Wind turbine power prediction method and system based on multi-interval feature screening and machine learning

Abstract

The present invention provides a wind power prediction method and system based on multi-interval feature screening and machine learning, comprising: selecting measurement points and obtaining historical wind turbine operation data at each measurement point; processing outliers in the historical data; generating feature data from the processed data, and performing feature screening in different power intervals; constructing a wind power prediction model, inputting several different sets of filtered feature data into the model for training, and obtaining multiple models; commencing prediction, obtaining real-time measurement point data from a database, determining intervals based on wind power, processing outliers on the real-time data, and directly generating features for the interval; inputting the features into the corresponding model to predict wind power generation. The present invention utilizes machine learning technology in a wind power generation system to construct a model to predict wind power generation, thereby effectively preserving the complex nonlinear relationships between parameters and predicting wind turbine power generation, so as to adjust the power supply structure in advance.

Description

Technical Field

[0001] The present invention relates to the field of integration of computers and wind power generation services, and more specifically, to a wind power prediction method and system based on multi-interval feature screening and machine learning. Background Art

[0002] With the increasing demand for energy conservation and emission reduction and the decreasing cost of renewable energy generation, the global wind power and other new energy power generation industries are developing rapidly. In 2021, China's cumulative installed wind power capacity reached 328.5GW, accounting for 13.8% of the country's total installed power generation capacity, second only to hydropower and thermal power.

[0003] As one of the most reliable and viable renewable energy sources, wind power continues to receive strong government support in terms of scale, technology, and subsidies. However, due to environmental factors, wind power generation is difficult to sustain and is highly variable. This variability poses a significant challenge to integrating wind power into power grids. To maintain a balance between power generation and consumption, fluctuations in wind power generation require rapid replacement with other power sources. Therefore, to ensure grid stability and power supply security, highly accurate power generation forecasts are essential. Summary of the Invention

[0004] In order to address the deficiencies in the prior art, the present invention provides a wind power prediction method and system based on multi-interval feature screening and machine learning to improve the accuracy of power prediction.

[0005] The present invention adopts the following technical solutions.

[0006] A wind power prediction method based on multi-interval feature screening and machine learning includes the following steps:

[0007] Step 1: Select relevant measuring points of the fan and obtain historical data of the fan operation at each measuring point;

[0008] Step 2: Detect and process abnormal data on historical data to obtain processed data;

[0009] Step 3: Generate and filter features on the processed data, divide the power intervals, and obtain high-correlation feature data for each power interval;

[0010] Step 4: Build a wind power prediction model and train the model based on highly correlated feature data, and set an update cycle to update the trained model;

[0011] Step 5: Acquire the real-time data of the wind turbine operation, and perform abnormal data detection and processing on the real-time data to obtain the power range and high-correlation feature data of the real-time data.

[0012] Preferably, the relevant measurement points of the wind turbine in step 1 include: wind speed, wind direction, ambient temperature, cabin temperature, pitch angle of the wind turbine blades, and wind power.

[0013] Preferably, step 2 further comprises: using a mutation point detection algorithm to detect mutation values ​​in historical data, and replacing mutation values ​​with an average value of adjacent wind turbine data as a replacement value.

[0014] Preferably, step 3 comprises the following steps:

[0015] Step 3.1, generate features for the processed data, including time features and statistical features;

[0016] In step 3.2, the wind power values ​​are sorted and divided into power intervals using a multi-interval feature screening method, the correlation between the features and the power in each power interval is calculated, and the high-correlation feature data in each power interval is screened out.

[0017] Preferably, the step 3.2 further includes:

[0018] Sort the wind power values ​​and divide them into N power intervals according to their sorting;

[0019] The correlation between the generated features and the power is calculated using the feature correlation analysis method in each interval;

[0020] A screening threshold is set for each interval to filter out features with high correlation and obtain high correlation feature data for each interval.

[0021] Preferably, the step 4 comprises the following steps:

[0022] Step 4.1: Build a wind power prediction model based on the GBDT algorithm;

[0023] Step 4.2: training the wind power prediction model in each power range based on the data obtained after screening in step 3, and obtaining the trained wind power prediction model corresponding to each power range;

[0024] Step 4.3, set the update period T1. After time T1, obtain new historical data from the database, retrain the model, and update the model.

[0025] Preferably, the step 4.2 further includes:

[0026] The high-correlation characteristic data of each power interval are used as the input X of the wind power prediction model, and the wind power after T time is used as the output value y of the wind power prediction model. The GBDT model is trained in each power interval and the trained wind power prediction model is saved.

[0027] Preferably, the step 5 further includes:

[0028] Obtain real-time data on wind turbine operation from the database, including wind speed, wind direction, ambient temperature, cabin temperature, pitch angles of wind turbine blades, and wind power;

[0029] Detect and process abnormal data in real-time data; this includes using a mutation point detection algorithm, combining mutation values ​​detected in historical data, setting a normal value range, detecting mutation values ​​in real-time data, and using the average value of adjacent wind turbine data as a replacement value for mutation values;

[0030] Combined with the power intervals divided in step 3, the real-time data is divided into corresponding power intervals according to the current wind power, and the features selected in step 3 are used to directly generate high-correlation features.

[0031] The present invention also provides a wind power prediction system based on multi-interval feature screening and machine learning, comprising: a data acquisition module, an abnormal data detection and processing module, a feature generation module, a feature screening module, a model training module, a model prediction module and a display module;

[0032] The data acquisition unit is used to select relevant measuring points based on experience and collect fan operation data;

[0033] The data preprocessing unit is used to preprocess the fan operation data to obtain preprocessed data;

[0034] The feature data unit is used to construct a feature calculation module, and the preprocessed data is input into the feature calculation module to obtain feature data;

[0035] The model building unit is used to build a wind power prediction GBDT model;

[0036] The model training unit is used to input the characteristic data into the wind power prediction model for training to obtain a trained wind power prediction model;

[0037] The model prediction unit is used to predict the wind turbine power generation using the trained wind power prediction model;

[0038] The display module is used to receive the wind turbine power prediction value obtained by the model prediction module and display it for user reference

[0039] The present invention also provides a terminal, comprising a processor and a storage medium;

[0040] The storage medium is used to store instructions;

[0041] The processor is configured to operate according to the instructions to execute the steps of the wind power prediction method based on multi-interval feature screening and machine learning.

[0042] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the wind power prediction method based on multi-interval feature screening and machine learning.

[0043] The beneficial effect of the present invention is that, compared with the prior art, the present invention combines abnormal data detection and processing technology in the wind power prediction system, and considers that the characteristics of different power intervals may be different. It uses the correlation calculation method to perform multi-interval feature screening, retains the characteristics of each interval, and uses the GBDT machine learning algorithm to construct a wind power prediction model to predict the wind power in different power intervals. In different power intervals, different models are trained with the characteristics of the corresponding intervals. Different models are called according to different powers during prediction, thereby improving the power prediction accuracy, effectively learning the complex nonlinear relationship between parameters, and predicting the wind turbine power generation so as to adjust the power supply structure in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 1 is a flow chart of a wind power prediction method based on multi-interval feature screening and machine learning in the present invention;

[0045] Figure 2 Schematic diagram of the feature generation and screening method of the present invention;

[0046] Figure 3 It is a structural diagram of the stroke power prediction model of the present invention;

[0047] Figure 4 It is a structural diagram of the wind power prediction system based on multi-interval feature screening and machine learning in the present invention;

[0048] Figure 5 This is a graph showing the results of the stroke power prediction in the present invention; DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without making creative efforts are all within the scope of protection of the present invention.

[0050] like Figure 1 As shown in the flowchart of the wind power prediction method based on multi-interval feature screening and machine learning, this embodiment provides a wind power prediction method based on multi-interval feature screening and machine learning. Specifically, it includes the following steps:

[0051] Step 1: Select relevant measuring points of the fan and obtain historical data of the fan operation at each measuring point.

[0052] According to the experience of professional workers, relevant measuring points of the fan are selected, and the historical operation data of the fan at each measuring point is obtained in batches from the database.

[0053] Preferably, relevant measurement points of the wind turbine include wind speed, wind direction, ambient temperature, cabin temperature, pitch angle of the wind turbine blades, and wind power.

[0054] Furthermore, a conventional wind turbine has three blades, so it is necessary to obtain the pitch angles of each of the three blades.

[0055] Step 2: Detect and process abnormal data on historical data to obtain processed data.

[0056] The mutation point detection algorithm is used to set the range of normal values ​​and detect mutation values ​​in historical data. For mutation values, the average value of adjacent wind turbine data is used as the replacement value.

[0057] Step 3: Generate and filter features on the processed data, divide the power intervals, and filter to obtain high-correlation feature data for each power interval.

[0058] Figure 2 Schematic diagram of the method for step 3.

[0059] like Figure 3 As shown, step 3 specifically includes the following steps:

[0060] Step 3.1, generate features for the data processed in step 2, and the generated features include time features and statistical features;

[0061] Among them, time features include year, quarter, month, week, etc.; statistical features include maximum value, mean, median, peak value, variance, etc.

[0062] Step 3.2: Use a multi-interval feature screening method to sort the wind power values ​​and divide them into power intervals, calculate the correlation between the features and power in each power interval, and screen out high-correlation feature data in each power interval;

[0063] Considering that the characteristics that affect power vary across different power intervals, a multi-interval feature screening method is adopted. First, the wind power values ​​are sorted and divided into N power intervals according to their ranking. For example, when N=3, the power is divided into three intervals: high, medium, and low. Then, within each interval, the correlation between the generated characteristics and power is calculated using the feature correlation analysis method. Finally, a screening threshold is set for each interval to filter out highly correlated features, thereby obtaining highly correlated feature data for each interval.

[0064] Step 4: Build a wind power prediction model and train it based on the highly correlated feature data, and set an update cycle to update the trained model;

[0065] Step 4.1: Build a wind power prediction model based on the GBDT algorithm;

[0066] Step 4.2: training the wind power prediction model in each power range based on the data obtained after screening in step 3, and obtaining the trained wind power prediction model corresponding to each power range;

[0067] Specifically, the high-feature data of each power interval after screening is used as the input X of the wind power prediction model, and the wind power after T time is used as the output value y of the wind power prediction model. The wind power prediction model is trained and saved in each power interval to obtain the corresponding trained wind power prediction model for each power interval;

[0068] The time T can be set to 3s, 6s or 9s.

[0069] Step 4.3: Set the update period T1. After time T1, obtain new historical data from the database, retrain the model, and update the model.

[0070] like Figure 3 As shown, Figure 3 The wind power prediction model uses a boosting tree structure. Specifically, the operation and training of the wind power prediction model also includes:

[0071] Input the wind turbine operating data at time t0 into the wind power prediction model;

[0072] The wind power at time t0+T is calculated using the first regression tree in the wind power prediction model. The predicted result pred is compared with the predetermined threshold a. If the predicted result pred reaches the predetermined threshold a, the predicted result is output. If the predicted result does not reach the predetermined threshold a, the next step is continued.

[0073] Construct a second regression tree in the gradient direction of residual reduction, train the second tree regression tree and calculate the residual b1 between the predicted result pred and the true value. If the predicted result pred+b1 meets the predetermined threshold a, output pred+b1. If pred+b1 still does not meet the predetermined threshold a, proceed to the next step.

[0074] Construct a third regression tree in the gradient direction of residual reduction, train the third regression tree to calculate the residual b2 between the predicted result pred and the true value, if the predicted result pred+b2 meets the predetermined threshold a, then output pred+b2, if the predicted result pred+b2 does not meet the predetermined threshold a, then continue to construct the fourth regression tree in the gradient direction of residual reduction;

[0075] And so on, until the threshold is met, the output is the sum of the predicted result pred and the residual of each regression tree. The training result of the last tree is based on the training of all the previous trees, so the last tree will be used for calculation.

[0076] Preferably, the update period T1 can be set to 1 minute or 2 minutes.

[0077] Step 5: Acquire real-time data of wind turbine operation, perform abnormal data detection and processing on the real-time data, and obtain the power range and high-correlation feature data of the real-time data;

[0078] Specifically, step 5 further includes:

[0079] The real-time data of the wind turbine operation is obtained from the database, including wind speed, wind direction, ambient temperature, cabin temperature, pitch angles of each wind turbine blade, and wind power.

[0080] Detect and process abnormal data in real-time data. Using a mutation point detection algorithm, combined with mutation values ​​detected in historical data, a normal range is set to detect mutation values ​​in real-time data. For mutation values, the average value of adjacent wind turbine data is used as a surrogate value.

[0081] Generate highly relevant features for real-time data. First, combine the power intervals divided in step 3 and divide the real-time data into corresponding power intervals according to the current wind power, such as low, medium, or high power intervals. Then, use the features selected in step 3 to directly generate highly relevant features without generating all features.

[0082] Step 6: Combined with the power range of the real-time data, the highly correlated feature data obtained in step 5 is input into the trained wind power prediction model of the corresponding power range to predict the wind turbine power generation.

[0083] Specifically, the high correlation features generated in step 5 are input into the trained wind power prediction model of the corresponding power interval obtained in step 4 to predict the wind turbine power generation after T time, and the predicted value is displayed through the display module, and the actual value of the wind turbine power generation is displayed after T time.

[0084] like Figure 4 As shown, this embodiment also provides a wind power prediction system based on multi-interval feature screening and machine learning. Figure 4 This is a structural diagram of a wind power prediction system based on multi-interval feature screening and machine learning. The system includes a data acquisition module, an abnormal data detection and processing module, a feature generation module, a feature screening module, a model training module, a model prediction module, and a display module.

[0085] Among them, the data acquisition module is used to obtain historical data and real-time data of wind turbine operation from the Vestore database;

[0086] The abnormal data detection and processing module is used to detect and process mutation values ​​of acquired historical data and real-time data to obtain processed data;

[0087] The feature generation module is used to generate features for the processed data. For historical data, it generates all time features and statistical feature data V0; for real-time data, it only generates high-correlation feature data for each power interval.

[0088] The feature screening module is used to select highly correlated features for each power interval from all feature data V0. First, the wind power is sorted and divided into different power intervals, such as low, medium, and high. Then, the correlation between V0 and wind power is calculated for each power interval, and thresholds are set to obtain highly correlated features for each interval.

[0089] The model training module is used to input the high-correlation characteristic data of each power interval into the wind power prediction model for training, and obtain the trained wind power prediction model corresponding to each power interval;

[0090] This module is also used to update the model. After time T1, new historical data is input and the model is retrained.

[0091] The model prediction module is used to predict the wind turbine power generation using the wind power prediction model trained for each power interval. The real-time data is judged according to the power size to determine the interval in which the processed real-time data is located. The wind power model of the corresponding interval is then called, and the generated high-correlation feature data is input for prediction to obtain the predicted value of the wind turbine power generation.

[0092] The display module is used to receive the wind turbine power generation prediction value obtained by the model prediction module and display it for user reference.

[0093] The system can implement wind power prediction by executing the above-mentioned steps 1 to 6 through the corresponding models.

[0094] In order to verify the effect of the present invention, the wind power of a certain wind turbine is predicted based on the above method, and the results are as follows Figure 5 As shown, Figure 5 The figure shows the comparison between the actual value and the predicted value of the running results, where the solid line is the actual value and the dotted line is the predicted value.

[0095] The beneficial effect of the present invention is that, compared with the existing technology, the present invention combines abnormal data detection and processing technology in the wind power prediction system, and considers that the characteristics of different power intervals may be different, uses the correlation calculation method to perform multi-interval feature screening, and uses the GBDT machine learning algorithm to construct a wind power prediction model to predict the wind power in different power intervals, thereby improving the power prediction accuracy, effectively learning the complex nonlinear relationship between parameters, and predicting the wind turbine power generation so as to adjust the power supply structure in advance.

[0096] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0097] Computer-readable storage media can be a tangible device that can hold and store the instructions used by the instruction execution device. Computer-readable storage media can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. Computer-readable storage media used herein is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagated by waveguides or other transmission media (for example, light pulses by fiber optic cables), or electrical signals transmitted by wires.

[0098] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0099] The computer program instructions for performing the disclosed operation can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. Computer-readable program instructions can be executed completely on a user's computer, partially on a user's computer, executed as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (such as, for example, utilizing an Internet service provider to connect via the Internet). In certain embodiments, by utilizing the state information of computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A wind turbine power prediction method based on multi-interval feature screening and machine learning, characterized in that: The following steps are involved: Step 1: Select relevant measuring points of the fan and obtain historical data of the fan operation at each measuring point; Step 2: Detect and process abnormal data on historical data to obtain processed data; Step 3: Generate and filter features on the processed data, divide the power intervals, and obtain high-correlation feature data for each power interval; Step 4: Build a wind turbine power generation prediction model and train the model based on highly correlated feature data, and set an update cycle to update the trained model; Training the model includes: training the wind turbine power generation prediction model in each power range to obtain a trained wind turbine power generation prediction model corresponding to each power range; Step 5: Acquire the real-time data of the wind turbine operation, and perform abnormal data detection and processing on the real-time data to obtain the power range and high-correlation feature data of the real-time data.

2. The wind turbine power generation prediction method based on multi-interval feature screening and machine learning according to claim 1 is characterized in that: The relevant measurement points of the wind turbine in step 1 include: wind speed, wind direction, ambient temperature, cabin temperature, pitch angle of the wind turbine blades, and wind turbine power generation.

3. The wind turbine power generation prediction method based on multi-interval feature screening and machine learning according to claim 1 is characterized in that: The step 2 further includes: using a mutation point detection algorithm to detect mutation values ​​in historical data, and replacing the mutation values ​​with an average value of adjacent wind turbine data as a replacement value.

4. The wind turbine power generation prediction method based on multi-interval feature screening and machine learning according to claim 1 is characterized in that: The step 3 comprises the following steps: Step 3.1, generate features for the processed data, including time features and statistical features; In step 3.2, the multi-interval feature screening method is used to sort the wind turbine power values ​​and divide the power intervals, calculate the correlation between the features and power in each power interval, and screen out the high-correlation feature data in each power interval.

5. The method for predicting wind turbine power generation based on multi-interval feature screening and machine learning according to claim 4, characterized in that: The step 3.2 further includes: Sort the wind turbine power values ​​and divide them into N power intervals according to the sorting; The correlation between the generated features and the power is calculated using the feature correlation analysis method in each interval; A screening threshold is set for each interval to filter out features with high correlation and obtain high correlation feature data for each interval.

6. The method for predicting wind turbine power generation based on multi-interval feature screening and machine learning according to claim 1, characterized in that: The step 4 comprises the following steps: Step 4.1: Build a wind turbine power prediction model based on the GBDT algorithm; Step 4.2: training the wind turbine power prediction model in each power range based on the data obtained after screening in step 3, and obtaining the trained wind turbine power prediction model corresponding to each power range; Step 4.3, set the update period T1. After time T1, obtain new historical data from the database, retrain the model, and update the model.

7. The method for predicting wind turbine power generation based on multi-interval feature screening and machine learning according to claim 6, characterized in that: The step 4.2 further includes: The high-correlation feature data of each power interval is used as the input X of the wind turbine power generation prediction model, and the wind turbine power generation after T time is used as the output value y of the wind turbine power generation prediction model. The GBDT model is trained in each power interval and the trained wind turbine power generation prediction model is saved.

8. The method for predicting wind turbine power generation based on multi-interval feature screening and machine learning according to claim 6, characterized in that: The step 5 further comprises: Obtain real-time data on wind turbine operation from the database, including wind speed, wind direction, ambient temperature, cabin temperature, pitch angles of wind turbine blades, and wind turbine power generation; Detect and process abnormal data in real-time data; this includes using a mutation point detection algorithm, combining mutation values ​​detected in historical data, setting a normal value range, detecting mutation values ​​in real-time data, and using the average value of adjacent wind turbine data as a replacement value for mutation values; Combined with the power intervals divided in step 3, the real-time data is divided into corresponding power intervals according to the current wind turbine power generation. The features selected in step 3 are used to directly generate high-correlation features.

9. A wind turbine power generation prediction system based on multi-interval feature screening and machine learning using the wind turbine power generation prediction method based on multi-interval feature screening and machine learning according to any one of claims 1 to 8, characterized in that: include: Data acquisition module, abnormal data detection and processing module, feature generation module, feature screening module, model training module, model prediction module and display module; The data acquisition unit is used to select relevant measuring points based on experience and collect fan operation data; The data preprocessing unit is used to preprocess the fan operation data to obtain preprocessed data; The feature data unit is used to construct a feature calculation module, and the preprocessed data is input into the feature calculation module to obtain feature data; The model building unit is used to build a GBDT model for wind turbine power generation prediction; The model training unit is used to input the characteristic data into the wind turbine power generation prediction model for training, thereby obtaining a trained wind turbine power generation prediction model; The model prediction unit is used to predict the wind turbine power generation power using the trained wind turbine power generation prediction model; The display module is used to receive the wind turbine power generation prediction value obtained by the model prediction module and display it for user reference.

10. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images