Load prediction method and device for wind generating set

Abstract

The invention relates to a load prediction method and device for a wind generating set. The load prediction method for the wind generating sets comprises the steps of determining a graph network with each wind generating set in the multiple wind generating sets as a node based on operation data of the multiple wind generating sets; based on the structural characteristics of the graph network, selecting a wind generating set for load testing; carrying out load test on the selected wind generating set; training and establishing a load prediction model for the plurality of wind generating sets based on the load test data and the operation data of the selected wind generating sets; and predicting the load of the target wind generating set through the load prediction model based on the operation data of the target wind generating set. According to the method and the device, load prediction and life evaluation of the whole unit can be carried out according to the topological structure of the graph network model only by carrying out load test on part of the units, so that the load prediction efficiency of the whole unit is remarkably improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind power generation, and more specifically, to a method and device for predicting the load of a wind power generator set. Background Art

[0002] In the design of wind turbines, it is necessary to determine the various loads generated by the wind turbines in the environment and under various operating conditions. The purpose is to perform strength analysis (including static strength analysis and fatigue strength analysis), dynamic analysis and life calculation on the wind turbine components, so as to ensure that the wind turbines can operate normally within their designed life. This work is the most basic work in the design of wind turbines, and all subsequent work is based on load calculation.

[0003] Usually, wind turbines operate under very complex and changeable natural conditions, and different external environments have different effects on the stress state of each component of the wind turbine. In order to achieve the purpose of load calculation and meet the design requirements, the load calculation of the wind turbine is based on the design conditions that include various important conditions that the wind turbine will bear. The determination of the calculation conditions should comprehensively consider the combination of external conditions, the probability of occurrence of the wind turbine operating state, and the action of the control and protection system, and be formulated in combination with the international standards for wind turbine design conditions and load states (for example, EIC61400).

[0004] The existing wind turbine load prediction technology can realize the load prediction of a single unit. The load prediction model of the wind turbine is established through the unit's operating data and the load data at the corresponding time. Based on this model, the operating data at different times are input into the model to perform the load prediction process of the wind turbine. For example, the existing wind turbine load calculation method can obtain the load simulation operation results of the unit by simulating the Bladed model of the unit under various working conditions. The load prediction method of the wind turbine mainly performs load tests on each unit to establish a load prediction model to achieve load prediction for the unit. Load testing for each unit to establish a load prediction model requires obtaining a large amount of data, resulting in problems such as long data testing time and low load prediction efficiency. Summary of the invention

[0005] In order to solve the above problems, the present disclosure proposes a load prediction method and device for a wind turbine generator set, a computing system, and a computer-readable storage medium.

[0006] According to one aspect of the present disclosure, a load prediction method for a wind turbine is provided, the load prediction method comprising: determining a graph network with each wind turbine in the plurality of wind turbines as a node based on operating data of the plurality of wind turbines; selecting a wind turbine for load testing based on structural characteristics of the graph network; performing a load test on the selected wind turbine; training and establishing a load prediction model for the plurality of wind turbines based on the load test data and operating data of the selected wind turbine; and predicting the load of the target wind turbine by means of the load prediction model based on the operating data of the target wind turbine.

[0007] Optionally, the number of the selected wind turbine generator sets is smaller than the number of the plurality of wind turbine generator sets.

[0008] Optionally, the operating data includes at least one of rotation speed related data, power related data, acceleration related data, angle related data, wind speed related data, temperature related data and operating flag related data.

[0009] Optionally, the step of determining, based on the operating data of the multiple wind turbines, a graph network with each of the multiple wind turbines as a node includes: determining the similarity between the wind turbines based on the operating data of the multiple wind turbines; and connecting each pair of nodes corresponding to a similarity greater than a predetermined similarity threshold to each other to construct the graph network.

[0010] Optionally, the predetermined similarity threshold may be an average value of similarities between various wind turbine generator sets.

[0011] Optionally, the step of selecting the wind turbines for load testing based on the structural characteristics of the graph network includes: determining the node with the largest number of connected edges for each graph network block in the graph network blocks that are not connected to each other; determining the wind turbine corresponding to the node with the largest number of connected edges in each graph network block as the wind turbine for load testing, or determining the wind turbine with the highest annual average wind speed among multiple wind turbines corresponding to multiple nodes with the largest number of connected edges in each graph network block as the wind turbine for load testing.

[0012] Optionally, the input data of the load prediction model includes operating data of the selected wind turbine generator set, and the target data of the load prediction model includes load test data of the selected wind turbine generator set.

[0013] Optionally, the step of training and establishing a load prediction model for the multiple wind turbines based on the load test data and operation data of the selected wind turbines includes: determining the predicted load of the selected wind turbine based on the adjacency matrix of the graph network, the input data and the preset weights of the load prediction model; in response to the difference between the load test data and the predicted load being greater than a predetermined error threshold, adjusting the size of the preset weight and re-determining the predicted load of the selected wind turbine; in response to the difference between the load test data and the predicted load being less than or equal to the predetermined error threshold, determining the current preset weight as the final weight of the load prediction model; and establishing the load prediction model based on the final weight.

[0014] Optionally, the load prediction method also includes: before training and establishing a load prediction model for the multiple wind turbines, integrating load test data and operation data with the same timestamp of the same wind turbine; and deleting abnormal operation data from the operation data of each wind turbine in the multiple wind turbines according to the wind turbine operation flag included in the operation data.

[0015] According to another aspect of the present disclosure, there is provided a load prediction device for a wind turbine generator set, the load prediction device comprising: a graph network determination unit, which determines a graph network with each of the multiple wind turbine generator sets as a node based on operation data of the multiple wind turbine generator sets; a selection unit, which selects a wind turbine generator set for load testing based on structural characteristics of the graph network; a load testing unit, which performs a load test on the selected wind turbine generator set; a prediction model determination unit, which trains and establishes a load prediction model for the multiple wind turbine generator sets based on the load test data and operation data of the selected wind turbine generator set; and a prediction unit, which predicts the load of the target wind turbine generator set through the load prediction model based on the operation data of the target wind turbine generator set.

[0016] Optionally, the number of the selected wind turbine generator sets is smaller than the number of the plurality of wind turbine generator sets.

[0017] Optionally, the operating data includes at least one of rotation speed related data, power related data, acceleration related data, angle related data, wind speed related data, temperature related data and operating flag related data.

[0018] Optionally, the graph network determination unit is further configured to: determine the similarity between each wind turbine generator set based on the operating data of the multiple wind turbine generator sets; and connect each pair of nodes corresponding to the similarity greater than a predetermined similarity threshold to each other to construct the graph network.

[0019] Optionally, the predetermined similarity threshold is an average value of similarities between various wind turbine generator sets.

[0020] Optionally, the selection unit is further configured to: determine the node with the largest number of connected edges for each graph network block in the graph network blocks that are not connected to each other; determine the wind turbine corresponding to the node with the largest number of connected edges in each graph network block as the wind turbine for load testing, or determine the wind turbine with the highest annual average wind speed among multiple wind turbines corresponding to multiple nodes with the largest number of connected edges in each graph network block as the wind turbine for load testing.

[0021] Optionally, the input data of the load prediction model includes operating data of the selected wind turbine generator set, and the target data of the load prediction model includes load test data of the selected wind turbine generator set.

[0022] Optionally, the prediction model determination unit is further configured to: determine the predicted load of the selected wind turbine generator set based on the adjacency matrix of the graph network, the input data and the preset weights of the load prediction model; in response to the difference between the load test data and the predicted load being greater than a predetermined error threshold, adjust the size of the preset weight and re-determine the predicted load of the selected wind turbine generator set; in response to the difference between the load test data and the predicted load being less than or equal to the predetermined error threshold, determine the current preset weight as the final weight of the load prediction model; and establish the load prediction model based on the final weight.

[0023] Optionally, the load prediction device also includes: a data integration unit, which integrates the load test data and operation data with the same timestamp of the same wind turbine before training and establishing the load prediction model for the multiple wind turbines; and a data screening unit, which deletes abnormal operation data from the operation data of each wind turbine in the multiple wind turbines according to the wind turbine operation flag included in the operation data.

[0024] According to another aspect of the present disclosure, a computing system is provided that includes at least one computing device and at least one storage device storing instructions, wherein the instructions, when executed by the at least one computing device, prompt the at least one computing device to execute the load prediction method for the wind turbine generator set as described above.

[0025] According to another aspect of the present disclosure, a computer-readable storage medium storing instructions is provided, wherein when the instructions are executed by at least one computing device, the at least one computing device is prompted to execute the load prediction method for the wind turbine generator set as described above.

[0026] By adopting the present disclosure, a real-time load prediction and life assessment system for wind turbines can be established based on a graph network model. Only load tests on some units are required to perform load prediction and life assessment on the entire field of units according to the topological structure of the graph network model. There is no need to perform "one machine one test" on the load of the units, which significantly improves the efficiency of load prediction for the entire field of units. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The above and / or other objects and advantages of the present disclosure will become more apparent through the following description of embodiments in conjunction with the accompanying drawings, in which:

[0028] Figure 1 is a flow chart showing a load prediction method for a wind turbine generator system according to an exemplary embodiment of the present disclosure;

[0029] Figure 2 is a diagram showing a representation of a graph network connection according to an example;

[0030] Figure 3 is a diagram illustrating a block partitioning of a graph network according to an example;

[0031] Figure 4 is a flow chart illustrating pre-processing of load data and full-field SCADA data after a load test is completed according to an exemplary embodiment of the present disclosure;

[0032] Figure 5 is a flow chart illustrating the establishment of a load prediction model according to an exemplary embodiment of the present disclosure;

[0033] Figure 6 is a block diagram showing a load prediction device for a wind turbine generator set according to an exemplary embodiment of the present disclosure;

[0034] Figure 7 is a block diagram illustrating a computing system including at least one computing device and at least one storage device storing instructions according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0035] Below, in conjunction with the accompanying drawings, a description of a specific embodiment is provided to help the reader obtain a comprehensive understanding of the method, device and / or system described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the method, device and / or system described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0036] Load refers to external forces and other factors that cause internal forces and deformations in structures or components. Traditional load calculation and analysis methods can only calculate and analyze the design loads of wind turbines, and cannot achieve real-time monitoring and analysis of wind turbine loads. There are methods for real-time load prediction and life assessment in the prior art. However, the load prediction methods in the prior art can only achieve load prediction for a single unit. If load prediction for all units is required, each unit needs to be load tested. The establishment of a load prediction model requires the operation data and corresponding load data of each unit.

[0037] In order to solve the above problems, the present disclosure proposes a load prediction method and device for a wind turbine generator set, a computing system and a computer-readable storage medium. According to the load prediction method and device for a wind turbine generator set disclosed in the present disclosure, it is not necessary to perform load tests on each wind turbine generator set separately, but only needs to perform load tests on some wind turbine generator sets to perform real-time load tests and life extension evaluation on all wind turbine generator sets.

[0038] Specific examples of the present disclosure are described below with reference to the accompanying drawings.

[0039] Figure 1 is a flow chart illustrating a load prediction method for a wind turbine generator system according to an exemplary embodiment of the present disclosure. Figure 2 is a diagram illustrating representations of graph network connections according to an example. Figure 3 is a diagram illustrating a graph network block partitioning according to an example.

[0040] like Figure 1As shown, in step S101, based on the operating data of multiple wind turbines, a graph network with each wind turbine in the multiple wind turbines as a node is determined. A graph network is a set of functions organized in a graph structure in a topological space for relational reasoning. It is composed of interconnected graph network blocks and can be specialized into various connectionist models, including feedforward neural networks, recursive neural networks, etc. Among them, the edges of the graph network refer to the links between nodes, the degree of a node represents the number of edges connecting the node to other nodes, and the adjacency matrix represents the matrix of adjacent relationships between nodes. Among them, the operating data includes at least one of the speed-related data (for example, generator speed, impeller speed, etc.), power-related data (for example, active power, etc.), acceleration-related data (for example, nacelle x-direction acceleration, nacelle y-direction acceleration, etc.), angle-related data (for example, pitch angle, yaw angle, etc.), wind speed-related data (for example, 1-minute average wind speed, etc.), temperature-related data (for example, nacelle temperature, ambient temperature, main bearing temperature, etc.) and operation flag-related data (for example, fan operation flag, power limit flag, shutdown flag, etc.).

[0041] In the example, based on the operating data of multiple wind turbines, the similarity between each wind turbine is determined; each pair of nodes corresponding to the similarity greater than a predetermined similarity threshold is connected to each other to construct a graph network. Here, the predetermined similarity threshold is the average value of the similarity between each wind turbine. The graph network is constructed based on the similarity, which provides a basis for the subsequent division of the graph network blocks and thus reduces the number of wind turbines for load testing. For example, each wind turbine generator set in the field is taken as a node of the graph network, and the operation data of each wind turbine generator set is used to establish the graph network structure. Specifically, the data with a high correlation with the load, such as the generator speed, active power, impeller speed, nacelle x-direction acceleration, nacelle y-direction acceleration, pitch angle, yaw angle data, etc., are taken as the benchmark data. The similarity between the benchmark data of different units is calculated by, for example, the Pearson correlation coefficient method, and the mean of the similarity of each unit is taken as the division condition. Two units (nodes) with a low similarity value (for example, less than 0.5) are identified as having no edge connection between each other. On the contrary, two units (nodes) with a high similarity value (for example, greater than or equal to 0.5) are identified as having an edge connection between each other. In addition, the specific connection form can be represented in the form of an adjacency matrix (if there is an edge connection between two nodes, the value of the corresponding position in the matrix is ​​1, otherwise it is 0). For example, according to the example, the representation form of the graph network connection can be as follows Figure 2 As shown, Figure 2 The adjacency matrix corresponding to the graph network connections shown is as follows:

[0042]

[0043] The row and column numbers both represent the node numbers. For example, since node 1 and node 2 are connected (e.g. Figure 2 As shown in the figure), the value of the first row and second column in the matrix is ​​1; since nodes 1 and 3 are not connected (as shown in the figure), Figure 2 ), so the value of the first row and third column in the matrix is ​​0.

[0044] In step S102, based on the structural characteristics of the graph network, wind turbine generator sets for load testing are selected. In the example, the number of the selected wind turbine generator sets is less than the number of the plurality of wind turbine generator sets.

[0045] In the example, for each of the graph network blocks that do not have connections between each other, a node with the largest number of connected edges (e.g., Figure 3 Node 3, node 8, and nodes 11 to 14 in each graph network block); the node with the largest number of connected edges in each graph network block (for example, Figure 3 The wind turbine generator set corresponding to the nodes 3 and 8 in the graph is determined as the wind turbine generator set for load testing, or the multiple nodes with the largest number of connected edges in each graph network block (for example, Figure 3 The wind turbine with the highest annual average wind speed among the multiple wind turbines corresponding to nodes 11 to 14 in the graph is determined as the wind turbine for load testing. Therefore, for each graph network block, only the wind turbine corresponding to one of the nodes needs to be load tested. For example, the specific graph partitioning method is: based on the similarity in the above node partitioning, if there are some nodes that have no edge links with a group of nodes and have edge links with another group of nodes (for example, Figure 3 As shown in the figure, nodes 1 to 5 are one graph network block, nodes 6 to 10 are another graph network block, and nodes 11 to 14 are another graph network block. There is no connection relationship between these three graph network blocks. In this case, the large graph needs to be divided, and then the load test is performed according to the number of divisions of the large graph (for example, Figure 3 As shown, three units need to be load tested), and the unit to be tested is the unit corresponding to the node with the largest number of connected edges (if the number of connected edges is the same, the unit site with the largest average wind speed in a year needs to be tested).

[0046] Therefore, according to the load testing method implemented according to the graph network model disclosed in the present invention, only the load data of a small number of units need to be tested to perform load prediction and life assessment of the entire unit, without the need to perform "one machine one test" on the load of the unit.

[0047] In step S103, a load test is performed on the selected wind turbine generator set. In an example, the loads of key parts of the wind turbine of the selected wind turbine generator set (eg, tower bottom bending moment load, tower top bending moment load, etc.) are tested.

[0048] In step S104, based on the load test data and operation data of the selected wind turbine generator set, a load prediction model for multiple wind turbine generator sets is trained and established. In the example, the input data of the load prediction model includes the operation data of the selected wind turbine generator set, and the target data of the load prediction model includes the load test data of the selected wind turbine generator set. For a detailed description of step S104, please refer to Figure 5 And its related description. In the example, based on the adjacency matrix of the graph network, the input data and the preset weights of the load prediction model, the predicted load of the selected wind turbine is determined; in response to the difference between the load test data and the predicted load being greater than the predetermined error threshold, the size of the preset weight is adjusted, and the predicted load of the selected wind turbine is re-determined; in response to the difference between the load test data and the predicted load being less than or equal to the predetermined error threshold, the current preset weight is determined as the final weight of the load prediction model; based on the final weight, a load prediction model is established. Through the above processing, a real-time load prediction model for a wind turbine can be quickly and accurately established based on a graph network model.

[0049] In step S105, based on the operating data of the target wind turbine generator set, the load of the target wind turbine generator set is predicted by the load prediction model. For example, the operating data (e.g., SCADA data) of the target wind turbine generator set within a preset time period can be input into the load prediction model, and the output result of the load prediction model is used as the real-time predicted load of the target wind turbine generator set. In addition, the load prediction model can be used to predict the loads at key positions of the wind turbine generator set (e.g., tower bottom bending moment load, tower top bending moment load, etc.), and when the predicted load is too large, the corresponding load reduction control strategy (including, for example, adjusting the pitch angle, adjusting the impeller speed, power control, etc.) is adopted to reduce the load of the key components of the wind turbine to prevent the wind turbine generator set from exceeding the load limit, thereby achieving the purpose of protecting the wind turbine generator set.

[0050] In addition, the load prediction method of a wind turbine generator set according to an exemplary embodiment of the present disclosure may further include: before training and establishing a load prediction model for multiple wind turbine generator sets, integrating load test data and operation data with the same timestamp of the same wind turbine generator set; according to the wind turbine operation flag bit included in the operation data, deleting abnormal operation data from the operation data of each wind turbine generator set in the multiple wind turbine generator sets. For a detailed description of the steps of preprocessing the load data and the entire field SCADA data, please refer to Figure 4Through the above processing, input data and target data with a unified format and standardization can be obtained, which is conducive to ensuring the accuracy of the load prediction model obtained by training.

[0051] According to the load testing method of the wind turbine generator set of the exemplary embodiment of the present disclosure, it is possible to establish a real-time load prediction and life assessment system for the wind turbine generator set based on a graph network model. Only load testing of some units is required to perform load prediction and life assessment of the entire set according to the topological structure of the graph network model. There is no need to perform "one machine one test" on the load of the unit, thereby significantly improving the efficiency of load prediction for the entire set.

[0052] Figure 4 is a flow chart illustrating pre-processing of load data and full-field SCADA data after a load test is completed according to an exemplary embodiment of the present disclosure.

[0053] After the load test is completed and before the load prediction model for multiple wind turbines is trained and established, the load test data and operation data of the same wind turbine with the same timestamp can be integrated; according to the wind turbine operation flag included in the operation data, the abnormal operation data is deleted from the operation data of each wind turbine in the multiple wind turbines to pre-process the load data and the SCADA data of the entire field. Figure 4 shown.

[0054] In step S401, the load data of the unit undergoing the load test and the SCADA data of the entire field are obtained.

[0055] In step S402, the load data is merged with the SCADA data according to the same timestamp and unit number. For example, the load data with the same timestamp and unit number are merged with the fields required by the SCADA data (the load data of the units that have not been subjected to load testing are all padded with 0 and merged with the SCADA units, and the required fields include generator speed, active power, impeller speed, cabin x-direction acceleration, cabin y-direction acceleration, pitch angle, yaw angle, 1-minute average wind speed, fan operation flag, cabin temperature, ambient temperature, main bearing temperature, etc.).

[0056] In step S403, the data of the normal operation of the unit is retained according to the unit operation law. For example, the data is processed using the flag bit of the normal operation of the SCADA data during the operation of the unit (for example, the power limit flag bit, the shutdown flag bit, etc.) to obtain the data of the normal operation of the unit (for example, the data whose flag bit indicates abnormal operation is eliminated, and the data without the flag bit is processed according to the normal operation status of the unit).

[0057] In step S404, the data with the same timestamp are arranged in sequence according to the unit number to obtain the graph model input data. For example, the data with the same timestamp are integrated in sequence according to the unit number, and finally the whole field data is integrated into a set of data. The obtained data format is shown in the following table (taking two units in the whole field as an example). The data is divided into training set and test set, and the subsequent real-time data still needs to be preprocessed using the above method.

[0058]

[0059] After the data processing is completed, the data format in which the SCADA data and the load data correspond one to one is obtained. Then, the SCADA data is used as the input data, and the load data at the same time is used as the target data, and is input into the graph network model. Taking the above table as an example, since the loads of three units are calculated, the size of the adjacency matrix is ​​3×3. Since only two of the units are tested, although the data of the three units are used as input data, only the semi-supervised form is used to update the weights using the errors of the two units. Through the above processing, input data and target data with a unified format and standardization can be obtained, which is conducive to ensuring the accuracy of the trained load prediction model.

[0060] Figure 5 is a flow chart illustrating the establishment of a load prediction model according to an exemplary embodiment of the present disclosure.

[0061] During the training phase, the following formula is used to update the graph network weights (the model will have multiple weight update formulas combined as a multi-layer network model):

[0062] H=σ(AXW)

[0063] Among them, σ represents the activation function, A represents the adjacency matrix, X is the input data, H is the predicted data, W is the weight that needs to be updated by the model, and the weight W is updated by calculating the error between H and the target data.

[0064] After the training phase is completed, the weights of the multi-layer network model have been updated, and the weights and advanced matrices of the model can represent the load of the entire unit. At this point, the SCADA data of any unit can be input into the model to obtain the predicted load data of the unit.

[0065] Finally, the real-time SCADA data of the entire field unit is input into the preprocessing program and the constructed real-time load prediction model of the entire field wind turbine generator set, so as to realize the real-time load prediction and life assessment of the entire field unit.

[0066] The detailed process of establishing a real-time load prediction model using graph network algorithms is as follows: Figure 5 shown.

[0067] In step S501, a graph network model is constructed based on the entire field SCADA data. For example, as described above, based on the operation data of multiple wind turbines, the similarity between each wind turbine is determined; each pair of nodes corresponding to a similarity greater than a predetermined similarity threshold is connected to each other to construct a graph network; based on the structural characteristics of the graph network, a wind turbine for load testing is selected.

[0068] In step S502, the load data of the wind turbine generator set selected for load test and the SCADA data of all wind turbine generator sets are obtained, and the data are preprocessed (for example, Figure 4 Subsequently, the processed SCADA data and the measured load data (SCADA data as input features, measured load data as output features) are divided into training data and test data (processed real-time data will also be transmitted after model training is completed).

[0069] In step S503, a load prediction model is established, and the training data is input into the load prediction model. Subsequently, in step S504, it is determined whether the loss converges. If it is determined in step S504 that the loss converges, the process proceeds to step S505, otherwise it returns to step S503 to update the model parameters. In step S505, it is determined whether the accuracy requirement is met. In step S505, the test data is compared with the output of the load prediction model. For example, if the difference between the test data and the output of the load prediction model is less than a predetermined threshold, it means that the accuracy requirement is met, otherwise it means that the accuracy requirement is not met. If it is determined in step S505 that the current load prediction model does not meet the accuracy requirement, the process proceeds to step S506 to determine whether the predetermined number of training times has been reached. If it is determined in step S506 that the predetermined number of training times has not been reached, the process returns to step S503 to update the model parameters to obtain an updated load prediction model, and the process proceeds to steps S504, S505 and S506 again for corresponding judgment. If it is determined in step S506 that the predetermined number of training times has been reached, the process proceeds to step S508 to output the load prediction model with the highest accuracy and its actual accuracy. In addition, if it is determined in step S505 that the load prediction model meets the accuracy requirement, the process proceeds to step S507 to determine the final load prediction model. Finally, in step S509, the load prediction model with the highest accuracy determined in step S508 or the final load prediction model determined in step S507 is used to predict the load of the target wind turbine generator set and perform life assessment based on the processed real-time data obtained in step S502.

[0070] Through the above processing, a real-time load prediction and life assessment system for wind turbines can be established quickly and accurately based on the graph network model, thereby significantly improving the accuracy and efficiency of load prediction for the entire unit.

[0071] Figure 6 is a block diagram illustrating a load prediction device for a wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0072] like Figure 6 As shown, a load prediction device 600 for a wind turbine generator set according to an exemplary embodiment of the present disclosure includes: a graph network determination unit 601, which determines a graph network with each wind turbine generator set in a plurality of wind turbine generator sets as a node based on operation data of the plurality of wind turbine generator sets; a selection unit 602, which selects a wind turbine generator set for load testing based on structural characteristics of the graph network; a load testing unit 603, which performs a load test on the selected wind turbine generator set; a prediction model determination unit 604, which trains and establishes a load prediction model for a plurality of wind turbine generator sets based on the load test data and operation data of the selected wind turbine generator set; and a prediction unit 605, which predicts the load of a target wind turbine generator set through a load prediction model based on the operation data of the target wind turbine generator set.

[0073] In an example, the selected number of wind turbines is less than the number of the plurality of wind turbines.

[0074] In the example, the operating data includes at least one of speed-related data (e.g., generator speed, impeller speed, etc.), power-related data (e.g., active power, etc.), acceleration-related data (e.g., nacelle x-direction acceleration, nacelle y-direction acceleration, etc.), angle-related data (e.g., pitch angle, yaw angle, etc.), wind speed-related data (e.g., 1-minute average wind speed, etc.), temperature-related data (e.g., nacelle temperature, ambient temperature, main bearing temperature, etc.) and operating flag-related data (e.g., wind turbine operating flag, power limit flag, shutdown flag, etc.).

[0075] In the example, the graph network determination unit 601 is further configured to: determine the similarity between each wind turbine generator set based on the operating data of multiple wind turbine generator sets; connect each pair of nodes corresponding to the similarity greater than a predetermined similarity threshold to each other to construct a graph network.

[0076] In an example, the predetermined similarity threshold comprises an average value of similarities between the wind turbines.

[0077] In the example, the selection unit 602 is further configured to: determine the node with the largest number of connected edges for each graph network block in the graph network blocks that are not connected to each other; determine the wind turbine corresponding to the node with the largest number of connected edges in each graph network block as the wind turbine for load testing, or determine the wind turbine with the highest annual average wind speed among multiple wind turbines corresponding to multiple nodes with the largest number of connected edges in each graph network block as the wind turbine for load testing.

[0078] In an example, input data of the load prediction model includes operating data of the selected wind turbine generator set, and target data of the load prediction model includes load test data of the selected wind turbine generator set.

[0079] In the example, the prediction model determination unit 604 is further configured to: determine the predicted load of the selected wind turbine generator set based on the adjacency matrix of the graph network, the input data and the preset weights of the load prediction model; in response to the difference between the load test data and the predicted load being greater than a predetermined error threshold, adjust the size of the preset weight and re-determine the predicted load of the selected wind turbine generator set; in response to the difference between the load test data and the predicted load being less than or equal to the predetermined error threshold, determine the current preset weight as the final weight of the load prediction model; and establish a load prediction model based on the final weight.

[0080] In the example, the load prediction device 600 also includes: a data integration unit, which integrates the load test data and operation data with the same timestamp of the same wind turbine before training and establishing a load prediction model for multiple wind turbines; and a data screening unit, which deletes abnormal operation data from the operation data of each wind turbine in the multiple wind turbines according to the wind turbine operation flag included in the operation data.

[0081] Combination of the above Figures 1 to 5 The specific operations shown are respectively Figure 6 The load prediction device 600 of the wind turbine generator set shown in the figure is executed by the corresponding units, and the specific operation details will not be repeated here. The load test device of the wind turbine generator set according to the exemplary embodiment of the present disclosure can establish a real-time load prediction and life assessment system for the wind turbine generator set based on the graph network model. Only the load test of some units is required to perform load prediction and life assessment of the entire set according to the topological structure of the graph network model, without the need to perform "one unit one test" on the load of the unit, thereby significantly improving the efficiency of load prediction of the entire set.

[0082] Figure 7is a block diagram illustrating a computing system including at least one computing device and at least one storage device storing instructions according to an exemplary embodiment of the present disclosure.

[0083] like Figure 7 As shown, a computing system 700 provided according to an exemplary embodiment of the present invention includes a computing device 701 and a storage device 702, wherein the storage device 702 stores computer executable instructions. When the computer executable instructions are executed by the computing device 701, the load prediction method for the wind turbine generator set described in any of the aforementioned embodiments is executed.

[0084] The computing device 701 is deployed in a server or client, and may also be deployed on a node device in a distributed network environment. In addition, the computing device 701 may be a PC computer, a tablet device, a personal digital assistant, a smart phone, a web application or other device capable of executing the above-mentioned instruction set. Here, the computing device is not necessarily a single computing device, but may also be any device or circuit capable of executing the above-mentioned instruction (or instruction set) individually or jointly. The computing device may also be a part of an integrated control system or a system manager, or may be configured as a portable electronic device interconnected with an interface locally or remotely (e.g., via wireless transmission). In the computing device, the processor includes a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor also includes an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0085] According to another aspect of the present disclosure, a computer-readable storage medium storing instructions is provided, and when the instructions are executed by at least one computing device, the at least one computing device is prompted to execute the load prediction method for a wind turbine generator set described in any of the aforementioned embodiments. Computer-readable storage media include magnetic media such as floppy disks and tapes, optical media (including compact disk (CD) ROM and DVD ROM), magneto-optical media such as floppy disks, hardware devices such as ROM, RAM, and flash memory designed to store and execute program commands. The instructions may include language codes executable by a computer using an interpreter and machine language codes generated by a compiler.

[0086] By adopting the present disclosure, a real-time load prediction and life assessment system for wind turbines can be established based on a graph network model. Only load tests on some units are required to perform load prediction and life assessment on the entire field of units according to the topological structure of the graph network model. There is no need to perform "one machine one test" on the load of the units, which significantly improves the efficiency of load prediction for the entire field of units.

[0087] The processes, methods or algorithms disclosed herein may be transmitted to or implemented by a processing device, a controller or a computer, which may include any existing programmable electronic control unit or a dedicated electronic control unit. Similarly, the processes, methods or algorithms may be stored in a variety of forms as data and instructions that can be executed by a controller or a computer, including but not limited to information being permanently stored on a non-writable storage medium (such as a ROM device) and information being variably stored on a writable storage medium (such as a floppy disk, a tape, a CD, a RAM device, and other magnetic and optical media). The processes, methods or algorithms may also be implemented in a software executable object. Optionally, the processes, methods or algorithms may be implemented in whole or in part using suitable hardware components (such as an ASIC, an FPGA, a state machine, a controller or other hardware components or devices) or a combination of hardware components, software components and firmware components.

[0088] Although the present disclosure includes specific examples, it will be apparent to those of ordinary skill in the art that various changes in form and detail may be made in these examples without departing from the spirit and scope of the claims and their equivalents. The examples described herein will be considered to be descriptive only and not for limiting purposes. The description of the features or aspects in each example will be considered to be applicable to similar features or aspects in other examples. Suitable results may be obtained if the described techniques are performed in a different order, and / or if the components in the described systems, architectures, devices, or circuits are combined in different ways and / or replaced or supplemented with other components or their equivalents in the described systems, architectures, devices, or circuits. Therefore, the scope of the present disclosure is not limited by specific embodiments, but by the claims and their equivalents, and all variations within the scope of the claims and their equivalents will be interpreted as included in the present disclosure.

Claims

1. A method for predicting load of a wind turbine generator set, characterized in that: The load prediction method comprises: Based on the operation data of the plurality of wind turbine generator sets, determining a graph network with each wind turbine generator set in the plurality of wind turbine generator sets as a node; Based on the structural characteristics of the graph network, a wind turbine generator set for load testing is selected; Carry out load tests on selected wind turbines; Based on the load test data and operation data of the selected wind turbine generator sets, training and establishing a load prediction model for the plurality of wind turbine generator sets; Based on the operating data of the target wind turbine generator set, the load of the target wind turbine generator set is predicted by the load prediction model.

2. The method for predicting the load of a wind turbine generator set according to claim 1, characterized in that: The operation data includes at least one of rotation speed related data, power related data, acceleration related data, angle related data, wind speed related data, temperature related data and operation flag related data.

3. The method for predicting the load of a wind turbine generator set according to claim 1, characterized in that: The step of determining a graph network with each wind turbine generator set in the plurality of wind turbine generator sets as a node based on the operation data of the plurality of wind turbine generator sets comprises: Determining similarities between the wind turbines based on the operating data of the plurality of wind turbines; Pairs of nodes corresponding to similarities greater than a predetermined similarity threshold are connected to each other to construct the graph network.

4. The method for predicting the load of a wind turbine generator set according to claim 3, characterized in that: The predetermined similarity threshold is an average value of the similarities between the wind turbine generator sets.

5. The method for predicting the load of a wind turbine generator set according to claim 3, characterized in that: The step of selecting a wind turbine generator set for load testing based on the structural characteristics of the graph network comprises: For graph network blocks that are not connected to each other, determine the nodes with the largest number of connected edges respectively; The wind turbine generator set corresponding to the node with the largest number of connected edges in each graph network block is determined as the wind turbine generator set for load testing, or the wind turbine generator set with the highest annual average wind speed among multiple wind turbine generator sets corresponding to multiple nodes with the largest number of connected edges in each graph network block is determined as the wind turbine generator set for load testing.

6. The method for predicting the load of a wind turbine generator set according to claim 1, characterized in that: The input data of the load prediction model includes the operating data of the selected wind turbine generator set, and the target data of the load prediction model includes the load test data of the selected wind turbine generator set.

7. The method for predicting the load of a wind turbine generator set according to claim 6, characterized in that: The step of training and establishing a load prediction model for the plurality of wind turbines based on the load test data and operation data of the selected wind turbines comprises: Determining a predicted load of the selected wind turbine generator set based on the adjacency matrix of the graph network, the input data and a preset weight of the load prediction model; In response to a difference between the load test data and the predicted load being greater than a predetermined error threshold, adjusting the size of the preset weight and re-determining the predicted load of the selected wind turbine generator set; In response to a difference between the load test data and the predicted load being less than or equal to the predetermined error threshold, determining a current preset weight as a final weight of the load prediction model; Based on the final weight, the load prediction model is established.

8. A load prediction device for a wind turbine generator set, characterized in that: The load prediction device comprises: A graph network determining unit, based on the operation data of the plurality of wind turbine generator sets, determines a graph network with each wind turbine generator set in the plurality of wind turbine generator sets as a node; A selection unit, which selects a wind turbine generator set for load testing based on the structural characteristics of the graph network; A load test unit, which performs load test on the selected wind turbine generator set; A prediction model determination unit, which trains and establishes load prediction models for the plurality of wind turbines based on load test data and operation data of the selected wind turbines; The prediction unit predicts the load of the target wind turbine generator set by using the load prediction model based on the operation data of the target wind turbine generator set.

9. A computing system comprising at least one computing device and at least one storage device storing instructions, characterized in that: When the instructions are executed by the at least one computing device, the at least one computing device is prompted to perform the load prediction method for a wind turbine generator set according to any one of claims 1-7.

10. A computer-readable storage medium storing instructions, characterized in that: When the instructions are executed by at least one computing device, the at least one computing device is prompted to execute the load prediction method for a wind turbine generator set according to any one of claims 1-7.

Images