Reliability analysis method and device for hybrid tower fans based on multi-fidelity numerical model

Abstract

The present invention discloses a reliability analysis method and device for a mixed-tower wind turbine based on a multi-fidelity numerical model, the method comprising the following steps: establishing two or more mixed-tower wind turbine dynamic models; inputting input data into each mixed-tower wind turbine dynamic model to obtain extreme response result data of different precisions; using low-precision response results to train different types of machine learning models to obtain multiple first proxy models; inputting high-precision response results into each first proxy model to obtain a predicted value, adjusting each first proxy model according to the predicted result, and training a second proxy model by combining each predicted result; obtaining initial sample points and inputting them into the second proxy model to obtain an extreme response result; and calculating reliability according to the extreme response result obtained by the second proxy model. The present invention can realize reliability analysis of a mixed-tower wind turbine under multiple failure modes, and improve the efficiency, accuracy and credibility of the analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine reliability analysis, and in particular to a mixed-tower wind turbine reliability analysis method and device based on a multi-fidelity numerical model. Background Art

[0002] Wind energy is a clean and renewable energy source, and its main form of utilization is wind power generation. With the continuous advancement of wind power generation technology, the unit capacity and tower height of wind turbines are also increasing. A mixed tower wind turbine refers to a wind turbine in which the tower is formed by a mixture of concrete and steel. The lower part of this type of tower is a prestressed reinforced concrete structure and the upper part is a steel tube structure. Therefore, the tower has high rigidity and good stability. However, in a mixed tower wind turbine, due to the particularity of its tower structure, it will face complex nonlinear problems when performing dynamic analysis, so the reliability analysis of this type of wind turbine is more difficult.

[0003] When conducting reliability analysis of mixed-tower wind turbines, the existing technology usually adopts the modal superposition method for multi-body dynamic modeling, and uses this model to perform linear time history analysis and reliability assessment. However, this type of method can only be applied to the analysis of traditional steel tower structures and linear material constitutive structures, and it is difficult to reveal the nonlinear characteristics of mixed-tower structures, resulting in a large deviation between the dynamic analysis results and the actual situation. And due to the inherent limitations of the modal superposition method, it is impossible to obtain accurate stress and strain of the structure. Therefore, when conducting reliability assessment, usually only the displacement or acceleration of the top of the tower can be used as a single discriminant index, which will further reduce the reliability of the assessment. In addition, the above-mentioned method of reliability analysis based solely on the dynamic model method requires a large and accurate number of samples to ensure the accuracy and credibility of the analysis results, and the numerical model of the mixed-tower wind turbine is often difficult to meet the requirements of reliability analysis. Therefore, it will take a lot of time and computing resources to conduct reliability assessment, and it is difficult to balance analysis efficiency and accuracy. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a hybrid tower wind turbine reliability analysis method and device based on a multi-fidelity numerical model with simple operation, low cost, high analysis efficiency, accuracy and reliability, which can quickly realize accurate reliability analysis of the hybrid tower wind turbine under multiple failure modes.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0006] A reliability analysis method for a mixed-tower fan based on a multi-fidelity numerical model comprises the following steps:

[0007] Two or more mixed-tower wind turbine dynamic models with different fidelity are established according to the structural parameters of the target wind turbine;

[0008] The sample sets with different input variables are used as input data and respectively input into each of the mixed-tower wind turbine dynamic models to respectively perform nonlinear time-history analysis of the tower under wind field coupling to obtain extreme value response result data with different accuracies, wherein the extreme value response result data includes any one or more data of tower top displacement, acceleration, structural stress, and strain;

[0009] Using low-precision response results whose precision is lower than a preset threshold in the extreme value response result data to train different types of machine learning models respectively, to obtain multiple first proxy models;

[0010] Inputting high-precision response results with precision higher than a preset threshold in the extreme value response result data into each of the first proxy models to obtain a prediction value, adjusting each of the first proxy models according to the prediction result, and training a second proxy model based on the prediction results of each of the first proxy models;

[0011] Obtaining initial sample points and inputting them into the second proxy model to obtain extreme value response result data of each sample point;

[0012] Reliability is calculated based on the extreme value response result data obtained by the second proxy model.

[0013] Furthermore, the hybrid tower wind turbine dynamic model includes a low-fidelity dynamic model with a fidelity lower than a preset threshold and a high-fidelity dynamic model with a fidelity higher than a preset threshold. By constructing a multi-body dynamic model including an impeller and a cabin in Simpack and creating a local server port, and constructing a nonlinear fiber beam unit model of a tower in OpenSees and creating a local server port, the dynamic response results of the impeller and the cabin of the current analysis step are received from Simpack through a joint simulation interface program defined in Simulink, and the dynamic response results of the tower of the current analysis step are output to the Simpack server port, and the dynamic response results of the tower are received from OpenSees, and the dynamic response results of the impeller and the cabin are output to the OpenSees server port, so as to realize the joint simulation of Simpack and OpenSees to obtain the low-fidelity dynamic model; by constructing a finite element model of the tower in Abaqus and constructing a multi-body dynamic model including an impeller and a cabin in Simpack, the high-fidelity dynamic model is obtained through the joint simulation of Simpack and Abaqus.

[0014] Furthermore, the steps of constructing the low-fidelity kinetic model include:

[0015] On the Simpack side, a multi-body dynamics model is established based on the structural parameters and material parameters of the impeller and nacelle. At the same time, a servo control system is defined and the wind field is defined as the environmental load of the impeller. The connection node between the nacelle and the top of the tower is defined as an interface node to output the dynamic response results of the impeller and nacelle, and receive the dynamic response results of the tower. Set the local server port, start Simpack as a server, and wait for TCP / IP communication access.

[0016] On the OpenSees side, a nonlinear fiber beam unit model of the tower is constructed according to the tower structure parameters and material parameters, the wind field is defined as the environmental load of the tower, and the connection node between the cabin and the top of the tower is defined as an interface node to output the dynamic response results of the tower, while receiving the dynamic response results of the impeller and cabin, and creating a local server port, starting OpenSees as a server, and waiting for TCP / IP communication access;

[0017] The interface program of joint simulation is defined in Simulink to realize the simulation data transmission and analysis step time coordination between Simulink and OpenSees. The Simpack joint simulation interface program is used to receive the dynamic response results of the impeller and nacelle of the current analysis step, and output the dynamic response results of the tower to the Simpack server port. The OpenSees joint simulation interface program is used to receive the dynamic response results of the tower of the current analysis step, and output the dynamic response results of the impeller and nacelle to the OpenSees server port.

[0018] Furthermore, the steps of constructing the high-fidelity kinetic model include:

[0019] In Abaqus, a three-dimensional model of the tower is established according to the tower structure parameters. After dividing the solid units and defining the finite element parameters, a nonlinear finite element model of the tower is established. The connection node between the cabin and the top of the tower is defined as an interface node to output the dynamic response results of the tower and receive the dynamic response results of the impeller and cabin at the same time.

[0020] In Simpack, a multi-body dynamics model is established based on the impeller, cabin structural parameters and material parameters. The servo control system and tower structure type are defined as Abaqus joint simulation entities, and the interface nodes are associated. The connection node between the cabin and the top of the tower is defined as the interface node.

[0021] The wind field is defined as the environmental load, and a joint simulation time-integrated solver is set up to achieve the simulation data transmission and analysis step time coordination between the time-integrated solver in Simpack and the nonlinear finite element solver in Abaqus.

[0022] Furthermore, the step of inputting high-precision response results in the extreme value response result data whose precision is higher than a preset threshold into each of the first proxy models to obtain a prediction value, adjusting each of the first proxy models according to the prediction result, and training the second proxy model based on the prediction results of each of the first proxy models includes:

[0023] The high-precision response result Y HIGH As input variables, they are input into each first proxy model In the above example, we get multiple predicted values.

[0024] Calculate each predicted value And high-precision response result Y HIGH The mean square error E j ;

[0025] According to the calculated mean square error E j For the model parameters θ j Perform optimization to update model parameters;

[0026] According to the error of each first proxy model, adjust the weight ω of each first proxy model j , using each ω j Form a comprehensive forecast value The calculation expressions are:

[0027]

[0028] in, represents the sum of errors of all first agent models;

[0029] Through multiple iterations, the parameters of each first proxy model are continuously optimized to make the model gradually approach the high-precision response result, and finally the second proxy model M is obtained. HIGH .

[0030] Furthermore, the initial sample points are generated based on the probability density function of the input variables, and the steps include: randomly selecting a certain number of sample points from the probability density function of each input variable by a random sampling method, combining the sample points of different input variables into a multidimensional sample set X 1 ,X 2 ,X 3 ,…,X n As the initial sample point, n represents the number of sample sets.

[0031] Further, the input variables include: any one or more of environmental input variables, model structure input variables, and model material input variables, the environmental input variables include the average wind speed and / or turbulence intensity of the wind field, the model structure input variables include the hybrid tower height and / or the hybrid tower diameter, and the model material input variables include the elastic modulus of concrete and / or the elastic modulus of steel; a sample point is randomly selected from each input variable, and sample points of different input variables are combined to obtain a sample set and the execution is repeated to obtain multiple groups of sample variable sets X 1 ,X 2 ,X 3 ,…,X m , where m represents the number of sample sets, and the sample variable set X 1 ,X 2 ,X 3 ,…,X m The system is divided into multiple parts according to a specified ratio to be input into each mixed-tower wind turbine dynamics model respectively.

[0032] Furthermore, the calculating reliability of the extreme value response result data obtained according to the second proxy model includes:

[0033] Statistical analysis is performed on the extreme value response result data, and the preset structural failure criterion Y C The failure probability p of the structure is calculated according to the following formula F :

[0034]

[0035] In the formula, I(Y i >Y C ) is the indicator function, when Y i >Y C The value is 1 when it is, otherwise it is 0;

[0036] The reliability index β is calculated based on the failure probability, and the calculation expression is:

[0037] β=-Φ -1 (p F )

[0038] where Φ is the cumulative distribution function of the standard normal distribution.

[0039] A computer device comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0040] A computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0041] Compared with the prior art, the advantages of the present invention are: the present invention uses deep neural networks, supervised learning and integrated learning, and uses the analysis results of multi-fidelity numerical models to train a high-precision proxy model, and uses the extreme response result data of the high-precision proxy model to calculate the reliability. It can fully consider the nonlinear characteristics of the mixed-tower wind turbine and can easily extract the stress and strain distribution of the structure, thereby realizing the reliability analysis of the mixed-tower wind turbine under multiple failure modes, greatly improving the stability, accuracy and credibility of the reliability analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the implementation process of the mixed-tower wind turbine reliability analysis method based on the multi-fidelity numerical model in this embodiment.

[0043] Figure 2 A schematic diagram of the process of implementing low-fidelity dynamics modeling in this embodiment.

[0044] Figure 3 A schematic diagram of the process of implementing high-fidelity dynamic modeling in this embodiment. DETAILED DESCRIPTION

[0045] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0046] As shown in the disclosure of the present invention, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not specifically refer to the singular, but may also include the plural. The words "first", "second" and similar words used in the disclosure of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, the words "include" or "comprise" and the like mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0047] Taking into account that mixed-tower wind turbines have different failure modes, it is difficult to accurately characterize the nonlinear characteristics of mixed-tower structures using a single model. The present invention aims at reliability analysis of mixed-tower wind turbines. By establishing two or more mixed-tower wind turbine dynamic models with different fidelity, each mixed-tower wind turbine dynamic model first performs nonlinear time-history analysis of the tower to obtain extreme response result data of different precisions, and then divides the extreme response result data into low-precision data and high-precision data. The low-precision data is used to train multiple machine learning models to form low-precision proxy models (first proxy models) of multiple mixed-tower wind turbines. The proxy models are used to complete The extreme response of the hybrid tower wind turbine is predicted, and the high-precision data is used to adjust and optimize the parameters of the low-precision proxy model. The prediction results of each low-precision proxy model are combined to train a high-precision proxy model (second proxy model) of the hybrid tower wind turbine. Finally, the extreme response result data of the high-precision model is used to calculate the reliability. The combination of deep neural networks, supervised learning and ensemble learning can fully explore the nonlinear characteristics of the hybrid tower wind turbine, greatly improve the accuracy and credibility of the reliability analysis results, and achieve both analysis accuracy and efficiency, and realize efficient and accurate reliability analysis of the hybrid tower wind turbine under multiple failure modes.

[0048] The following is a detailed description with reference to specific embodiments.

[0049] like Figure 1 As shown, the steps of the mixed-tower wind turbine reliability analysis method based on the multi-fidelity numerical model in this embodiment include:

[0050] Step S01. Establish two or more mixed-tower wind turbine dynamic models with different fidelity according to the structural parameters of the target wind turbine.

[0051] In this embodiment, the mixed-tower wind turbine dynamics model includes a low-fidelity dynamics model with a fidelity lower than a preset threshold and a high-fidelity dynamics model with a fidelity higher than a preset threshold, that is, it includes two types of low-fidelity dynamics models and high-fidelity dynamics models. Of course, more mixed-tower wind turbine dynamics models with different fidelity can be constructed according to actual needs, so as to comprehensively utilize dynamic models of different fidelity to realize nonlinear time-history analysis of the tower under wind field coupling, taking into account both analysis efficiency and analysis accuracy.

[0052] In the traditional wind turbine modeling method, Simpack software is usually used for modeling, but Simpack software cannot realize nonlinear simulation analysis of the tower, and the mixed tower structure has strong nonlinear properties. In view of the above problems, this embodiment decomposes the wind turbine system into two independent and mutually coupled subsystems: one set corresponds to the impeller and the cabin system, and the other set corresponds to the tower. The two subsystems are independently modeled through different software, that is, the impeller, the cabin system and the tower are modeled in different software, and the data interaction of the interface nodes is realized to achieve mutual coupling, so as to accurately capture the interaction between the impeller, the cabin system and the tower, so that the model can fully reflect the dynamic response of the wind turbine in the actual working environment, thereby realizing the nonlinear modeling of the mixed tower.

[0053] For the low-fidelity numerical model, this embodiment specifically constructs a multi-body dynamics model including an impeller and a cabin in Simpack and creates a local server port, and constructs a nonlinear fiber beam unit model of the tower in OpenSees and creates a local server port, receives the dynamic response results of the impeller and cabin of the current analysis step from Simpack through the joint simulation interface program defined in Simulink, and outputs the dynamic response results of the tower to the Simpack server port, and receives the dynamic response results of the tower of the current analysis step from OpenSees, and outputs the dynamic response results of the impeller and cabin to the OpenSees server port, so as to realize the joint simulation of Simpack and OpenSees to obtain a low-fidelity dynamic model. The low-fidelity numerical model constructed by the above method can realize the nonlinear finite element simulation of the tower and efficiently obtain a large amount of nonlinear dynamic time-history analysis data.

[0054] Simpack and OpenSees software are two softwares with different functions, among which OpenSees is a software for modeling and analysis of civil engineering structures, which can perform nonlinear response analysis of concrete and steel structures, but is not suitable for aerodynamic analysis. Simpack is a software for multi-body system simulation, which can realize wind field and blade aerodynamic analysis, but cannot realize nonlinear modeling of reinforced concrete structures. In this embodiment, by using Simpack to model the impeller and cabin of the wind turbine, the geometric nonlinearity and material nonlinear characteristics of the tower can be modeled in the OpenSees software through nonlinear fiber beam units. At the same time, the nonlinear fiber beam unit can also output the stress and strain data of the structure to realize the modeling of the tower. Simulink software is used as a bridge between the two softwares, and data communication and joint simulation between Simpack and OpenSees software are realized through Simulink software. Through the TCP / IP communication protocol, the dynamic response data of the impeller, cabin system and tower are transmitted in real time at the sub-analysis step time node, thereby combining Simpack, Simulink and OpenSees to realize efficient nonlinear time history analysis of mixed tower wind turbines.

[0055] In a specific application embodiment, the steps of constructing a low-fidelity kinetic model include:

[0056] On the Simpack side, a multi-body dynamics model is established based on the structural parameters and material parameters of the impeller and nacelle. At the same time, a servo control system is defined and the wind field is defined as the environmental load of the impeller. The connection node between the nacelle and the top of the tower is defined as an interface node to output the dynamic response results of the impeller and nacelle, and receive the dynamic response results of the tower. Set the local server port, start Simpack as a server, and wait for TCP / IP communication access.

[0057] On the OpenSees side, a nonlinear fiber beam unit model of the tower is constructed according to the tower structure parameters and material parameters, the wind field is defined as the environmental load of the tower, and the connection node between the cabin and the top of the tower is defined as an interface node to output the dynamic response results of the tower, while receiving the dynamic response results of the impeller and cabin, and creating a local server port, starting OpenSees as a server, and waiting for TCP / IP communication access;

[0058] The interface program of joint simulation is defined in Simulink to realize the simulation data transmission and analysis step time coordination between Simulink and OpenSees. The Simpack joint simulation interface program is used to receive the dynamic response results of the impeller and nacelle of the current analysis step, and output the dynamic response results of the tower to the Simpack server port. The OpenSees joint simulation interface program is used to receive the dynamic response results of the tower of the current analysis step, and output the dynamic response results of the impeller and nacelle to the OpenSees server port.

[0059] The following is an example of using the above method to construct a low-fidelity dynamics model in a specific application embodiment. Figure 2 As shown, the detailed steps are:

[0060] Step 1. Build a multibody dynamics model in Simpack, create a local server port, and wait for TCP / IP communication.

[0061] Step 1.1, establish a multi-body dynamic model based on the impeller and nacelle structural parameters (such as impeller diameter, airfoil data and other parameters) and material parameters (such as elastic modulus and density of blades and other parameters), and define the servo control system, such as the pitch control mechanism of the blades and the torque control mechanism of the generator.

[0062] Step 1.2: Define the wind field as the environmental load of the impeller, including parameters such as average wind speed, turbulence intensity, and wind speed time series.

[0063] Step 1.3: Define the connection node between the nacelle and the top of the tower as an interface node to output the dynamic response results of the impeller and nacelle, and receive the dynamic response results of the tower.

[0064] Step 1.4: Define the time integration solver, including parameters such as solver type and analysis step size.

[0065] Step 1.5, set the local server port, start Simpack as a server, and wait for TCP / IP communication access.

[0066] Step 2: Build a nonlinear fiber beam unit model of the tower in OpenSees, create a local server port, and wait for TCP / IP communication.

[0067] Step 2.1: Establish a nonlinear beam fiber beam unit model based on tower structure parameters (such as diameter, wall thickness, prestressing and other parameters) and material parameters (such as elastic modulus and density of concrete and steel and other parameters).

[0068] Step 2.2: Define the wind field as the environmental load of the tower, such as average wind speed, turbulence intensity, wind speed time history and other parameters.

[0069] Step 2.3: Define the connection node between the nacelle and the top of the tower as an interface node to output the dynamic response results of the tower and receive the dynamic response results of the impeller and nacelle at the same time.

[0070] Step 2.4: Define the nonlinear finite element solver, including parameters such as solver type and analysis step size.

[0071] Step 2.5, set the local server port, start OpenSees as a server, and wait for TCP / IP communication access.

[0072] Step 3: Define the interface program of joint simulation in Simulink to realize the simulation data transmission and analysis step time coordination between different servers, and ensure the consistency of the simulation process of each subsystem. Among them, the Simpack joint simulation interface program is mainly responsible for receiving the dynamic response results of the impeller and nacelle of the current analysis step, and outputting the dynamic response results of the tower to the Simpack server port; the OpenSees joint simulation interface program is mainly responsible for receiving the dynamic response results of the tower of the current analysis step, and outputting the dynamic response results of the impeller and nacelle to the OpenSees server port.

[0073] A low-fidelity mixed-tower wind turbine dynamic model can be constructed in the above manner. This model can be used to efficiently perform nonlinear time-history analysis of the tower under wind field coupling, thereby efficiently obtaining a large amount of nonlinear dynamic time-history analysis data.

[0074] Regarding the high-fidelity dynamic model, this embodiment specifically constructs a finite element model of the tower in Abaqus and a multi-body dynamic model including the impeller and the nacelle in Simpack, and obtains a high-fidelity dynamic model through joint simulation of Simpack and Abaqus. It can realize refined three-dimensional modeling and nonlinear finite element simulation of the tower, and obtain accurate nonlinear dynamic time-history analysis data.

[0075] In a specific application embodiment, the steps of constructing a high-fidelity kinetic model include:

[0076] In Abaqus, a three-dimensional model of the tower is established according to the tower structure parameters. After dividing the solid units and defining the finite element parameters, a nonlinear finite element model of the tower is established. The connection node between the cabin and the top of the tower is defined as an interface node to output the dynamic response results of the tower and receive the dynamic response results of the impeller and cabin at the same time.

[0077] In Simpack, a multi-body dynamics model is established based on the impeller, cabin structural parameters and material parameters. The servo control system and tower structure type are defined as Abaqus joint simulation entities, and the interface nodes are associated. The connection node between the cabin and the top of the tower is defined as the interface node.

[0078] The wind field is defined as the environmental load, and a joint simulation time-integrated solver is set up to achieve the simulation data transmission and analysis step time coordination between the time-integrated solver in Simpack and the nonlinear finite element solver in Abaqus.

[0079] As finite element analysis software, Abaqus can support the simulation of complex nonlinear problems such as nonlinear material behavior, large deformation, and contact problems, but it is not suitable for aerodynamic analysis. The joint simulation engine in Simpack software can be used to realize the joint simulation between Simpack and Abaqus software. In this embodiment, the impeller, the cabin system, and the tower are taken as two subsystems, and the mixed tower is finely modeled in the Abaqus software. The interface node is defined as the connection node between the cabin and the top of the tower, so that Simpack and Abaqus can be combined to form a high-fidelity dynamic model to realize accurate nonlinear time-history analysis of the mixed-tower wind turbine.

[0080] The following is an example of using the above method to construct a high-fidelity dynamic model (high-fidelity wind turbine numerical model) in a specific application embodiment. Figure 3 As shown, the detailed steps are:

[0081] Step 1: Construct a refined finite element model of the tower in Abaqus.

[0082] Step 1.1: Establish a refined three-dimensional model based on the tower structure parameters (such as diameter, wall thickness, height, etc.).

[0083] Step 1.2: Divide the entity elements, define the finite element parameters (such as nonlinear material constitutive, prestress, component contact properties, etc.), and establish a nonlinear finite element model.

[0084] Step 1.3: Define the connection node between the nacelle and the top of the tower as an interface node to output the dynamic response results of the tower and receive the dynamic response results of the impeller and nacelle at the same time.

[0085] Step 1.4: Convert the nonlinear finite element model of the tower into a "*.inp" file that can be read by Simpack.

[0086] Step 2: Build a multi-body dynamics model in Simpack.

[0087] Step 2.1: Establish a multi-body dynamics model based on the impeller and nacelle structural parameters (such as impeller diameter, airfoil data, etc.) and material parameters (such as elastic modulus and density of blades, etc.). At the same time, define the servo control system, which mainly includes the pitch control mechanism of the blades and the torque control mechanism of the generator.

[0088] Step 2.2: Define the tower structure type as an Abaqus co-simulation entity and import the "*.inp" file.

[0089] Step 2.3: Associate the interface node and define the connection node between the cabin and the top of the tower as the interface node.

[0090] Step 2.4: Define the wind field as an environmental load, including parameters such as average wind speed, turbulence intensity, and wind speed time series.

[0091] Step 2.5: Set up the joint simulation time integration solver to realize the simulation data transmission and analysis step time coordination between the Simpack time integration solver and the Abaqus nonlinear finite element solver.

[0092] A high-fidelity dynamic model can be constructed in the above manner, and the model can be used to realize accurate nonlinear time-history analysis of hybrid tower wind turbines.

[0093] Step S02. Using sample sets with different input variables as input data and inputting them into each mixed-tower wind turbine dynamics model respectively, so as to perform nonlinear time history analysis of the tower under wind field coupling, and obtain extreme value response result data with different accuracies.

[0094] Optionally, the input variables may include environmental input variables, model structure input variables, model material input variables, etc. The environmental input variables may include the average wind speed and turbulence intensity of the wind farm, the model structure input variables may include the hybrid tower height, the hybrid tower diameter, etc., and the model material input variables may include the elastic modulus of concrete, the elastic modulus of steel, etc.

[0095] To form a sample set of input variables, a sample point can be randomly selected from each input variable, and the sample points of different input variables can be combined to obtain a sample set and repeated to obtain multiple sets of sample variable sets X 1 ,X 2 ,X 3 ,…,X m , where m represents the number of sample sets, and the sample variable set X 1 ,X 2 ,X 3 ,…,X mBy dividing it into multiple parts according to a specified ratio and inputting them into each mixed-tower wind turbine dynamic model respectively, extreme response result data with different accuracies can be obtained. The extreme response result data may include tower top displacement, acceleration, structural stress, strain and other data.

[0096] For example, the mixed tower wind turbine dynamics model includes two models, a low-fidelity dynamics model and a high-fidelity dynamics model. The sample variable set formed in the above manner can be divided into two parts X according to a ratio of 10:1 (or other ratios a:b, where a>>b can be configured). LOW,a , X HIGH,b , a+b=m, and then the sample set X LOW,a Provided to the low-fidelity dynamics model, X HIGH,b Provided to high-fidelity dynamic models, through nonlinear dynamic time-history analysis, a large amount of low-precision extreme response result data Y can be obtained. LOW and a small amount of high-precision extreme response result data Y HIGH .

[0097] Step S03. Use low-precision response results in the extreme value response result data whose precision is lower than a preset threshold to train different types of machine learning models respectively, and obtain multiple first proxy models.

[0098] Nonlinear response prediction problems usually involve complex input and output relationships, and a single model is difficult to capture these complex nonlinear relationships. In this embodiment, the input data set used for training includes various variables such as environmental variables, structural variables, and material variables. At the same time, extreme response result data such as displacement, acceleration, stress, and strain are used as target variables for supervised learning. Through ensemble learning, multiple learning models are combined to form multiple first proxy models (low-precision proxy models). Each model can capture different aspects of the data. Each proxy model can simultaneously predict multiple response results of the structure, analyze the failure model of the structure from multiple dimensions, and thus more comprehensively evaluate the reliability and safety of the structure.

[0099] Specifically, given a large number of low-precision data sets (input variable X LOW And the extreme response result data Y LOW ), through different types of machine learning models (such as deep neural networks, decision trees, support vector machines, etc.), multiple low-precision proxy models can be trained k is the number of low-precision proxy models. Each proxy model can not only predict the displacement and acceleration extreme responses of the mixed-tower wind turbine, but also predict the stress and strain extreme responses of the structure, thereby realizing the reliability analysis of the mixed-tower wind turbine under multiple failure modes, greatly improving the accuracy and credibility of the reliability analysis results.

[0100] Step S04. Input high-precision response results in the extreme value response result data with accuracy higher than a preset threshold into each first proxy model to obtain a prediction value, adjust each first proxy model according to the prediction result, and train the second proxy model based on the prediction results of each first proxy model.

[0101] In this embodiment, a high-precision data set (input variable X HIGH ) input different first proxy models (low-precision proxy models), compare the predicted values ​​with the high-precision extreme response result data, adjust and optimize the low-fidelity model, and then integrate the predicted results of each first proxy model (low-precision proxy model) to train a second proxy model (high-precision proxy model). A high-precision proxy model can be obtained, which can better describe complex nonlinear relationships.

[0102] Step S401. The high-precision response result Y HIGH As input variables, they are input into each first proxy model In the above example, we get multiple predicted values. Right now:

[0103]

[0104] Step S402. Calculate each predicted value And high-precision response result Y HIGH The mean square error E j , the calculation expression is:

[0105]

[0106] Step S403. Based on the calculated mean square error E j For the model parameters θ j Perform optimization to update the model parameters.

[0107] Alternatively, the model parameters θ can be quantized using the gradient descent method as follows: j Perform optimization to update the model parameters:

[0108]

[0109] Where η is the learning rate;

[0110] Step S404: Adjust the weight ω of each first proxy model according to the error of each first proxy model. j , using each ω j Form a comprehensive forecast value The calculation expressions are:

[0111]

[0112] in, represents the sum of errors of all first-surrogate models.

[0113] Step S405. Re-execute steps S401 to S404 to continuously optimize the parameters of each first proxy model through multiple iterations, so that the model gradually approaches the high-precision response result Y HIGH , and finally get the second proxy model M HIGH , that is, a high-precision proxy model.

[0114] Optionally, a weighted voting mechanism may be used to integrate the prediction results of multiple first proxy models to train a second proxy model (a high-precision proxy model).

[0115] Step S05: Obtain initial sample points and input them into the second proxy model to obtain extreme value response result data of each sample point.

[0116] Optionally, the initial sample points may be generated based on the probability density function of the input variables, the steps comprising: randomly selecting a certain number of sample points from the probability density function of each input variable by a random sampling method, combining the sample points of different input variables into a multidimensional sample set X 1 ,X 2 ,X 3 ,…,X n As the initial sample point, n represents the number of sample sets. The probability density function of the input variable can be determined using historical data, experimental data, or theoretical distribution. For example, the distribution type can include normal distribution, lognormal distribution, uniform distribution, Weibull distribution, etc.

[0117] After generating initial sample points based on the probability density function of the input variables to obtain multiple sets of sample sets representing possible input conditions, the generated sample sets are input into the second proxy model (high-precision proxy model), and then the second proxy model (high-precision proxy model) predicts the initial sample points to obtain the extreme value response result data Y of each sample point. 1 ,Y 2 ,Y 3 ,…,Y n .

[0118] Step S06: Calculate reliability based on the extreme value response result data obtained by the second proxy model.

[0119] Optionally, the extreme value response result data Y of each sample point obtained by the second proxy model can be 1 ,Y 2 ,Y 3 ,…,Y nAccording to the structural failure criterion, it is judged whether the structure begins to fail. The failure probability of the mixed tower fan structure can be calculated by combining the judgment results of the failure status of each sample point, and then the reliability index can be calculated based on the failure probability. The above structural failure criterion can be set according to the status of each parameter when the mixed tower fan begins to fail.

[0120] Specifically, the step of calculating the reliability according to the extreme value response result data obtained by the second proxy model may be:

[0121] Statistical analysis is performed on the extreme value response result data, and the preset structural failure criterion Y C The failure probability p of the structure is calculated according to the following formula F :

[0122]

[0123] In the formula, I(Y i >Y C ) is the indicator function, when Y i >Y C The value is 1 when it is, otherwise it is 0;

[0124] The reliability index β is calculated based on the failure probability, and the calculation expression is:

[0125] β=-Φ -1 (p F ) (7)

[0126] where Φ is the cumulative distribution function of the standard normal distribution.

[0127] Through the above-mentioned method, the present invention can obtain a high-precision proxy model by training the analysis results of the multi-fidelity numerical model through deep neural network, supervised learning and integrated learning, effectively improving the prediction stability and accuracy, thereby realizing accurate and efficient reliability analysis of mixed-tower wind turbines. Compared with traditional methods, the present invention can fully consider the nonlinear characteristics of mixed-tower wind turbines and can conveniently extract the stress and strain distribution of the structure, thereby realizing the reliability analysis of mixed-tower wind turbines under multiple failure modes, greatly improving the accuracy and credibility of the reliability analysis results.

[0128] This embodiment further provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0129] It is understandable that the above method of this embodiment can be executed by a single device, such as a computer or server, etc., and can also be applied to a distributed scenario and completed by multiple devices in cooperation with each other. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps in the above method of this embodiment, and multiple devices interact to complete the above method. The processor can be implemented in the form of a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., for executing related programs to implement the above method of this embodiment. The memory can be implemented in the form of a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device. The memory can store an operating system and other applications. When the above method of this embodiment is implemented by software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0130] This embodiment further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0131] Those skilled in the art will appreciate that the above-mentioned embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0132] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A reliability analysis method for a mixed-tower wind turbine based on a multi-fidelity numerical model, characterized in that the steps include: Two or more mixed-tower wind turbine dynamic models with different fidelity are established according to the structural parameters of the target wind turbine; The sample sets with different input variables are used as input data and respectively input into each of the mixed-tower wind turbine dynamic models to respectively perform nonlinear time-history analysis of the tower under wind field coupling to obtain extreme value response result data with different accuracies, wherein the extreme value response result data includes any one or more data of tower top displacement, acceleration, structural stress, and strain; Using low-precision response results whose precision is lower than a preset threshold in the extreme value response result data to train different types of machine learning models respectively, to obtain multiple first proxy models; Inputting high-precision response results with precision higher than a preset threshold in the extreme value response result data into each of the first proxy models to obtain a prediction value, adjusting each of the first proxy models according to the prediction result, and training a second proxy model based on the prediction results of each of the first proxy models; Obtaining initial sample points and inputting them into the second proxy model to obtain extreme value response result data of each sample point; Reliability is calculated based on the extreme value response result data obtained by the second proxy model.

2. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to claim 1 is characterized in that: The hybrid tower wind turbine dynamics model includes a low-fidelity dynamics model with a fidelity lower than a preset threshold and a high-fidelity dynamics model with a fidelity higher than a preset threshold. By constructing a multi-body dynamics model including an impeller and a cabin in Simpack and creating a local server port, and by constructing a nonlinear fiber beam unit model of a tower in OpenSees and creating a local server port, the dynamic response results of the impeller and the cabin of the current analysis step are received from Simpack through a joint simulation interface program defined in Simulink, and the dynamic response results of the tower of the current analysis step are output to the Simpack server port, and the dynamic response results of the tower are received from OpenSees, and the dynamic response results of the impeller and the cabin are output to the OpenSees server port, so as to realize the joint simulation of Simpack and OpenSees to obtain the low-fidelity dynamics model; by constructing a finite element model of the tower in Abaqus and a multi-body dynamics model including an impeller and a cabin in Simpack, the high-fidelity dynamics model is obtained through the joint simulation of Simpack and Abaqus.

3. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to claim 2 is characterized in that: The steps of constructing the low-fidelity kinetic model include: On the Simpack side, a multi-body dynamics model is established based on the structural parameters and material parameters of the impeller and nacelle. At the same time, a servo control system is defined and the wind field is defined as the environmental load of the impeller. The connection node between the nacelle and the top of the tower is defined as an interface node to output the dynamic response results of the impeller and nacelle, and receive the dynamic response results of the tower. Set the local server port, start Simpack as a server, and wait for TCP / IP communication access. On the OpenSees side, a nonlinear fiber beam unit model of the tower is constructed according to the tower structure parameters and material parameters, the wind field is defined as the environmental load of the tower, and the connection node between the cabin and the top of the tower is defined as an interface node to output the dynamic response results of the tower, while receiving the dynamic response results of the impeller and cabin, and creating a local server port, starting OpenSees as a server, and waiting for TCP / IP communication access; The interface program of joint simulation is defined in Simulink to realize the simulation data transmission and analysis step time coordination between Simulink and OpenSees. The Simpack joint simulation interface program is used to receive the dynamic response results of the impeller and nacelle of the current analysis step, and output the dynamic response results of the tower to the Simpack server port. The OpenSees joint simulation interface program is used to receive the dynamic response results of the tower of the current analysis step, and output the dynamic response results of the impeller and nacelle to the OpenSees server port.

4. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to claim 2 is characterized in that: The steps of constructing the high-fidelity kinetic model include: In Abaqus, a three-dimensional model of the tower is established according to the tower structure parameters. After dividing the solid units and defining the finite element parameters, a nonlinear finite element model of the tower is established. The connection node between the cabin and the top of the tower is defined as an interface node to output the dynamic response results of the tower and receive the dynamic response results of the impeller and cabin at the same time. In Simpack, a multi-body dynamics model is established based on the impeller, cabin structural parameters and material parameters. The servo control system and tower structure type are defined as Abaqus joint simulation entities, and the interface nodes are associated. The connection node between the cabin and the top of the tower is defined as the interface node. The wind field is defined as the environmental load, and a joint simulation time-integrated solver is set up to achieve the simulation data transmission and analysis step time coordination between the time-integrated solver in Simpack and the nonlinear finite element solver in Abaqus.

5. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to claim 1 is characterized in that: The step of inputting high-precision response results in the extreme value response result data whose precision is higher than a preset threshold into each of the first proxy models to obtain a prediction value, adjusting each of the first proxy models according to the prediction result, and training the second proxy model based on the prediction results of each of the first proxy models comprises: The high-precision response result Y HIGH As input variables, they are input into each first proxy model In the above example, we get multiple predicted values. Right now Calculate each predicted value And high-precision response result Y HIGH The mean square error E j ; According to the calculated mean square error E j For the model parameters θ j Perform optimization to update model parameters; According to the error of each first proxy model, adjust the weight ω of each first proxy model j , using each ω j Form a comprehensive forecast value The calculation expressions are: in, represents the sum of errors of all first agent models; Through multiple iterations, the parameters of each first proxy model are continuously optimized to make the model gradually approach the high-precision response result, and finally the second proxy model M is obtained. HIGH .

6. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to any one of claims 1 to 5, characterized in that: The initial sample points are generated based on the probability density function of the input variables, and the steps include: randomly selecting a certain number of sample points from the probability density function of each input variable by a random sampling method, combining the sample points of different input variables into a multidimensional sample set X1, X2, X3, ..., X n As the initial sample point, n represents the number of sample sets.

7. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to any one of claims 1 to 5, characterized in that: The input variables include any one or more of environmental input variables, model structure input variables, and model material input variables, wherein the environmental input variables include the average wind speed and / or turbulence intensity of the wind field, the model structure input variables include the height of the hybrid tower and / or the diameter of the hybrid tower, and the model material input variables include the elastic modulus of concrete and / or the elastic modulus of steel; a sample point is randomly selected from each input variable, and sample points of different input variables are combined to obtain a sample set, and the execution is repeated to obtain multiple groups of sample variable sets X1, X2, X3, ..., X m , where m represents the number of sample sets, and the sample variable set X1, X2, X3, …, X m The system is divided into multiple parts according to a specified ratio to be input into each mixed-tower wind turbine dynamics model respectively.

8. The reliability analysis method of a mixed-tower wind turbine based on a multi-fidelity numerical model according to any one of claims 1 to 5, characterized in that: The reliability calculation of the extreme value response result data obtained according to the second proxy model includes: Statistical analysis is performed on the extreme value response result data, and the preset structural failure criterion Y C The failure probability p of the structure is calculated according to the following formula F : In the formula, I(Y i >Y C ) is the indicator function, when Y i >Y C The value is 1 when it is, otherwise it is 0; The reliability index β is calculated based on the failure probability, and the calculation expression is: β=-Φ -1 (p F ) where Φ is the cumulative distribution function of the standard normal distribution.

9. A computer device comprising a processor and a memory, wherein the memory is used to store a computer program, wherein: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images