Wind power generation power prediction method and system

Abstract

The invention discloses a wind power generation power prediction method and system, and relates to the technical field of wind power generation. The method comprises the steps of obtaining a data source of operation of a wind turbine generator, preprocessing the data source, constructing a power prediction model, obtaining a preliminary prediction result of wind power generation power based on the power prediction model and the preprocessed data source, and evaluating and optimizing the preliminary prediction result to obtain a final prediction result. The method has high prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method and system for predicting wind power generation power. Background Art

[0002] Wind power generation is the process of converting wind energy into electrical energy. The wind turbine captures wind energy through the wind rotor. The rotation of the wind rotor drives the generator to generate electricity. Wind speed, wind direction and other meteorological conditions directly affect the power output of wind power generation.

[0003] The relationship between wind power generation and wind speed can be described by a wind power curve, where a typical wind power curve includes three key points: starting wind speed, rated wind speed and shutdown wind speed. The power generation capacity of the wind turbine is zero below the starting wind speed, reaches the maximum value near the rated wind speed, and stops generating electricity above the shutdown wind speed.

[0004] In addition, due to factors such as the uncertainty of wind speed, the complexity, timeliness and spatial scale of meteorological conditions, the power of wind power generation is uncertain, which affects the stability of subsequent wind power grid connection. Therefore, it is generally necessary to predict the wind power generation power first, and then adjust the grid voltage appropriately according to the prediction results to allow the wind power grid connection to proceed smoothly.

[0005] Traditional wind power generation is generally predicted by empirical analysis of wind turbine performance and wind speed, using simple empirical formulas for prediction, and then manually selecting the corresponding grid-connected voltage. However, on the one hand, for some areas, the wind speed and wind direction vary greatly, which affects the accuracy of wind power generation prediction. On the other hand, manually selecting the grid-connected voltage cannot achieve accurate matching between the wind power grid-connected voltage and the grid grid-connected voltage, resulting in power loss during the wind power grid-connected process.

[0006] Chinese patent (CN116227671A) discloses an optimization method for a wind power joint prediction model based on a long short-term memory neural network. By introducing a grid search method, all possibilities of parameters are quickly and effectively compared, which greatly improves the performance of the algorithm. This is the simplest and most widely used hyperparameter search algorithm. It determines the optimal value by finding all points within the search range. With a larger search range and a smaller step size, the grid search has a high probability of finding the global optimal value. However, this method only calculates the data once and cannot determine the accuracy of the calculated prediction results. Summary of the invention

[0007] In order to overcome the deficiencies in the background technology, the present invention proposes a method with high prediction accuracy.

[0008] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0009] A method for predicting wind power generation power, the specific steps are:

[0010] Obtain the data source of wind turbine operation and pre-process the data source;

[0011] Build a power prediction model;

[0012] Based on the power prediction model and the preprocessed data source, the preliminary prediction result of wind power generation is obtained;

[0013] The preliminary prediction results are evaluated and optimized to obtain the final prediction results.

[0014] Furthermore, the data sources include meteorological data, historical power generation data and geographic information data.

[0015] Furthermore, the preprocessing process of the data source includes missing value processing and data cleaning.

[0016] Furthermore, the method for constructing the power prediction model is specifically as follows:

[0017] Use LSTM model to predict short-term wind power generation;

[0018] Use the SARIMA model to predict long-term wind power generation;

[0019] The wind power generation calculated by the LSTM model and the SARIMA model is fused to obtain preliminary prediction results.

[0020] Furthermore, the data fusion is implemented by using a weighted average algorithm.

[0021] Furthermore, the evaluation and optimization steps of the preliminary prediction results are specifically as follows:

[0022] The root mean square error, mean absolute error and mean absolute percentage error were used to evaluate the preliminary forecast results;

[0023] If the evaluation result indicates that it is accurate, then the preliminary prediction result will be the final prediction result;

[0024] If the evaluation result indicates that it is inaccurate, the parameters of the power prediction model are optimized and adjusted, the wind power generation power is re-predicted, and the calculated preliminary prediction result is optimized again.

[0025] Furthermore, the parameter optimization and adjustment method of the power prediction model adopts a parameter updating method of a gradient descent algorithm.

[0026] A wind power generation power prediction system comprises a data acquisition unit, a data preprocessing unit, a prediction calculation unit, an evaluation optimization unit and a visualization unit which are sequentially connected in communication;

[0027] The data acquisition unit is used to receive the required data from the wind turbine operation database and the meteorological and geographical database;

[0028] The data preprocessing unit is used to perform preprocessing operations on the acquired data;

[0029] The prediction calculation unit writes a power prediction model, and performs calculations based on preprocessed data to obtain a preliminary prediction result of wind power;

[0030] The evaluation and optimization unit is used to evaluate the preliminary prediction results, select whether to optimize the power prediction model according to the situation, and output the final prediction results;

[0031] The visualization unit is used to receive the required data and the final prediction result, generate a wind power generation prediction report, and display it on a visualization device.

[0032] Further, the prediction calculation unit includes an LSTM model module, a SARIMA model module and a data fusion module, wherein the LSTM model module and the SARIMA model module respectively receive the preprocessing data of the preprocessing unit and calculate the corresponding wind power generation power;

[0033] The data fusion module is used to fuse the wind power generation power of the LSTM model module and the SARIMA model module to obtain a preliminary prediction result.

[0034] Furthermore, the visualization unit is provided with a user interface for allowing a user to retrieve historical prediction data.

[0035] Due to the adoption of the above technical solution, the present invention has at least the following beneficial effects:

[0036] The present invention discloses a wind power generation power prediction method and system, which can enhance the accuracy of the prediction results by evaluating and optimizing the preliminary prediction results.

[0037] In addition, by predicting and merging the short-term and long-term power generation of wind turbines, the preliminary prediction results can be made more accurate, thereby correspondingly improving the accuracy of the final prediction results;

[0038] In addition, the visualization unit in the wind power generation prediction system can display the final prediction results, which helps customers to intuitively observe the prediction results of wind power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a method of the present invention is shown in FIG. DETAILED DESCRIPTION

[0040] The technical solution of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention:

[0041] Embodiment 1 is:

[0042] Combined with Figure 1 The wind power generation power prediction method described above comprises the following specific steps:

[0043] Obtain the data source of wind turbine operation and pre-process the data source;

[0044] In particular, the data sources include meteorological data, historical power generation data, and geographic information data. The meteorological data includes wind speed v, wind direction θ, temperature T, air pressure P, and humidity H. The historical power generation data includes the historical actual power generation P at different wind speeds. 发 ,Geographic information includes the height h of the wind turbine, which are factors that affect the power generated by the wind turbine;

[0045] The meteorological data and geographic information data need to be obtained from the meteorological and geographic database, while the historical power generation data is obtained from the wind turbine operation database.

[0046] In addition, using data sources instead of single data as input data for subsequent processing can improve the diversity of data selection, thereby ensuring that the subsequent power prediction model processing is more comprehensive, which can make the processing results more accurate.

[0047] In addition, the preprocessing step can ensure that the data input into the subsequent power prediction model is of higher quality, thereby reducing the prediction bias caused by data quality issues.

[0048] Build a power prediction model;

[0049] The power prediction model can predict the power generation of wind turbines and obtain the predicted power generation data;

[0050] Based on the power prediction model and the preprocessed data source, the preliminary prediction result of wind power generation is obtained;

[0051] The preliminary prediction results are evaluated and optimized to obtain the final prediction results.

[0052] Embodiment 2 is:

[0053] On the basis of Example 1, the preprocessing step is further limited, that is, the preprocessing process of the data source includes missing value processing and data cleaning;

[0054] The missing value processing is used to supplement the missing data in the data source, and in this embodiment, linear interpolation can be used for calculation and supplementation, and the calculation process is:

[0055] Assume that in the time series data, y i and i+1 are known adjacent data points, and y k is the missing data point. The goal of linear interpolation is to i and i+1 To estimate y k , generally applicable to wind speed and temperature data;

[0056] The formula for linear interpolation is as follows:

[0057]

[0058] Where: t i and t i+1 is the timestamp; t k is the timestamp of the missing data point.

[0059] In some cases, if the relationship between data points is not linear, you can consider using polynomial interpolation. Polynomial interpolation can fill in missing values ​​by fitting a polynomial. For wind power forecasting, if there is a nonlinear relationship between data points, polynomial interpolation may provide better results, such as wind direction θ, temperature T, air pressure P, humidity H, historical actual power generation P at different wind speeds 发 and the height h of the wind turbine,

[0060] And the formula for polynomial interpolation is as follows:

[0061]

[0062] Where P(x) is a polynomial function used to represent the data set; a i are the coefficients of the polynomial, which can be solved by the least squares method or other methods; n is the degree of the polynomial; x is the data value;

[0063] That is, when there are n+1 data points, we can construct an n-degree polynomial P(x) that will be i The value is y i For example, if there are three data points, then a quadratic polynomial (n=2) can be constructed, and the polynomial will evaluate to the corresponding y values ​​at these three points.

[0064] Before data cleaning, use To detect outliers in a normally distributed data stream, for outliers in a non-normally distributed data stream, a method based on the median absolute deviation can be used for detection;

[0065] Then the outliers are removed from the data stream, which is the data cleaning process.

[0066] Embodiment 3 is:

[0067] On the basis of Example 2, the power prediction model is further limited, that is, the method for constructing the power prediction model is specifically as follows:

[0068] Use LSTM model to predict short-term wind power generation;

[0069] Use the SARIMA model to predict long-term wind power generation;

[0070] The wind power generation calculated by the LSTM model and the SARIMA model is fused to obtain preliminary prediction results.

[0071] The calculation formula of the LSTM model is:

[0072] f t =σ(W f +[h t-1 , x t ]+b f );

[0073] i t =σ(Wi+[h t-1 ,x t ]+b i );

[0074]

[0075]

[0076] o t =σ(W o +[h t-1 ,x t ]+b o );

[0077] h t =o t *tanh(C t );

[0078] Among them, f t is the forget gate; t-1 Represents the hidden state of the previous moment, which contains the information of the previous time step; X t Represents the input at the current moment, including the information of the current time step; i t is the input gate; is a candidate memory unit; C t is the unit state; o t is the output gate; σ is the activation function; tanh is the activation function; W i , W Cand W o are weight matrices, b i , b C and b o All are bias terms.

[0079] And the forget gate can help the model decide which historical information is no longer important and can be forgotten. For example, if the wind speed has not changed much in the past few days, the forget gate may decide to discard this information and only keep the recent change trend;

[0080] The input gate and candidate memory cells jointly decide which new information should be stored in the cell state. For example, new wind speed data or temperature data may be very important, so the input gate will decide to keep this information;

[0081] The cell state saves information about the past and current state, which is very important for predicting future wind power. For example, the cell state can help the model remember the wind speed change trend over a period of time in the past, so as to make more accurate predictions;

[0082] The output gate helps the model decide what information should be output at the current time step. For example, if the cell state contains important wind speed change information, the output gate will decide whether to pass this information to the next time step or as the final prediction result.

[0083] In addition, for the SARIMA model:

[0084] Assume y t is the observed value of the time series at time point t, X t is the vector containing the exogenous variables, β is the coefficient vector of the exogenous variables, p, d, q are the order of the non-seasonal part, and P, D, Q, s are the order and cycle length of the seasonal part.

[0085] Among them, the non-seasonal part:

[0086] Autoregressive part (AR): represents

[0087] Differential part (I): expressed as (1-B) d , where d is the order of the difference;

[0088] Moving average part (MA): expressed as in:

[0089]

[0090] Seasonal section:

[0091] Seasonal autoregressive part (SAR): denoted as φ(B s ),in:

[0092] φ(B)=1-φ1B s -φ2B s2 -φ3B s3 …-φ P B sP .

[0093] Seasonal difference part (SI): expressed as (1-B s ) D , where D is the order of seasonal differencing.

[0094] Seasonal Moving Average (SMA): denoted as θ(B s ),in:

[0095]

[0096] Combining the above parts, the SARIMAX model can be expressed as:

[0097]

[0098] in: is a non-seasonal autoregressive polynomial; φ(B s ) is the seasonal autoregressive polynomial; (1-B) d It is a non-seasonal difference operation; (1-B s ) D is the seasonal difference operation; is a non-seasonal moving average polynomial; θ(B s ) is the seasonal moving average polynomial; c is the constant term; β is the coefficient vector of the exogenous variable; X t is the time series of the exogenous variable; t is the error term, which is usually assumed to be a white noise series.

[0099] The backward shift operator B shifts the time series backward by one time unit.

[0100] Polynomial φ(B s ), and θ(B s ): represent the autoregressive and moving average parts respectively.

[0101] Differential operation: via (1-B) d and (1-B s ) D Difference the time series to make it stationary.

[0102] X t Includes impact t Other variables, such as wind speed, temperature, etc. The coefficient β measures the effect of these variables on yt impact.

[0103] Furthermore, suppose we have a SARIMAX(1,1,1)(1,1,1)_12 model and we have exogenous variables X t , the model can be written as:

[0104]

[0105] Here: p=1, d=1, q=1 represents the non-seasonal part, and P=1, D=1, Q=1, s=12 represents the seasonal part.

[0106] In particular, the data fusion is implemented using a weighted average algorithm, and the specific calculation formula is:

[0107]

[0108] Where ω1 and ω2 are weights, i t is the prediction power output by the LSTM model, y t is the forecast power output by the SARIMA model.

[0109] Embodiment 4 is:

[0110] Based on Example 3, the evaluation and optimization steps of the preliminary prediction results are further defined as follows:

[0111] The root mean square error, mean absolute error and mean absolute percentage error were used to evaluate the preliminary forecast results;

[0112] If the evaluation result indicates that it is accurate, then the preliminary prediction result will be the final prediction result;

[0113] If the evaluation result indicates that it is inaccurate, the parameters of the power prediction model are optimized and adjusted, the wind power generation power is re-predicted, and the calculated preliminary prediction result is optimized again.

[0114] In particular, the parameter optimization and adjustment method of the power prediction model adopts a parameter updating method of a gradient descent algorithm.

[0115] The calculation formula of the parameter update method of the gradient descent algorithm is:

[0116]

[0117] Among them, θ is the parameter vector of the model, and the parameter vector θ is the model parameter that needs to be optimized, η is the learning rate, which determines the step size of the parameter update at each iteration. is the gradient of the loss function J(θ) at the current parameter θ, that is, the partial derivative of the loss function with respect to the parameter θ.

[0118] As an example,

[0119] First, the root mean square error RMSE, mean absolute error MAE and mean absolute percentage error MAPE between the preliminary prediction results and the true value are calculated. If RMSE, MAE and MAPE are small, the preliminary prediction results are considered to be more accurate; otherwise, the optimization step is entered.

[0120] The optimization process is:

[0121] 1) Parameter adjustment: Check whether the parameters of the LSTM model, such as the number of layers, number of neurons, and learning rate, are reasonable, and appropriately increase the number of LSTM layers or the number of neurons.

[0122] 2) Feature Engineering: Check if additional features need to be added, such as lagged features or seasonal features of wind speed.

[0123] 3) Data preprocessing: Check whether the data is properly normalized and try different normalization methods.

[0124] 4) Model retraining: Retrain the LSTM model and SARIMA model using the adjusted parameters.

[0125] 5) Repeat evaluation: Re-evaluate the optimized model and calculate new RMSE, MAE, and MAPE.

[0126] 6) If the new evaluation results show that RMSE, MAE and MAPE are significantly reduced, the prediction results of the optimized model are considered to be more accurate.

[0127] 7) If you are still not satisfied, continue to adjust the model parameters, feature selection or data preprocessing methods until you get a satisfactory prediction result.

[0128] A wind power generation power prediction system comprises a data acquisition unit, a data preprocessing unit, a prediction calculation unit, an evaluation optimization unit and a visualization unit which are sequentially connected in communication;

[0129] The data acquisition unit is used to receive the required data from the wind turbine operation database and the meteorological and geographical database;

[0130] The data preprocessing unit is used to perform preprocessing operations on the acquired data;

[0131] The prediction calculation unit writes a power prediction model, and performs calculations based on preprocessed data to obtain a preliminary prediction result of wind power;

[0132] The evaluation and optimization unit is used to evaluate the preliminary prediction results, select whether to optimize the power prediction model according to the situation, and output the final prediction results;

[0133] The visualization unit is used to receive the required data and the final prediction result, generate a wind power generation prediction report, and display it on a visualization device.

[0134] In addition, the prediction calculation unit includes an LSTM model module, a SARIMA model module and a data fusion module, wherein the LSTM model module and the SARIMA model module respectively receive the preprocessing data of the preprocessing unit and calculate the corresponding wind power generation power;

[0135] The data fusion module is used to fuse the wind power generation power of the LSTM model module and the SARIMA model module to obtain a preliminary prediction result.

[0136] In particular, the visualization unit is provided with a user interface for allowing the user to retrieve historical prediction data, wherein the user interface can be used to connect to an external account, so that in addition to the local account, the external account can also query the wind power prediction results obtained in the system.

[0137] The parts of the present invention that are not described in detail are prior art. It is obvious to those skilled in the art that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention; therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is limited by the appended claims rather than the above description. Therefore, it is intended to include all changes within the meaning and scope of the equivalent elements of the claims in the present invention, and any figure marks in the claims should not be regarded as limiting the claims involved.

Claims

1. A method for predicting wind power generation, characterized in that: include Obtain the data source of wind turbine operation and pre-process the data source; Build a power prediction model; Based on the power prediction model and the preprocessed data source, the preliminary prediction result of wind power generation is obtained; The preliminary prediction results are evaluated and optimized to obtain the final prediction results.

2. The method for predicting wind power generation according to claim 1, wherein: The data sources include meteorological data, historical power generation data and geographic information data.

3. The method for predicting wind power generation according to claim 2, wherein: The preprocessing process of the data source includes missing value processing and data cleaning.

4. The method for predicting wind power generation according to claim 3, characterized in that: The method for constructing the power prediction model is specifically as follows: Use LSTM model to predict short-term wind power generation; Use the SARIMA model to predict long-term wind power generation; The wind power generation calculated by the LSTM model and the SARIMA model is fused to obtain preliminary prediction results.

5. The method for predicting wind power generation according to claim 4, characterized in that: The data fusion is implemented by using a weighted average algorithm.

6. The method for predicting wind power generation according to claim 5, characterized in that: The evaluation and optimization steps of the preliminary prediction results are specifically as follows: The root mean square error, mean absolute error and mean absolute percentage error were used to evaluate the preliminary forecast results; If the evaluation result indicates that it is accurate, then the preliminary prediction result will be the final prediction result; If the evaluation result indicates that it is inaccurate, the parameters of the power prediction model are optimized and adjusted, the wind power generation power is re-predicted, and the calculated preliminary prediction result is optimized again.

7. The method for predicting wind power generation according to claim 6, wherein: The parameter optimization and adjustment method of the power prediction model adopts a parameter updating method of a gradient descent algorithm.

8. A wind power generation power prediction system, comprising a data acquisition unit, a data preprocessing unit, a prediction calculation unit, an evaluation optimization unit and a visualization unit which are sequentially connected in communication; The data acquisition unit is used to receive the required data from the wind turbine operation database and the meteorological and geographical database; The data preprocessing unit is used to perform preprocessing operations on the acquired data; The prediction calculation unit writes a power prediction model, and performs calculations based on preprocessed data to obtain a preliminary prediction result of wind power; The evaluation and optimization unit is used to evaluate the preliminary prediction results, select whether to optimize the power prediction model according to the situation, and output the final prediction results; The visualization unit is used to receive the required data and the final prediction result, generate a wind power generation prediction report, and display it on a visualization device.

9. The wind power generation prediction system according to claim 8, characterized in that: The prediction calculation unit includes an LSTM model module, a SARIMA model module and a data fusion module, wherein the LSTM model module and the SARIMA model module respectively receive the preprocessing data of the preprocessing unit and calculate the corresponding wind power generation; The data fusion module is used to fuse the wind power generation power of the LSTM model module and the SARIMA model module to obtain a preliminary prediction result.

10. The wind power generation power prediction system according to claim 8, characterized in that: The visualization unit is provided with a user interface for allowing a user to retrieve historical forecast data.

Images