Long-Term Prediction and Extrapolation Method for Offshore Wind Power Output Characteristics Based on Convolutional Neural Network

Abstract

The present invention relates to a long-term prediction extrapolation method for the output characteristics of offshore wind power based on a convolutional neural network. Through the convolutional neural network analysis method, the method obtains the quantitative extraction of the three-dimensional atmospheric structure characteristics, thereby using the spatial feature correlation analysis of meteorological element characteristics and the historical active power of regional offshore wind power stations in the same period to give the key characteristic factors affecting the active power of regional offshore wind power, and then establishing an extrapolation model for the long-term prediction of the output characteristics of regional offshore wind power to provide an overall trend analysis of offshore wind power for the power grid. The present invention obtains the correlation model of the output characteristics of regional offshore wind power between different seasons through the climate factor correlation analysis method, thereby establishing a long-term estimation of the main trend of the output characteristics of cross-seasonal wind power generation considering climate characteristic analysis.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy, and particularly relates to a long-term prediction extrapolation method for the output characteristics of offshore wind power based on a convolutional neural network. Background Art

[0002] Wind power generation is a relatively mature and widely applied power generation type in the technical field of new energy. Among them, offshore wind power relies on the stable operation of the wind on the sea for power generation, with higher fan utilization rates, larger single-unit installed capacities, and being close to the power consumption load. It has obvious advantages and prospects. Compared with onshore wind power generation, due to significant differences in multiple factors such as the underlying surface, circulation, and climate, the ocean surface is relatively flat, the wind has basically no resistance, is not affected by the undulation of terrain, and has relatively high and stable average wind speeds. Moreover, the ocean surface area is occasionally affected by extreme weather processes such as typhoons. The impact on the output characteristics of offshore wind power cannot be ignored. It is urgent to better improve the output prediction level of offshore wind power to support the development of offshore wind power. Summary of the Invention

[0003] The purpose of the present invention is to provide a long-term prediction extrapolation method for the output characteristics of offshore wind power based on a convolutional neural network. Through the convolutional neural network analysis method, quantitative extraction of the three-dimensional atmospheric structure characteristics is obtained, so as to use the spatial feature correlation analysis of meteorological element characteristics and the historical synchronous active power of regional offshore wind power stations to give the key characteristic factors affecting the active power of regional offshore wind power, and then establish an extrapolation model for long-term prediction of the output characteristics of regional offshore wind power, providing overall trend analysis data of offshore wind power generation for provincial power grids.

[0004] To achieve the above purpose, the technical solution of the present invention is: a long-term prediction extrapolation method for the output characteristics of offshore wind power based on a convolutional neural network, including the following steps:

[0005] S1. Calibrate the time stamps of the output data of each offshore wind power in the region, and calculate the total regional output every 15 minutes, that is, the regional wind power output;

[0006] S2. According to the time series of the regional offshore wind power output, apply the decision tree classification algorithm to classify and identify the wind types of the daily time series, and extract three main categories: high-wind days, low-wind days, and calm-wind days, to form the time series of the daily regional wind power output wind types for each month in history; based on the time series of the daily regional wind power output wind types for each month, obtain the probability of high-wind days, low-wind days, and calm-wind days for each month;

[0007] S3. Extract climate index analysis variables from climate data. Select three indices, namely the Arctic Oscillation (AO), the North Atlantic Oscillation (NAO), and the Pacific-North American teleconnection (PNA), which are closely related to ocean surface winds, as climate index analysis variables. All three indices are monthly average indices.

[0008] S4. Establish multiple linear regression estimation models for the monthly probability time series of strong wind days, light wind days, and calm wind days respectively with AO, NAO, and PNA.

[0009] S5. Extract the forecast information on the cross-seasonal time scale from the future forecasts of AO, NAO, and PNA released by authoritative meteorological research institutions, and substitute it into the multiple linear regression estimation model to obtain the cross-seasonal regional wind power output specific cross-level estimation results.

[0010] In an embodiment of the present invention, the specific implementation of step S1 is as follows:

[0011] Sort out the offshore wind power output data, select all the historical offshore wind farm stations in the same period in the region to form a set, and use the Energy Management System (EMS) to obtain the output data of each wind turbine every 15 minutes.

[0012] For the output data of each wind turbine every 15 minutes obtained, remove the power-limiting data caused by active power control, and obtain the regional wind power output data every 15 minutes through summation calculation. The time length of the regional wind power output data to be calculated shall be not less than 3 years.

[0013] In an embodiment of the present invention, the specific implementation of step S2 is as follows:

[0014] Adopt the look-up table method to process the regional wind power output data every 15 minutes into time series with daily, monthly, and annual time scales, where the daily, monthly, and annual representations are respectively i, j, k ; the regional wind power output data at a certain moment is denoted as P ( i, j, k );

[0015] Using the 10-meter wind speed data and adopting the ideal wind power conversion model, calculate the ideal regional wind power output data denoted as P C ( i, j, k );

[0016] Select the CART decision tree calculation program. The training set consists of P ( i, j, k ) and P C ( i, j, k ); Taking the day as the time unit, for P ( i, j, k) Classify the wind type characteristics to obtain the wind type category to which any day in the period to be analyzed belongs. The wind type categories include three main categories: high wind days, light wind days, and calm wind days, which are respectively denoted as heavy, weak, mediocre ; The daily characteristics of the regional wind power output after processing are respectively denoted as P ( heavy, j, k )、 P ( weak, j, k ) and P ( mediocre, j, k );

[0017] Statistically obtain the probabilities of P(heavy, j, k), P(weak, j, k), and P(mediocre, j, k) in any month.

[0018] In an embodiment of the present invention, the specific implementation of step S3 is as follows:

[0019] Let the change curves of the monthly wind types heavy, weak, mediocre within a certain period and the monthly P ( heavy, j, k ), P ( weak, j, k ) and P ( mediocre, j, k ) probabilities be used as quantitative indicators to describe the characteristics of the regional wind power output;

[0020] Calculate the correlation coefficients between the monthly AO, NAO, PNA indices and P ( heavy, j, k ), P ( weak, j, k ) and P ( mediocre , j, k ), and use the t test method to determine the correlation significance;

[0021] Obtain the climate factor index combinations related to the monthly wind power output characteristics P(heavy, j, k), P(weak, j, k), and P(mediocre, j, k).

[0022] In an embodiment of the present invention, the specific implementation of step S4 is as follows:

[0023] Let the monthly AO, NAO, and PNA be the independent variables of multiple regression, and establish multiple linear regression equations with the historical monthly wind type probabilities P ( heavy, j, k ), P ( weak, j, k ) and P ( mediocre, j, k ) as the dependent variables;

[0024] Estimate the parameters using the least squares method to obtain the respective estimation models of P(heavy, j, k), P(weak, j, k), and P(mediocre, j, k).

[0025] In an embodiment of the present invention, the specific implementation of step S5 is as follows:

[0026] Access and analyze the cross-seasonal prediction results of oceanic AO, NAO, and PNA released by authoritative meteorological agencies;

[0027] Use a multiple regression prediction model to obtain the probability characteristics of regional oceanic wind power output in the next 3 - 6 months;

[0028] Compare the wind type probability conditions of regional wind power output in each month within the next 3 - 6 months. Based on the historical wind type probability statistics in step S2, obtain the year-on-year and month-on-month conditions of strong wind days, light wind days, and moderate wind days in that month, and give the estimated result of wind power output in that month.

[0029] Compared with the prior art, the present invention has the following beneficial effects: Through the climate factor correlation analysis method, the present invention obtains the correlation model of the regional offshore wind power output characteristics between different seasons, thereby establishing a long-term estimation of the main trend of the cross-seasonal wind power generation output characteristics considering climate feature analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow block diagram of the long-term prediction extrapolation method for offshore wind power output characteristics based on convolutional neural network of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The following will specifically describe the technical solutions of the present invention with reference to the accompanying drawings.

[0032] The present invention provides a long-term prediction extrapolation method for offshore wind power output characteristics based on convolutional neural network, including the following steps:

[0033] S1. Calibrate the time stamps of each offshore wind power output data in the region, and calculate the regional total output every 15 minutes, that is, the regional wind power output;

[0034] S2. According to the time series of regional offshore wind power output, apply the decision tree classification algorithm to classify and label the wind types of the daily time series, and extract three main categories: strong wind days, light wind days, and moderate wind days, to form the time series of the daily regional wind power output wind types of each month; Based on the time series of the daily regional wind power output wind types of each month, obtain the probability of strong wind days, light wind days, and moderate wind days in each month;

[0035] S3. Extract climate index analysis variables from climate data, and select three indices, namely the Arctic Oscillation (AO), the North Atlantic Oscillation (NAO), and the Pacific-North American teleconnection pattern (PNA), which are closely related to ocean surface winds, as climate index analysis variables. All three indices are monthly average indices.

[0036] S4. Establish multiple linear regression estimation models for the monthly probability time series of strong wind days, light wind days, and calm wind days respectively with AO, NAO, and PNA.

[0037] S5. Extract the cross-seasonal time scale forecast information from the future forecasts of AO, NAO, and PNA released by authoritative meteorological research institutions, and substitute it into the multiple linear regression estimation model to obtain the cross-seasonal regional wind power generation output specific cross-level estimation results.

[0038] The following is the specific implementation process of the present invention.

[0039] As Figure 1 shown, a long-term prediction extrapolation method for the characteristics of offshore wind power output based on a convolutional neural network of the present invention is specifically implemented as follows:

[0040] (1) Processing of regional offshore wind power output data

[0041] 1) Sort out the offshore wind power output data, select all the offshore wind farm stations in the same historical period in the region to form a set, and use the output data of each wind turbine every 15 minutes obtained from the Energy Management System (EMS).

[0042] 2) Conduct quality control on the EMS output data of each wind farm station, remove the power curtailment data caused by active power control, and obtain the regional wind power output data every 15 minutes through summation calculation.

[0043] 3) The time length of the regional wind power output data to be calculated shall not be less than 3 years.

[0044] (2) Application of climate analysis data

[0045] 1) Based on the meteorological reanalysis data produced by the National Centers for Environmental Prediction (NCEP) of the United States and the National Center for Atmospheric Research (NCAR) of the United States, according to the time length of the wind power output data in the area to be calculated, extract the meteorological element data in the reanalysis data.

[0046] 2) Process the meteorological elements in the meteorological reanalysis data, and refer to the calculation formulas of the important ocean surface climate factors AO, NAO, and PNA in the authoritative literatures [1]-[3] to obtain the monthly AO, NAO, and PNA values in the corresponding period, and respectively form the historical time series of these indices.

[0047] 3) Based on the AO, NAO, and PNA prediction information of the ocean surface released by the authoritative meteorological agency (the World Meteorological Organization), the monthly prediction value time series of AO, NAO, and PNA of the ocean surface across seasons is obtained through interpolation.

[0048] (3) Decision tree classification of regional wind power output characteristics

[0049] 1) Using the look-up table method, the 15-minute data of regional wind power output is processed into time series with daily, monthly, and annual time scales, where the daily, monthly, and annual representations are respectively i, j, k . The regional wind power output data at a specific moment is denoted as P ( i, j , k ).

[0050] 2) Using the 10-meter wind speed data in the meteorological reanalysis data and adopting the ideal wind power conversion model, the ideal regional wind power output data is calculated and denoted as P C ( i, j, k ).

[0051] The ideal wind power conversion model is the model for converting wind speed into wind energy under ideal conditions, E = (ρATv 3 ) / 2, where E is the wind power, A is the cross-sectional area, T is the time, and v is the wind speed.

[0052] 3) Select the CART (ClassificationAnd Regression Tree) decision tree calculation program, and the training set consists of P ( i, j, k ) and P C ( i, j, k ).

[0053] 4) Taking a day as the time unit, classify the P ( i, j, k ) wind type characteristics, and obtain the wind type category to which any day in the analysis period belongs. The wind type categories include 3 main categories: high-wind days, low-wind days, and calm-wind days, which are respectively denoted as heavy, weak, mediocre .

[0054] 5) The daily characteristics of the regional wind power output in the processed ocean surface area are respectively denoted as P ( heavy, j, k ), P ( weak, j, k ) and P ( mediocre, j, k ).

[0055] 6) Statistically obtain the P ( heavy, j, k ) in any month, P( weak, j, k )and P ( mediocre, j, k ) probability.

[0056] (4) Analysis of the exponential correlation between regional wind power output characteristics and historical climate factors over the same period

[0057] 1) Suppose the wind types for each month in a certain period of time heavy, weak, mediocre The change curve and the month P ( heavy , j, k ), P ( weak, j, k )and P ( mediocre, j, k ) probability is a quantitative indicator to describe the wind power output characteristics of the ocean surface area.

[0058] 2) Calculate monthly AO, NAO, PNA index and P ( heavy, j, k ), P ( weak, j, k )and P ( mediocre, j, k ) and use t Test methods were used to determine the significance of the association.

[0059] 3) Determine the dependent variable combination method of the proposed multivariate linear regression equation, that is, obtain the monthly wind power output characteristics P ( heavy, j, k ), P ( weak, j, k )and P ( mediocre, j, k ) related climate factor index combination.

[0060] (5) Multivariate regression prediction model of regional wind power output characteristics based on multiple climate factor indexes

[0061] 1) Let monthly AO, NAO, and PNA be the independent variables of multiple regression, and establish the dependent variables as the probability of weather type in each month during the same period in history. P ( heavy, j, k ), P ( weak, j, k )and P ( mediocre, j, k ) is the multivariate linear regression equation.

[0062] 2) Use the least squares method to estimate the parameters and get P ( heavy, j, k ), P ( weak, j, k )and P ( mediocre, j, k ) respective estimation models.

[0063] (6)Long-term estimation of wind power output characteristics in cross-seasonal regions

[0064] 1) Access and analyze the cross-seasonal prediction results of AO, NAO, and PNA over the ocean surface released by authoritative meteorological agencies.

[0065] 2) Use the multiple regression prediction model for wind power output characteristics in the ocean surface area proposed by the present invention to obtain the probability characteristics of wind power output in the ocean surface area of the region for the next 3 - 6 months

[0066] 3) Compare the wind type probability of wind power output in each month within the next 3 - 6 months. Based on the historical wind type probability statistics in step (2), obtain the year-on-year and month-on-month conditions of the high-wind days, low-wind days, and calm-wind days in that month, and give the estimation result of wind power output in that month.

[0067] References:

[0068] [1] AO: Thompson D W J and Wallace J M. The Arctuc Oscillation signature in the wintertime geopotential height and temperature fields. Geophysical Research Letters, 1998, 25(9): 1297 - 1300.

[0069] [2] NAO: Walker, G T. EW Bliss. 1932. World weather V. Mem. Roy. Meteor. Sci., 4: 53 - 84.

[0070] [3] PNA: Wallace J M, Gutzler D S. Teleconnections in the geopotential height field during the Northern Hemisphere winter, Mon Wea Rev, 1981, 109(4): 784 - 812.

[0071] The above are the preferred embodiments of the present invention. All changes made according to the technical solution of the present invention, when the functional effects produced do not exceed the scope of the technical solution of the present invention, shall fall within the protection scope of the present invention.

Claims

1. A long-term prediction extrapolation method for the output characteristics of offshore wind power based on convolutional neural network, characterized in that, it includes the following steps: S1. Calibrate the time stamps of the output data of each offshore wind power in the region, and calculate the total regional output every 15 minutes, that is, the regional wind power output; S2. According to the time series of the offshore regional wind power output, apply the decision tree classification algorithm to classify and label the wind types of the daily time series, and extract 3 main categories of high-wind days, low-wind days, and moderate-wind days to form the time series of the daily regional wind power output wind types for each historical month; Based on the time series of the daily regional wind power output wind types for each month, obtain the probability of high-wind days, low-wind days, and moderate-wind days for each month; S3. Extract climate index analysis variables from climate data, and select 3 indexes closely related to the ocean surface wind, namely the Arctic Oscillation AO, the North Atlantic Oscillation NAO, and the Pacific-North American teleconnection PNA, as climate index analysis variables. All 3 indexes are monthly average indexes; S4. Establish multiple linear regression estimation models for the monthly high-wind day probability time series, low-wind day probability time series, and moderate-wind day probability time series with AO, NAO, and PNA respectively; S5. Extract the cross-seasonal time scale forecast information in the future forecasts of AO, NAO, and PNA released by authoritative meteorological research institutions, and substitute them into the multiple linear regression estimation model to obtain the cross-seasonal regional wind power generation output special cross-level estimation results; The specific implementation of the step S1 is as follows: Sort out the offshore wind power output data, select all offshore wind farm stations in the same historical period in the region to form a set, and use the energy management system EMS to obtain the output data of each fan every 15 minutes; For the output data of each fan obtained every 15 minutes, remove the power-limiting data caused by active power control, and obtain the regional wind power output data every 15 minutes through summation calculation; The time length of the regional wind power output data to be calculated shall not be less than 3 years; The specific implementation of the step S2 is as follows: Using the look-up table method, the 15-minute data of regional wind power output is processed into time series with daily, monthly, and annual time scales, where the daily, monthly, and annual representations are respectively i, j, k ; the regional wind power output data at a certain moment is denoted as P ( i, j, k ); Using the 10-meter wind speed data and adopting the ideal wind power conversion model, the calculated ideal regional wind power output data is denoted as P C ( i , j, k ) Select the CART decision tree calculation program. The training set consists of P ( i, j, k ) and P C ( i, j, k ). Taking days as the time unit, classify the wind type characteristics of P ( i, j, k ) to obtain the wind type category to which any day in the period to be analyzed belongs. The wind type categories include 3 main categories: high wind days, light wind days, and calm wind days, which are respectively denoted as heavy, weak, mediocre . The daily characteristics of the regional wind power output after processing are respectively denoted as P ( heavy, j, k ), P ( weak, j, k ) and P ( mediocre, j, k ). Statistically obtained for any month P ( heavy, j, k )、 P ( weak, j, k ) and P ( mediocre, j, k ) probabilities.

2. The long-term prediction extrapolation method for the output characteristics of offshore wind power based on convolutional neural network according to claim 1, characterized in that, the specific implementation of the step S3 is as follows: Set the monthly wind types within a certain period heavy, weak, mediocre of the change curve and the monthly P ( heavy, j, k ), P ( weak, j, k ), and P ( mediocre, j, k ) probability are used as quantitative indicators to describe the characteristics of regional wind power output; Calculate the monthly AO, NAO, and PNA indices and P ( heavy, j, k ), P ( weak, j, k ) and P ( mediocre, j , k ) correlation coefficients, and use t test method to determine the significance of the correlation; Obtain the monthly wind power output characteristics P ( heavy, j, k )、 P ( weak, j, k ) and P ( mediocre, j, k ) related climate factor index combinations.

3. The long-term prediction extrapolation method for the output characteristics of offshore wind power based on a convolutional neural network according to claim 2, characterized in that, the specific implementation of step S4 is as follows: Taking the monthly AO, NAO, and PNA as the independent variables of multiple regression, multiple linear regression equations are established with the probability of each monthly wind pattern in the same historical period as the dependent variable P ( heavy, j, k )、 P ( weak, j, k ) and P ( mediocre, j, k ) respectively; Estimate the parameters using the least squares method to obtain P ( heavy, j, k )、 P ( weak, j, k ) and P ( mediocre, j, k )'s respective estimation models.

4. The long-term prediction extrapolation method for the output characteristics of offshore wind power based on a convolutional neural network according to claim 1, characterized in that, the specific implementation of step S5 is as follows: Access and analyze the trans-seasonal prediction results of sea surface AO, NAO, and PNA released by authoritative meteorological agencies; Use a multiple regression prediction model to obtain the probability characteristics of regional sea surface wind power output in the next 3-6 months; Compare the wind type probability of regional wind power output in each month within the next 3-6 months, and based on the historical wind type probability statistics in step S2, obtain the year-on-year and month-on-month conditions of the high-wind days, low-wind days, and average-wind days in that month, and give the estimated result of the wind power output in that month.

Images