interval is output through the wind power time series.
[0060] The step 1 specifically includes two sub-steps: model parameter initialization and data preprocessing and importing. The specific method of model parameter initialization is:
[0061] Step 1.11: Initialize the hidden layer coefficients and thresholds of the extreme learning machine;
[0062] Step 1.12: Set the upper and lower quantile values of the quantile regression model based on the confidence interval.
[0063] The specific method of data preprocessing is:
[0064] Step 1.21: Import wind power data and normalize the time series;
[0065] Step 1.22: Construct the input and output sample set of the time series Among them, x i and i are wind power, are the sample input and measured values of the i-th group, and S is the number of samples.
[0066] The step 2 specifically constructs a composite quantile regression model, introduces the acuity index optimization into the cost function of the traditional quantile regression, and combines it with reliability optimization to form a composite index. The specific method is:
[0067] Step 2.1: The introduced quantile regression acuity is as follows:
[0068]
[0069] Among them, G() is the sensitivity optimization term, T is the total number of samples, g() is the output of the extreme learning machine, and w α The upper and lower quantile percentages are and α The output coefficient of
[0070] Step 2.2: The cost function of the composite quantile regression model is as follows:
[0071]
[0072] Among them, α is the confidence level, and is the transition variable, λ is the adjustment coefficient, and the cost function is optimized based on the linear programming method;
[0073] Step 2.3: The above cost function satisfies the following constraints:
[0074]
[0075] 0≤g(x t ,w α )≤1
[0076] -|β α,j |≤w α,j ≤|β α,j |
[0077]
[0078]
[0079] Among them, y t is the measured value.
[0080] The step 3 specifically balances the weights of reliability and sensitivity in the composite indicator by adjusting the coefficient, adjusting the prediction interval performance, and optimizing the adjustment coefficient according to the training set prediction interval performance. The specific method is:
[0081] Step 3.1: Complete the optimization of the composite quantile regression model according to the adjustment coefficient value and calculate the prediction interval of the training set;
[0082] Step 3.2: Calculate the reliability of the prediction interval as follows:
[0083] |ACE|=|PICP-PINC|
[0084] Among them, |ACE| is the coverage deviation of the prediction interval, which is an indicator to measure the reliability of the prediction interval, PICP is the percentage of the prediction interval covering the measured value, and PINC is the rated confidence level.
[0085] Step 3.3: Calculate the overall performance of the prediction interval as follows:
[0086]
[0087]
[0088]
[0089] in, Provide a comprehensive score for the prediction interval to comprehensively evaluate the performance of the prediction interval; Rate each point; is the prediction interval width; and are the upper and lower quantiles of the prediction interval, respectively.
[0090] The step 4 is specifically to optimize the output coefficient of the composite quantile regression model according to the optimal adjustment coefficient, complete the training, and output the prediction interval through the wind power time series, which specifically includes the following steps:
[0091] Step 4.1: According to the optimization results of the prediction interval adjustment coefficient, based on the training set input sample set, the linear programming method is used to optimize the composite quantile regression model, the optimal output coefficient is calculated, and the model training is completed;
[0092] Step 4.2: Import the input of the test sample and calculate the prediction interval based on the trained model.
[0093] In order to make those skilled in the art better understand the technical solution of the present invention and verify the effectiveness of the method of the present invention, the actual wind farm power is used as an example for detailed introduction. The composite quantile regression model (CQR) of the present invention is compared with the existing methods, including the extreme learning machine model (BELM) based on the bootstrap sampling method, the quantile regression model (QR), and the linear quantile regression model (LQR).
[0094] The power time series of an offshore wind farm in my country is used to verify the ultra-short-term probability interval prediction performance of the model. The data resolution is 15 minutes, and the probability prediction of wind power is carried out one hour ahead. The data uses the 15-day wind power time series from the four seasons of 2019. Each set of case tests takes 15 days of data, the first 11 days of data are used for training, and the last 4 days of data are used for testing.
[0095] The performance of interval prediction is generally evaluated by observing two indicators, the average coverage deviation (ACE) and the score. The calculation method of ACE is as follows:
[0096] |ACE|=|PICP-PINC|
[0097] Among them, PICP is the actual interval coverage, PINC is the rated confidence, and the closer the ACE value is to 0, the better.
[0098] When the coverage deviation is close to 0, the interval width should be as narrow as possible. The calculation formula is as follows:
[0099]
[0100] in, is the width of the i-th predicted value under the interval confidence interval α. The average interval width (AW) indicates the sensitivity of the prediction interval. and are the upper and lower boundaries of the interval corresponding to the i samples respectively.
[0101] The interval comprehensive score takes into account two factors: interval coverage deviation and interval width. The calculation formula is as follows:
[0102]
[0103]
[0104] It is the comprehensive score of the interval, and the value is a negative number. The closer the value is to 0, the better the comprehensive performance of the prediction interval. For the indicators to evaluate the prediction interval, the comprehensive score of the interval is mainly considered. The average coverage deviation is generally used to refer to the coverage performance of the interval, that is, the coverage needs to be close to the rated confidence level to prove that the prediction interval has good reliability.
[0105] Figure 2 The relationship between different regulation coefficients and prediction interval performance of the wind power training set data for 15 days in mid-January. The numerical relationship between the prediction interval comprehensive score (Intervalscore), coverage (PICP) and regulation coefficient (Regulation coefficient) is shown in the figure. In the figure, the confidence level is 90%, the prediction scale is 1 hour ahead, and the resolution of the regulation coefficient is 0.01. As shown in the figure, when λ=0.06, the prediction interval performance is optimal. Based on this training result, the quantile regression model is used as a reference, including LQR and QR, and the performance comparison is shown in Table 1.
[0106] Table 1 January data test results
[0107]
[0108] The comparison of prediction interval performance is shown in Table 1. The CQR proposed in the present invention has the best comprehensive prediction interval performance, optimal sensitivity and good reliability, which verifies the effectiveness of the proposed model.
[0109] Table 2 Test results of four seasons data
[0110]
[0111] Table 2 shows the comparison of the method proposed in this invention with several classic methods. The 15-day data sets used are from April, July, October and January 2019, respectively, to test the performance of the prediction model in different seasons. The comprehensive performance is the decisive indicator for evaluating the prediction interval, and the accuracy of interval coverage is combined to comprehensively compare the results of probability prediction. The test results of the data of each month under different confidence levels show that the prediction interval obtained by the method of this invention has the best performance.
[0112] Figures 3 to 6The prediction results of the four seasons with a confidence interval of 90% and one hour ahead are shown, and the data is used after normalization. It can be seen that the time series of wind power is complex and it is difficult for parametric modeling to accurately estimate the distribution of errors and accurately give a prediction interval. It can be seen from the figure that the interval constructed by the method of the present invention has a good prediction effect. In summary, the present invention can achieve high-quality ultra-short-term prediction intervals for wind power and can be used in practical engineering applications.
Claims
1. A method for offshore wind power probability prediction based on composite quantile regression, characterized in that: The steps include: (1) Importing normalized historical wind power time series to construct sample data, including processing the data, initializing the coefficients from the input layer to the hidden layer and the threshold of the hidden layer of the extreme learning machine model, and predicting the rated confidence interval; (2) Construct a composite quantile regression model, introduce the acuity index optimization into the cost function of traditional quantile regression, and combine it with reliability optimization to form a composite index; The cost function of the composite quantile regression model is as follows: Among them, α is the confidence level, and is the transition variable, λ is the adjustment coefficient, and the cost function is optimized based on the linear programming method; In the composite quantile regression model constructed in this step, the introduced quantile regression sensitivity is as follows: Among them, G() is the sensitivity optimization term, T is the total number of samples, g() is the output of the extreme learning machine, and w α The upper and lower quantile percentages are and α The output coefficient of The cost function of the composite quantile regression model satisfies the following constraints: 0≤g(x t ,w α )≤1 -|β α,j |≤w α,j ≤|β α,j Among them, y t is the measured value; (3) balancing the weights of reliability and sensitivity in the composite indicator through an adjustment coefficient, adjusting the prediction interval performance, and optimizing the adjustment coefficient based on the prediction interval performance of the training set; In this step, the weights of reliability and sensitivity in the composite index are balanced by adjusting the coefficient, the prediction interval performance is adjusted, and the adjustment coefficient is optimized according to the training set prediction interval performance, which specifically includes the following steps: (3.1) Complete the optimization of the composite quantile regression model according to the adjustment coefficient value and calculate the prediction interval of the training set; (3.2) Calculate the reliability of the prediction interval. The calculation expression is as follows: |ACE|=|PICP-PINC| Among them, |ACE| is the coverage deviation of the prediction interval, which is an indicator to measure the reliability of the prediction interval, PICP is the percentage of the prediction interval covering the measured value, and PINC is the rated confidence level; (3.3) Calculate the comprehensive performance of the prediction interval. The calculation process is as follows: in, Provide a comprehensive score for the prediction interval to comprehensively evaluate the performance of the prediction interval; Rate each point; is the prediction interval width; and are the upper and lower quantiles of the prediction interval, respectively; (4) Based on the optimal adjustment coefficient, the output coefficient of the composite quantile regression model is optimized, the training is completed, and the prediction interval is output through the wind power time series.
2. The offshore wind power probability prediction method based on composite quantile regression according to claim 1 is characterized in that: In step (1), the data is processed, that is, the coefficients from the input layer to the hidden layer and the threshold of the hidden layer of the extreme learning machine model are initialized, the rated confidence interval is predicted, the normalized historical wind power time series is imported, and the sample is constructed, which specifically includes the following steps: (1.1) Initialize the hidden layer coefficients and thresholds of the extreme learning machine in sequence; (1.2) Set the boundary quantile percentages of the quantile regression model so that the upper quantile percentage minus the lower quantile percentage is the confidence interval percentage; (1.3) Import the historical wind power time series and normalize the series; (1.4) Construct the input and output sample set of the time series, whose expression is as follows: Among them, x i and i are wind power, are the input and output of the i-th group of samples, and S is the number of samples.
3. The offshore wind power probability prediction method based on composite quantile regression according to claim 1 is characterized in that: In step (4), the output coefficient of the composite quantile regression model is optimized according to the optimal adjustment coefficient, the training is completed, and the prediction interval is output through the wind power time series, which specifically includes the following steps: (4.1) According to the optimization results of the prediction interval adjustment coefficient, based on the training set input sample set, the linear programming method is used to optimize the composite quantile regression model, calculate the optimal output coefficient, and complete the model training; (4.2) Import the input of the test sample and calculate the prediction interval based on the trained model.