Blade clearance value prediction method and device based on neural network model

Abstract

The embodiment of the invention provides a blade clearance value prediction method and device based on a neural network model. The method comprises the following steps: acquiring fan operation data of each fan, and preprocessing the fan operation data to obtain a target operation data set; performing model training on the plurality of preset neural network models based on the target operation data set to obtain target neural network models corresponding to the plurality of preset neural network models; combining the plurality of target neural network models to obtain an optimal combined neural network model; obtaining fan operation data with a to-be-predicted clearance value; and inputting the fan operation data of which the clearance value is to be predicted into the optimal combined neural network model to obtain a corresponding target clearance value prediction result. In this way, the optimal combined neural network model with the high blade clearance value prediction accuracy can be obtained through model training and combination, and therefore the blade clearance value is accurately predicted through the model, and the tower sweeping risk is reduced.

Description

Technical Field

[0001] The present disclosure relates to the field of wind power, and in particular to the field of blade clearance value prediction technology. Background Art

[0002] Currently, the scale of wind turbines is getting larger and larger, and the enlargement of wind turbines has brought about changes in long blades and high towers, which has led to frequent problems of blade tower sweeping. Therefore, blade clearance value monitoring is currently the most effective and direct method to prevent blade tower sweeping. However, due to cost control, not all wind turbines are equipped with clearance devices, and at the same time, the clearance devices are restricted by weather conditions such as fog and thunderstorms. Therefore, even if the clearance devices are installed, there will be frequent error reports of the clearance value. Therefore, it is urgent to accurately predict the blade clearance value of wind turbines to reduce the risk of tower sweeping, ensure the safety of the unit, and reduce power generation losses. Summary of the Invention

[0003] The present disclosure provides a method, device, equipment, and storage medium for predicting blade clearance value based on a neural network model.

[0004] According to the first aspect of the present disclosure, a method for predicting blade clearance value based on a neural network model is provided. The method includes: Obtaining the wind turbine operation data of each wind turbine; Preprocessing the wind turbine operation data of each wind turbine to obtain a target operation data set; Based on the target operation data set, training multiple preset neural network models to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively; Combining multiple target neural network models to obtain an optimal combined neural network model; Obtaining the wind turbine operation data for which the clearance value is to be predicted; Inputting the wind turbine operation data for which the clearance value is to be predicted into the optimal combined neural network model to obtain a corresponding target clearance value prediction result.

[0005] In the above aspect and any possible implementation manner, a further implementation manner is provided. The wind turbine operation data of each wind turbine includes the measured blade clearance value of each wind turbine at each moment and a plurality of first clearance-related parameters related to the measured blade clearance value of each wind turbine. The plurality of first clearance-related parameters include: the nacelle position, generator speed, active power, pitch angle, clearance state and clearance heartbeat, original wind direction, and original wind speed of each wind turbine; Each group of operation data sets in the target operation data set includes: generator speed, active power, pitch angle, measured blade clearance value, actual wind direction, and nacelle-facing wind speed; Preprocessing the fan operation data of each fan to obtain a target operation data set, including: Removing invalid data from the fan operation data of each fan; Calculating the actual wind direction of each fan according to the nacelle position and the original wind direction of each fan; Calculating the nacelle-facing wind speed of each fan according to the original wind direction and the original wind speed of each fan.

[0006] In the above aspect and any possible implementation manner, a further implementation manner is provided. The removing of invalid data from the fan operation data of each fan includes: Removing data with a generator speed lower than a preset minimum speed from the fan operation data of each fan; Removing data with abnormal clearance state and clearance heartbeat from the fan operation data of each fan; Removing data with unchanged measured blade clearance value and original wind speed from the fan operation data of each fan; Removing data with a change amount of the measured blade clearance value greater than a clearance value threshold and a change amount of the original wind speed greater than a wind speed threshold from the fan operation data of each fan.

[0007] In the above aspect and any possible implementation manner, a further implementation manner is provided. The training of multiple preset neural network models based on the target operation data set to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively includes: Grouping the target operation data set according to preset grouping parameters; wherein, the preset grouping parameters include fan model parameters, blade parameters, hub height parameters, topographic features of the wind farm location, and the distance between fans; Grouping the grouped target operation data set again at a preset time interval to obtain several groups of operation data sets; Training multiple preset neural network models respectively based on the several groups of operation data sets to obtain the target neural network models corresponding to the multiple preset neural network models respectively.

[0008] In the above aspect and any possible implementation manner, a further implementation manner is provided. Each group of operation data sets in the several groups of operation data sets includes: measured blade clearance value and multiple second clearance correlation parameters related to the measured blade clearance value; the second clearance correlation parameters include: generator speed, active power, pitch angle, and nacelle-facing wind speed; Based on the several groups of operation data sets, training multiple preset neural network models respectively to obtain the target neural network models corresponding to the multiple preset neural network models, including: Input the multiple second clearance correlation parameters of each group of operation data sets in the several groups of operation data sets into the multiple preset neural network models to obtain the clearance value prediction results corresponding to each group of operation data sets output by the multiple preset neural network models; Calculate the clearance value difference between the clearance value prediction results corresponding to each group of operation data sets output by each preset neural network model and the measured blade clearance value in each group of operation data sets; Calculate the average value of the clearance value differences of each group of operation data sets corresponding to each preset neural network model; Judge whether the average value of the clearance value differences of each group of operation data sets corresponding to each preset neural network model is lower than the preset clearance value threshold; Adjust the model parameters of the preset neural network models whose average value of the clearance value differences is not lower than the preset clearance value threshold among the multiple preset neural network models until the average value of the clearance value differences of the several groups of operation data sets corresponding to each preset neural network model is lower than the preset clearance value threshold, and then stop the adjustment to obtain the target neural network models corresponding to the multiple preset neural network models respectively.

[0009] In the above aspects and any possible implementation manners, a further implementation manner is provided. The combining the multiple target neural network models to obtain the optimal combined neural network model includes: Obtain multiple groups of fan test data sets; each group of fan test data sets includes: the measured blade clearance value of a test fan and multiple third clearance correlation parameters related to the measured blade clearance value of the test fan; the multiple third clearance correlation parameters include: the generator speed, active power, pitch angle and nacelle orientation wind speed of the test fan; Input the multiple third clearance correlation parameters in each group of fan test data sets into the objective function with preset weight coefficients to be optimized, to obtain the root mean square error between the predicted blade clearance value and the measured blade clearance value corresponding to the multiple groups of fan test data sets under each candidate value of the preset weight coefficients, where the objective function is composed of multiple target neural network models and the preset weight coefficients, and the predicted blade clearance value corresponding to the multiple groups of fan test data sets is output by the multiple target neural network models respectively after each group of fan test data sets is input into the objective function; The root mean square error between the predicted blade clearance values corresponding to the multiple groups of fan test data sets and the measured blade clearance values at each candidate value of the preset weight coefficient output according to the objective function is obtained to get the optimal combined neural network model.

[0010] In the above aspects and any possible implementation manners, a further implementation manner is provided. The method of obtaining the optimal combined neural network model according to the root mean square error between the predicted blade clearance values corresponding to the multiple groups of fan test data sets and the measured blade clearance values at each candidate value of the preset weight coefficient output according to the objective function includes: Taking the minimum root mean square error between the predicted blade clearance values and the measured blade clearance values at each candidate value as the objective, select the optimal candidate value from multiple candidate values of the preset weight coefficient; Combine the multiple target neural network models according to the optimal candidate value of the preset weight coefficient to obtain the optimal combined neural network model.

[0011] According to the second aspect of the present disclosure, a device for predicting blade clearance value based on a neural network model is provided. The device includes: A first acquisition module, configured to acquire the fan operation data of each fan; A processing module, configured to preprocess the fan operation data of each fan to obtain a target operation data set; A training module, configured to perform model training on multiple preset neural network models based on the target operation data set to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively; A combination module, configured to combine multiple target neural network models to obtain an optimal combined neural network model; A second acquisition module, configured to acquire the fan operation data with the clearance value to be predicted; A prediction module, configured to input the fan operation data with the clearance value to be predicted into the optimal combined neural network model to obtain a corresponding target clearance value prediction result.

[0012] According to the third aspect of the present disclosure, an electronic device is provided. The electronic device includes: a memory and a processor, and a computer program is stored on the memory. When the processor executes the program, the method as described above is implemented.

[0013] According to the fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method according to the first aspect of the present disclosure is implemented.

[0014] In the present disclosure, after obtaining the fan operation data of each fan, the fan operation data of each fan can be preprocessed to obtain a target operation data set with high data integrity and standardization. Then, the target operation data set is used to train multiple preset neural network models to obtain the target neural network models corresponding to the respective preset neural network models. Furthermore, the target neural network models are combined to obtain an optimal combined neural network model with higher accuracy. Then, the fan operation data with the net clearance value to be predicted is input into the optimal combined neural network model, and a target net clearance value prediction result with higher accuracy can be obtained. Thus, the optimal combined neural network model is used to accurately predict the blade net clearance value of the fan without a net clearance device, and the blade net clearance value of the fan with a net clearance device can also be corrected. In this way, the risk of tower sweeping can be reduced, the safety of the unit can be ensured, and the power generation loss can be reduced.

[0015] It should be understood that the content described in the summary of the invention section is not intended to limit the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In combination with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present disclosure will become more apparent. The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. In the drawings, the same or similar reference numerals represent the same or similar elements, where: Figure 1 shows a flowchart of a method for predicting the blade net clearance value based on a neural network model according to an embodiment of the present disclosure; Figure 2 shows a training schematic diagram of a BP neural network according to an embodiment of the present disclosure; Figure 3 shows a training schematic diagram of an LSTM neural network according to an embodiment of the present disclosure; Figure 4 shows a block diagram of a device for predicting the blade net clearance value based on a neural network model according to an embodiment of the present disclosure; Figure 5 shows a block diagram of an exemplary electronic device capable of implementing the embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts shall fall within the scope of protection of the present disclosure.

[0018] In addition, the term "and / or" in this article is merely a relational description of associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0019] Figure 1 The flowchart of a blade clearance value prediction method 100 based on a neural network model according to an embodiment of the present disclosure is shown. The method 100 includes: Step 110: Obtain the fan operation data of each fan; The fan operation data includes the measured blade clearance value of each fan at each moment and a plurality of first clearance associated parameters related to the measured blade clearance value. The plurality of first clearance associated parameters include: the nacelle position, generator speed, active power, pitch angle, clearance state and clearance heartbeat, original wind direction, and original wind speed of each fan.

[0020] The above fan operation data belongs to historical operation data and is the operation data collected by fans equipped with clearance devices.

[0021] Step 120: Preprocess the fan operation data of each fan to obtain a target operation data set; Step 130: Based on the target operation data set, perform model training on a plurality of preset neural network models to obtain the trained target neural network models corresponding to the plurality of preset neural network models respectively; Step 140: Combine a plurality of the target neural network models to obtain an optimal combined neural network model; Step 150: Obtain the fan operation data for which the clearance value is to be predicted; Step 160: Input the fan operation data for which the clearance value is to be predicted into the optimal combined neural network model to obtain a corresponding target clearance value prediction result.

[0022] After obtaining the fan operation data of each fan, the fan operation data of each fan can be preprocessed to obtain a target operation data set with high data integrity and standardization. Then, the target operation data set is used to train multiple preset neural network models to obtain the target neural network models corresponding to the respective preset neural network models. Furthermore, the target neural network models are combined to obtain an optimal combined neural network model with higher accuracy. Then, the fan operation data with the clearance value to be predicted is input into the optimal combined neural network model, and a target clearance value prediction result with higher accuracy can be obtained. Thus, the optimal combined neural network model is used to accurately predict the blade clearance value of the fan without a clearance device installed, and the blade clearance value of the fan with a clearance device installed can also be corrected. In this way, the risk of tower sweeping can be reduced, the safety of the unit can be ensured, and the power generation loss can be reduced.

[0023] In some embodiments, the fan operation data of each fan includes the measured blade clearance value of each fan at each moment and a plurality of first clearance-related parameters related to the measured blade clearance value of each fan. The plurality of first clearance-related parameters include: the nacelle position, generator speed, active power, pitch angle, clearance status and clearance heartbeat, original wind direction, and original wind speed of each fan; The fan operation data is recorded in the SCADA system, and the original wind direction and original wind speed are the wind direction and wind speed recorded in the SCADA system; The clearance status (ClearanceValid) and the clearance heartbeat (ClearanceHb) are two values related to the clearance value recorded in the SCADA system.

[0024] Each set of operation data in the target operation data set includes: generator speed, active power, pitch angle, measured blade clearance value, actual wind direction, and nacelle-facing wind speed; The preprocessing of the fan operation data of each fan to obtain the target operation data set includes: Eliminating the invalid data in the fan operation data of each fan; Calculating the actual wind direction of each fan according to the nacelle position and the original wind direction of each fan; The actual wind direction is the oncoming wind direction in front of the nacelle. The original wind speed is used to characterize the wind speed of the actual wind direction.

[0025] Calculating the nacelle-facing wind speed of each fan according to the original wind direction and the original wind speed of each fan;

[0026] By eliminating invalid data in the fan operation data of each fan, the effectiveness of the fan operation data can be ensured. The original wind direction is used to represent the deviation angle between the actual wind direction and the nacelle orientation, and the original wind speed is used to represent the wind speed of the actual wind direction. Therefore, based on the nacelle positions and the original wind directions of each fan, the actual wind direction of each fan can be accurately calculated. Similarly, based on the original wind directions and original wind speeds of each fan, the nacelle orientation wind speed of each fan can be accurately calculated, so as to accurately predict the blade clearance value using the actual wind direction and the nacelle orientation wind speed.

[0027] The specific calculation steps for the actual wind direction and the nacelle orientation wind speed are as follows: Based on the nacelle positions (NP1, NP2... NPn) (i.e., the nacelle orientation angles), the wind direction data (i.e., the original wind directions: ye1, ye2... yen) (the deviation angle between the actual wind direction and the nacelle orientation), and the wind speed data (i.e., the original wind speed, which is also the wind speed of the actual wind direction: ws1, ws2... wsn), calculate the actual wind direction (wd1, wd2... wdn) and the nacelle orientation wind speed (wsm1, wsm2... wsmn) for the time (t1, t2... tn). The calculation formulas are as follows:

[0028]

[0029] In some embodiments, the elimination of invalid data in the fan operation data of each fan includes: Eliminating data in the fan operation data of each fan where the generator speed is lower than the preset minimum speed; Eliminating data where the generator speed ratorspeed is 0. For the generator speeds (rs1, rs2... rsn) corresponding to the time series (t1, t2... tn), the data retention rules are as follows:

[0030] Eliminating data in the fan operation data of each fan where the clearance state and the clearance heartbeat are abnormal; For the clearance states (clv1, clv2... clvn) and the clearance heartbeats (chb1, chb2... chbn) corresponding to the time series (t1, t2... tn), the data retention rules are as follows:

[0031]

[0032]

[0033] Eliminating data in the fan operation data of each fan where the measured blade clearance value and the original wind speed remain unchanged; For the measured clearance value cv and the original wind speed ws corresponding to the time (t1, t2... tn), for the stiffness rule, the data retention rule is as follows: For any time period (tm1 - tm2)

[0034]

[0035] And

[0036] Eliminate the data in the fan operation data of each fan where the change amount of the measured blade clearance value is greater than the clearance value threshold and the change amount of the original wind speed is greater than the wind speed threshold.

[0037] For the outlier rule at any time ti, the data retention rule is as follows:

[0038] And

[0039] In some embodiments, the method of training the multiple preset neural network models based on the target operation dataset to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively includes: Group the target operation dataset according to preset grouping parameters; wherein, the preset grouping parameters include fan model parameters, blade parameters, hub height parameters, terrain features of the wind farm location, and the distance between fans; The blade parameters include: geometric parameters, aerodynamic parameters, stiffness parameters, mass and inertia parameters, etc.; The terrain features include: plain, hilly land, and mountain; The distance between fans varies according to the terrain features. For example, the distance between fans in a plain can be 5 kilometers, the distance between fans in hilly land can be 3 kilometers, and the distance between fans in a mountain can be 2 kilometers; Group the grouped target operation dataset again at preset time intervals to obtain several groups of operation datasets; For example: Based on the preprocessed data (unit rotation speed rs, active power rp, pitch angle pp, clearance value cv, actual wind direction wd, and wind speed in the direction of the unit wsm), select data for a continuous time period. First, perform the first grouping according to the fan model parameters, blade parameters, hub height parameters, topographic features of the wind farm location, and the distance between fans. Data with different model parameters, different blade parameters, different hub height parameters, different topographic features of the wind farm location, and different distances between fans are grouped separately. Then, take data with a time interval of 60 s within each group for secondary grouping, and extract continuous 60 s data from all the processed data as a data group.

[0040] Based on the several groups of operation data sets, model training is performed on multiple preset neural network models respectively to obtain the target neural network models corresponding to the multiple preset neural network models.

[0041] By grouping the target operation data set according to the preset grouping parameters, and then grouping the grouped target operation data set again according to the preset time interval, several groups of operation data sets can be obtained, thereby realizing the secondary grouping of the operation data set, ensuring the rationality and effectiveness of each operation data set. Then, based on the several groups of operation data sets, model training is performed on multiple preset neural network models respectively, which can ensure the accuracy of the target neural network models corresponding to the multiple preset neural network models.

[0042] In some embodiments, each group of operation data sets in the several groups of operation data sets includes: the measured blade clearance value and multiple second clearance-related parameters related to the measured blade clearance value; the second clearance-related parameters include: generator rotation speed, active power, pitch angle, and wind speed in the direction of the nacelle. The performing model training on multiple preset neural network models respectively based on the several groups of operation data sets to obtain the target neural network models corresponding to the multiple preset neural network models includes: Inputting the multiple second clearance-related parameters of each group of operation data sets in the several groups of operation data sets into the multiple preset neural network models to obtain the clearance value prediction results corresponding to each group of operation data sets output by the multiple preset neural network models. The multiple preset neural network models can be BP (back propagation) neural network models and LSTM (Long Short-Term Memory) neural network models.

[0043] The value of each parameter in the second clearance-related parameters may be multiple. Therefore, the clearance value prediction results corresponding to each group of operation data sets can also be multiple.

[0044] For example, each set of operating data sets includes generator speed x1 = (rs1, rs2... rs60), active power x2 = (rp1, rp2... rp60), pitch angle x3 = (pp1, pp2... pp60), actual wind direction x4 = (wd1, wd2... wd60), and unit-facing wind speed x5 = (wsm1, wsm2... wsm60) data, and the measured blade clearance value cvp1 = (cv1, cv2... cv60), where the first 5 are the second clearance correlation parameters. Inputting them into the BP neural network model can obtain the predicted result of the clearance value. =(cv1, cv2……cv60) BP , that is, there are also 60 predicted results of the clearance value.

[0045] Obtain the clearance value difference between the predicted clearance value corresponding to each set of operating data sets output by each preset neural network model and the measured blade clearance value in each set of operating data sets. Calculate the average value of the clearance value differences of each set of operating data sets corresponding to each of the preset neural network models. Determine whether the average value of the clearance value differences of each set of operating data sets corresponding to each of the preset neural network models is lower than the preset clearance value threshold. Adjust the model parameters of the preset neural network models whose average value of the clearance value differences is not lower than the preset clearance value threshold among the multiple preset neural network models until the average value of the clearance value differences of several sets of operating data sets corresponding to each of the preset neural network models is lower than the preset clearance value threshold, and then stop the adjustment to obtain the target neural network models corresponding to each of the multiple preset neural network models.

[0046] As described above, obtain the predicted clearance value output by the BP neural network model =(cv1, cv2……cv60) BP Calculate the clearance value difference between the 60 predicted clearance values cv in and the corresponding measured blade clearance values, then calculate the average value of the clearance value differences of this set of operating data sets, and determine whether the average value of the clearance value differences is lower than the preset clearance value threshold. If it is not lower than the preset clearance value threshold, it means that the accuracy of the predicted clearance value is low and the difference from the measured clearance value is too large. Therefore, it is necessary to adjust the model parameters in the BP neural network model. Similarly, if the average value of the clearance values of other sets of operating data sets corresponding to this BP neural network model is not lower than the preset clearance value threshold, continue to adjust the model parameters of the BP neural network model until the average value of the clearance value differences of several sets of operating data sets corresponding to all BP neural network models is lower than the preset clearance value threshold, and then stop the adjustment to obtain a trained BP neural network model with high accuracy of the blade clearance value.

[0047] After obtaining the net clearance value prediction results corresponding to each set of operating data sets output by multiple preset neural network models, by calculating the net clearance value differences between the net clearance value prediction results corresponding to each set of operating data sets output by each preset neural network model and the measured blade net clearance values in each set of operating data sets, and then calculating the average value of the net clearance value differences corresponding to each set of operating data sets of each preset neural network model, it can be determined whether the average value of the net clearance value differences corresponding to each set of operating data sets of each preset neural network model is lower than a preset net clearance value threshold. Adjust the model parameters of the preset neural network models whose average value of the net clearance value differences is not lower than the preset net clearance value threshold until the average values of the net clearance value differences corresponding to several sets of operating data sets of each preset neural network model are all lower than the preset net clearance value threshold, and then stop the adjustment to obtain the target neural network models corresponding to each of the multiple preset neural network models. In this way, the training of multiple neural network models can be realized to obtain multiple target neural network models with high accuracy of blade net clearance value.

[0048] In some embodiments, the combining the multiple target neural network models to obtain an optimal combined neural network model includes: Obtain multiple sets of fan test data sets; each set of fan test data sets includes: the measured blade net clearance value of a test fan and multiple third net clearance related parameters related to the measured blade net clearance value of the test fan; the third net clearance related parameters include: the generator speed, active power, pitch angle, and nacelle facing wind speed of the test fan. Of course, some of the multiple sets of fan test data sets may overlap with the above-mentioned target operating data sets, or completely overlap or be completely different. Correspondingly, some of the multiple test fans corresponding to the multiple sets of fan test data sets may be some of the above-mentioned fans, or all of the above-mentioned fans, or other fans completely different from the above-mentioned fans. Those skilled in the art can flexibly select according to needs, and the present application does not make any limitations.

[0049] Input the multiple third clearance-related parameters in each set of fan test data sets into the objective function with the preset weight coefficients to be optimized, so as to obtain the root mean square error between the predicted blade clearance values corresponding to the multiple sets of fan test data sets and the measured blade clearance values output by the objective function under each candidate value of the preset weight coefficients. Among them, the objective function is composed of multiple target neural network models and the preset weight coefficients, and the predicted blade clearance values corresponding to the multiple sets of fan test data sets are respectively output by the multiple target neural network models after each set of fan test data sets is input into the objective function; specifically, under each candidate value, after the multiple third clearance-related parameters in each set of fan test data sets are input into the objective function, each target neural network model included in the objective function outputs a predicted blade clearance value corresponding to this set of fan test data sets; for example: there are 3 sets of fan test data sets and two neural network models, namely the A target neural network model and the B target neural network model. Then, under each candidate value, these two target neural network models will both output 3 predicted blade clearance values. The objective function is to calculate the root mean square error between the 3 predicted blade clearance values corresponding to the 3 sets of fan test data sets output by the A target neural network model, the 3 predicted blade clearance values corresponding to the 3 sets of fan test data sets output by the B target neural network model, and the measured blade clearance values corresponding to the 3 sets of fan test data sets.

[0050] Obtain the optimal combined neural network model according to the root mean square error between the predicted blade clearance values corresponding to the multiple sets of fan test data sets and the measured blade clearance values output by the objective function under each candidate value of the preset weight coefficients.

[0051] Combining multiple target neural network models can obtain a combined neural network model. However, different candidate values of the weight coefficients of the combined neural network model may affect the accuracy of the finally output blade clearance values. Therefore, the multiple third clearance-related parameters in each set of fan test data sets can be respectively input into the objective function with the preset weight coefficients to be optimized, so as to obtain the root mean square error between the predicted blade clearance values corresponding to the multiple sets of fan test data sets and the measured blade clearance values output by the objective function under each candidate value of the preset weight coefficients. Then, combined with the root mean square error, the optimal combined neural network model can be obtained.

[0052] In some embodiments, the obtaining of the optimal combined neural network model according to the root mean square error between the predicted blade clearance values corresponding to the multiple sets of fan test data sets and the measured blade clearance values output by the objective function under each candidate value of the preset weight coefficients includes: With the goal of minimizing the root mean square error between the predicted blade clearance value and the measured blade clearance value for each candidate value, select the optimal candidate value from multiple candidate values of the preset weight coefficient; Combine the multiple target neural network models according to the optimal candidate value of the preset weight coefficient to obtain the optimal combined neural network model.

[0053] By aiming to minimize the root mean square error between the predicted blade clearance value and the measured blade clearance value for each candidate value, the optimal candidate value can be selected from multiple candidate values of the preset weight coefficient, and then the multiple target neural network models can be combined according to the optimal candidate value of the preset weight coefficient to obtain the optimal combined neural network model (for example: optimal combined neural network model = * target BP network model + (1 - ) * target LSTM network model, where the value of is the optimal candidate value).

[0054] A method for intelligent prediction of the clearance of a wind turbine based on a combined neural network according to the present invention includes: S1: Preprocess the operation data of each wind turbine to obtain a target operation data set: Exclude shutdown state data, organize the SCADA operation data, and exclude data in the shutdown state, that is, the generator speed ratorspeed is less than 10. For the generator speed (rs1, rs2... rsn) corresponding to the time series (t1, t2... tn), the data retention rule is as follows:

[0055] Exclude abnormal clearance state data, organize the blade-to-tower clearance distance value state (ClearanceValid) and clearance heartbeat (ClearanceHb) in the SCADA operation data, and exclude data with abnormal states and stopped heartbeats. For the clearance state (clv1, clv2... clvn) and clearance heartbeat (chb1, chb2... chbn) corresponding to the time series (t1, t2... tn), the data retention rule is as follows:

[0056]

[0057]

[0058] Stiff wild point data elimination: Further analyze the data obtained after out-of-limit data elimination to eliminate stiffness and wild points in the data. For the measured clearance value cv and the original wind speed ws corresponding to the time (t1, t2... tn), for the stiffness rule, the data retention rules are as follows: For any time period (tm1 - tm2)

[0059]

[0060] And

[0061] For the wild point rule of any time ti, the data retention rules are as follows:

[0062]

[0063] Actual wind direction and wind speed calculation of the nacelle orientation: Calculate the actual wind direction (wd1, wd2... wdn) and the wind speed of the nacelle orientation (wsm1, wsm2... wsmn) according to the nacelle position (NP1, NP2... NPn) (nacelle orientation angle), wind direction data (i.e., the original wind direction, ye1, ye2... yen) (the deviation angle between the actual wind direction and the nacelle orientation), and wind speed data (i.e., the original wind speed, ws1, ws2... wsn, representing the wind speed of the actual wind direction). For the time (t1, t2... tn), the calculation formulas are as follows:

[0064]

[0065] Data selection and grouping: Select data for continuous time periods based on the organized data (generator set speed rs, active power rp, pitch angle pp, clearance value cv, actual wind direction wd, and wind speed of the nacelle orientation wsm). First, conduct the first grouping according to the model parameters and the location of the wind farm. Data with different model parameters and different wind farm locations are grouped separately. Take data with a time period of 60s for secondary grouping within each group. Extract all continuous 60s data groups from all the processed data, and randomly select 4 / 5 of the data groups as the training set and 1 / 5 as the validation set.

[0066] S2: Train the neural network model: Clearance value prediction model based on the BP neural network model. The model flow chart is as Figure 2As shown in the figure, for the q-th group of training data, the input training set includes the unit speed x1 = (rs1, rs2... rs60), active power x2 = (rp1, rp2... rp60), pitch angle x3 = (pp1, pp2... pp60), actual wind direction x4 = (wd1, wd2... wd60), and unit-facing wind speed x5 = (wsm1, wsm2... wsm60) data, and x6 = (cv1, cv2... cv60). According to the input training set data, calculate the data of the middle hidden layer. The input signal of the hidden layer passes through the sigmoid activation function to obtain the output signal of the hidden layer , and the calculation process is as follows:

[0067] All the output signals of the hidden layer ( , ... ) are linearly weighted to obtain the input signal of the output layer , and then through the sigmod activation function, the predicted clearance value cvp1q of the output layer is finally obtained. The calculation process is as follows:

[0068]

[0069] where is the connection weight between the j-th neuron in the hidden layer and the neuron in the output layer.

[0070] According to the comparison between the predicted clearance value cvp1q ( Figure 2 unifies the predicted clearance value into cvp1 in the formula) output by the BP neural network model and the measured clearance value x6, is adjusted to obtain the target BP neural network model.

[0071] The clearance value prediction model based on the LSTM neural network is shown in Figure 3 . The input parameters of the q-th group of the model are the unit operation parameters. After the new predicted group of operation parameters are input, the old unit operation parameter data information in the memory cells is deleted through the forgetting gate. The formula is as follows:

[0072]

[0073] where is the weight of the forgetting gate layer, is the bias term of the forgetting gate layer, The prediction result cvp2q-1 of the clearance value for the (q - 1)-th training group (i.e., when the input parameter is , the clearance prediction value obtained by the LSTM neural network under the current parameter settings), where Figure 3 the prediction results of the clearance values are unified into cvp2.

[0074] After deleting the old unit operation data information, and are used as inputs, and the updated value of the sigmoid layer of the input gate layer is updated. Subsequently, a new candidate value vector is created in the tanh layer and added to the state. The calculation process formula is as follows:

[0075]

[0076] In the formula: is the weight of the input gate layer, is the bias term of the input gate layer. Subsequently, the memory cell state is updated, and the formula is as follows:

[0077] Finally, and are used as inputs, and the part of the cell state output is determined through the sigmoid layer to update the clearance value Oq, which is multiplied by the assignment of the tanh layer to obtain the final output clearance prediction value cvp2q. The calculation formula is as follows

[0078]

[0079] In the formula, is the weight of the output gate layer, is the bias term of the output gate layer.

[0080] According to the deviation between the predicted clearance value output by the LSTM neural network model and the measured clearance value , the weights and bias terms of the above layers are adjusted to obtain the target LSTM neural network model.

[0081] S3: Model verification module (1) Verify the trained target BP neural network model and target LSTM neural network model based on the validation set data.

[0082] (2) Combine the target BP neural network model and the target LSTM neural network model, and when combining, aim to minimize the root mean square error between the predicted blade clearance value and the measured blade clearance values in multiple groups of fan test data sets (where the following formula Y is the objective function, used to calculate the root mean square error between the predicted blade clearance values and the measured blade clearance values in multiple groups of fan test data sets when the preset weight coefficient takes different values (i.e., different candidate values)), and optimize the preset weight coefficient in the combined model , the formula is as follows: Y =

[0083] =(cv1, cv2... cv60) BP (that is is a matrix composed of the predicted blade clearance values corresponding to multiple groups of fan test data sets output by the target BP neural network model under a certain value of ) =(cv1, cv2... cv60) LSTM (that is is a matrix composed of the predicted blade clearance values corresponding to multiple groups of fan test data sets output by the target LSTM neural network model under a certain value of ) =(cv1, cv2... cv60) test (that is is a matrix composed of the measured blade clearance values of multiple groups of fan test data sets) In the formula, n is the number of sets in the multiple groups of fan test data sets used for testing, takes values in (0 - 1), and different values of are used as candidate values and substituted into the above formula respectively to obtain different Y values, and then the optimal value is determined according to the minimum value of the formula Y (that is, the minimum root mean square error), and then the above target BP neural network and LSTM neural network are combined using to obtain the optimal combined neural network model (for example: the optimal combined neural network model = * target BP network model + (1 - ) * target LSTM network model), which is used to predict the target clearance value prediction result of any group of fan operation data in the later stage.

[0084] S4: Input the fan operation data with the clearance value to be predicted into the combined neural network model to obtain the corresponding target clearance value prediction result. For any given fan operation data , through the calculation of the optimal combined neural network model, predict the predicted result of the target clearance value corresponding to the operating data of the fan . For the fan of the clearance device, when the clearance heartbeat stops, the clearance signal is interrupted, or the clearance jumps, the predicted result of the target clearance value can also be used to correct the measured clearance value thereof.

[0085] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present disclosure is not limited by the described action sequence, because according to the present disclosure, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present disclosure.

[0086] The above is the introduction of the method embodiments. The following further illustrates the solution of the present disclosure through device embodiments.

[0087] Figure 4 shows a block diagram of a blade clearance value prediction device 400 based on a neural network model according to an embodiment of the present disclosure. As Figure 4 shown, the device 400 includes: A first acquisition module 410, configured to acquire the operating data of each fan; A processing module 420, configured to preprocess the operating data of each fan to obtain a target operating data set; A training module 430, configured to perform model training on a plurality of preset neural network models based on the target operating data set to obtain the trained target neural network models corresponding to the plurality of preset neural network models respectively; A combination module 440, configured to combine a plurality of the target neural network models to obtain an optimal combined neural network model; A second acquisition module 450, configured to acquire the operating data of the fan whose clearance value is to be predicted; A prediction module 460, configured to input the operating data of the fan whose clearance value is to be predicted into the optimal combined neural network model to obtain a corresponding predicted result of the target clearance value.

[0088] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the described modules can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0089] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device and a non-transitory computer-readable storage medium storing computer instructions.

[0090] Figure 5 FIG. shows a schematic block diagram of an electronic device 800 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0091] The device 800 includes a computing unit 801 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0092] A plurality of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a magnetic disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0093] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 executes the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of method 100 described above can be executed. Alternatively, in other embodiments, the computing unit 801 can be configured to execute method 100 in any other suitable manner (e.g., by means of firmware).

[0094] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), system on a chip systems (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special or general-purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0095] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program code is executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0096] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0097] In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0098] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0099] A computing system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, or a server of a distributed system, or a server incorporating a blockchain.

[0100] It should be understood that the various forms of processes shown above can be used, with steps reordered, added or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and there is no limitation herein. The above specific embodiments do not constitute a limitation on the protection scope of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A method for predicting the blade clearance value based on a neural network model, characterized in that, Including: Obtaining the fan operation data of each fan; Preprocessing the fan operation data of each fan to obtain a target operation data set; Based on the target operation data set, training multiple preset neural network models to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively; Combining multiple said target neural network models to obtain an optimal combined neural network model; Obtaining the fan operation data with the net clearance value to be predicted; Inputting the fan operation data with the net clearance value to be predicted into the optimal combined neural network model to obtain a corresponding target net clearance value prediction result.

2. The method according to claim 1, wherein: The fan operation data of each fan includes the measured blade net clearance value of each fan at each moment and multiple first net clearance correlation parameters related to the measured blade net clearance value of each fan. The multiple first net clearance correlation parameters include: the nacelle position, generator speed, active power, pitch angle, net clearance state and net clearance heartbeat, original wind direction, and original wind speed of each fan; Each group of operation data sets in the target operation data set includes: generator speed, active power, pitch angle, measured blade net clearance value, actual wind direction, and nacelle-facing wind speed; The preprocessing of the fan operation data of each fan to obtain a target operation data set includes: Eliminating invalid data in the fan operation data of each fan; Calculating the actual wind direction of each fan according to the nacelle position and original wind direction of each fan; Calculating the nacelle-facing wind speed of each fan according to the original wind direction and original wind speed of each fan.

3. The method according to claim 2, wherein: The eliminating of invalid data in the fan operation data of each fan includes: Eliminating data in the fan operation data of each fan where the generator speed is lower than the preset minimum speed; Eliminating data in the fan operation data of each fan where the net clearance state and net clearance heartbeat are abnormal; Eliminating data in the fan operation data of each fan where both the measured blade net clearance value and the original wind speed remain unchanged; Eliminating data in the fan operation data of each fan where the change amount of the measured blade net clearance value is greater than the net clearance value threshold and the change amount of the original wind speed is greater than the wind speed threshold.

4. The method according to claim 1, wherein: The training of multiple preset neural network models based on the target operation data set to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively includes: Grouping the target operation data set according to preset grouping parameters; wherein, the preset grouping parameters include fan model parameters, blade parameters, hub height parameters, and the terrain features of the wind farm location, as well as the distance between fans; Grouping the grouped target operation data set again at a preset time interval to obtain several groups of operation data sets; Based on the several groups of operation data sets, respectively training multiple preset neural network models to obtain the target neural network models corresponding to the multiple preset neural network models respectively.

5. The method according to claim 4, wherein: Each of the several sets of operation datasets includes: the measured blade clearance value and a plurality of second clearance-related parameters associated with the measured blade clearance value; the second clearance-related parameters include: generator speed, active power, pitch angle, and nacelle-facing wind speed; Based on the several sets of operation datasets, model training is respectively performed on a plurality of preset neural network models to obtain the target neural network models corresponding to the plurality of preset neural network models, including: Input the plurality of second clearance-related parameters of each set of operation datasets in the several sets of operation datasets into the plurality of preset neural network models to obtain the clearance value prediction results corresponding to each set of operation datasets output by the plurality of preset neural network models; Calculate the clearance value difference between the clearance value prediction result corresponding to each set of operation datasets output by each preset neural network model and the measured blade clearance value in each set of operation datasets; Calculate the average value of the clearance value differences of each set of operation datasets corresponding to each preset neural network model; Determine whether the average value of the clearance value differences of each set of operation datasets corresponding to each preset neural network model is lower than a preset clearance value threshold; Adjust the model parameters of the preset neural network models in which the average value of the clearance value differences is not lower than the preset clearance value threshold among the plurality of preset neural network models until the average value of the clearance value differences of the several sets of operation datasets corresponding to each preset neural network model is lower than the preset clearance value threshold, and then stop the adjustment to obtain the target neural network models corresponding to the plurality of preset neural network models.

6. The method according to any one of claims 1 to 5, wherein The combination of the plurality of target neural network models to obtain an optimal combined neural network model includes: Obtain multiple sets of wind turbine test datasets; each set of wind turbine test datasets includes: the measured blade clearance value of a test wind turbine and a plurality of third clearance-related parameters associated with the measured blade clearance value of the test wind turbine; the plurality of third clearance-related parameters include: the generator speed, active power, pitch angle, and nacelle-facing wind speed of the test wind turbine; Input the plurality of third clearance-related parameters in each set of wind turbine test datasets into an objective function with preset weight coefficients to be optimized, to obtain the root mean square error between the predicted blade clearance value corresponding to the multiple sets of wind turbine test datasets and the measured blade clearance value at each candidate value of the preset weight coefficients, wherein the objective function is composed of a plurality of target neural network models and the preset weight coefficients, and the predicted blade clearance value corresponding to the multiple sets of wind turbine test datasets is output by the plurality of target neural network models respectively after each set of wind turbine test datasets is input into the objective function; According to the root mean square error between the predicted blade clearance value corresponding to the multiple sets of wind turbine test datasets and the measured blade clearance value at each candidate value of the preset weight coefficients output by the objective function, obtain the optimal combined neural network model.

7. The method according to claim 6, wherein The root mean square error between the predicted blade clearance values corresponding to the multiple groups of fan test data sets and the measured blade clearance values output according to the objective function at each candidate value of the preset weight coefficient is obtained to get the optimal combined neural network model, including: Taking the minimum root mean square error between the predicted blade clearance values and the measured blade clearance values at each candidate value as the objective, an optimal candidate value is selected from multiple candidate values of the preset weight coefficient; The multiple target neural network models are combined according to the optimal candidate value of the preset weight coefficient to obtain the optimal combined neural network model.

8. A blade clearance value prediction device based on a neural network model, characterized in that, Including: A first acquisition module, configured to acquire the fan operation data of each fan; A processing module, configured to preprocess the fan operation data of each fan to obtain a target operation data set; A training module, configured to perform model training on multiple preset neural network models based on the target operation data set to obtain the trained target neural network models corresponding to the multiple preset neural network models respectively; A combination module, configured to combine multiple target neural network models to obtain an optimal combined neural network model; A second acquisition module, configured to acquire the fan operation data for which the blade clearance value is to be predicted; A prediction module, configured to input the fan operation data for which the blade clearance value is to be predicted into the optimal combined neural network model to obtain a corresponding target blade clearance value prediction result.

9. An electronic device, characterized in that, Including: A memory and a processor, The memory stores a computer program, and when the processor executes the program, the method described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by the processor corresponding to the electronic device, the electronic device can implement the method for predicting the blade clearance value based on the neural network model described in any one of claims 1-7.

Images