阅读说明：本技术 一种风电机组载荷在线预测的智能方法及风电机组 (Intelligent method for online prediction of load of wind turbine generator and wind turbine generator ) 是由 王文亮 叶少青 袁凌 潘磊 于 2020-06-30 设计创作，主要内容包括：本发明公开了一种风电机组载荷在线预测的智能方法及应用其的风电机组,通过将控制器模型B植入到风电机组的PLC中,在线获得实时的预测载荷,并用于风电机组特定工况载荷控制或保护控制等；控制器模型B的获得方法为：在各个风况下进行风机模型和控制器模型A的联合仿真,获得控制器模型A输出的可测数据和风机模型输出的载荷数据,并对所述数据进行深度学习获得载荷预测模型,将载荷预测模型嵌入所述控制器模型A得到控制器模型B。本发明将深度学习后的控制器模型植入风电机组PLC中,不需要批量增加额外的载荷测量设备及载荷传感器,通过控制器模型对载荷进行实时拟合即可实现对风电机组载荷实时地预测,且成本较低。(The invention discloses an intelligent method for online prediction of wind turbine load and a wind turbine applying the same.A controller model B is implanted into a PLC of the wind turbine to obtain real-time predicted load online and is used for load control or protection control and the like of the wind turbine under specific working conditions; the method for obtaining the controller model B comprises the following steps: and performing combined simulation of the fan model and the controller model A under each wind condition to obtain measurable data output by the controller model A and load data output by the fan model, performing deep learning on the data to obtain a load prediction model, and embedding the load prediction model into the controller model A to obtain a controller model B. According to the method, the controller model after deep learning is implanted into the wind turbine PLC, extra load measuring equipment and load sensors do not need to be added in batches, the load of the wind turbine can be predicted in real time by fitting the load in real time through the controller model, and the cost is low.)

1. An intelligent method for online prediction of wind turbine generator load is characterized in that a controller model B is implanted into a PLC of a wind turbine generator to obtain a real-time predicted load online;

the controller model B is obtained by the following method:

s101, performing combined simulation on a fan model and a controller model A under each wind condition, and acquiring measurable data output by the controller model A and load data output by the fan model;

and S102, carrying out deep learning on the measurable data and the load data obtained in the S101 to predict the load, and embedding a load prediction model obtained by the deep learning into the controller model A to obtain a controller model B.

2. The intelligent method for wind turbine load online prediction according to claim 1, wherein in the step S101:

the measurable data output by the controller model A at least comprises variables of left and right acceleration of an engine room, front and back acceleration of the engine room, rotating speed, wind direction, torque, pitch angle, propeller adjusting speed and historical values of the variables;

the load data output by the fan model includes, but is not limited to, the components of tower bottom and tower top loads.

3. The intelligent method for wind turbine load online prediction according to claim 1, wherein in the step S101:

the joint simulation of the fan model and the controller model under each wind condition is carried out by using software Bladed, and the controller model is required to be made into a Bladed external controller interface format and wind condition data are prepared.

4. The intelligent method for wind turbine load online prediction according to claim 1, wherein in the step S101:

the wind conditions include all design wind conditions required by the IEC standard specification.

5. The intelligent method for wind turbine load online prediction according to claim 1, wherein the deep learning employs a non-linear autoregressive neural Network (NARX) with external input; the input items of the learning are measurable data output by the controller model A, and the output items of the learning are load data output by the fan model.

6. The intelligent online prediction method for the load of the wind turbine generator set according to claim 1, characterized in that if the method is used for the first time, after the real-time predicted load is obtained online, the actually measured load is obtained by using load measuring equipment;

then, calibrating and correcting the predicted load according to the actually measured load; or if the difference between the predicted load and the measured load is too large, the model needs to be trained and corrected through field data.

7. The intelligent online prediction method for the load of the wind turbine generator as claimed in claim 6, wherein the wind turbine generator in the first use adopts a wind turbine generator prototype, and the prototype test meets the IEC load test standard specification requirement of the wind turbine generator.

8. The intelligent method for wind turbine generator load online prediction according to claim 1, wherein a result recording period of the predicted load is a PLC execution period.

9. The intelligent online load prediction method for the wind turbine generator according to any one of claims 1 to 8, wherein the obtained predicted load is used for load control to reduce the running load of the wind turbine generator under a specific working condition; or the load of the wind turbine generator is controlled to be out of limit to protect the wind turbine generator; or for evaluating the service life of the wind turbine;

in S102, the load prediction model obtained by deep learning and the control or evaluation algorithm are embedded in the controller model a to obtain a controller model B.

10. A wind turbine applying the intelligent method for wind turbine load online prediction according to any one of claims 1 to 9, wherein the controller model B is embedded in a PLC of the wind turbine.

Technical Field

The invention relates to the field of wind power control, in particular to an intelligent method for online prediction of load of a wind turbine generator and the wind turbine generator.

Background

In recent years, the control system of the wind turbine generator has stronger and stronger computing capability, and various modern algorithms can be applied to the wind turbine generator. In the past, the corresponding relation between the design load and the actual load of the wind turbine generator can be compared only by adding extra expensive equipment to a prototype. The load condition of the mass production fan is unknown, and the load difference caused by the process and the assembly can be judged only manually.

Therefore, it is obvious that the conventional wind turbine generator set has inconvenience and defects in load prediction, and further improvement is needed. How to create a new method which is low in cost and can predict the load of the wind turbine generator on line in real time, and the method is used for controlling the load of the wind turbine generator under specific working conditions to reduce the running load of the wind turbine generator or protect the wind turbine generator, which becomes an extremely improved target in the current industry.

Disclosure of Invention

The invention aims to solve the technical problem of providing an intelligent method for online prediction of the load of a wind turbine generator and the wind turbine generator, so that the online real-time prediction of the load of the wind turbine generator can be realized without adding extra expensive equipment and without manual judgment.

In order to solve the technical problems, the invention adopts the following technical scheme:

an intelligent method for online prediction of a wind turbine load is characterized in that a controller model B is implanted into a PLC of the wind turbine to obtain a real-time predicted load online; the controller model B is obtained by the following method:

s101, performing combined simulation on a fan model and a controller model A under each wind condition, and acquiring measurable data output by the controller model A and load data output by the fan model;

and S102, carrying out deep learning on the measurable data and the load data obtained in the S101 to predict the load, and embedding a load prediction model obtained by the deep learning into the controller model A to obtain a controller model B.

Further, in the step S101: the measurable data output by the controller model A at least comprises variables of left and right acceleration of an engine room, front and back acceleration of the engine room, rotating speed, wind direction, torque, pitch angle, propeller adjusting speed and historical values of the variables; the load data output by the fan model includes, but is not limited to, the components of tower bottom and tower top loads.

Further, in the step S101: the joint simulation of the fan model and the controller model under each wind condition is carried out by using software Bladed, and the controller model is required to be made into a Bladed external controller interface format and wind condition data are prepared.

Further, in the step S101: the wind conditions include all design wind conditions required by the IEC standard specification.

Further, the deep learning employs a non-linear autoregressive neural network NARX with external input; the input items of the learning are measurable data output by the controller model A, and the output items of the learning are load data output by the fan model.

Further, if the device is used for the first time, after real-time predicted load is obtained on line, actual measured load is obtained by using load measuring equipment, and then the predicted load is calibrated and corrected according to the actual measured load; or if the difference between the predicted load and the measured load is too large, the model needs to be trained and corrected through field data.

Further, the wind turbine generator used for the first time adopts a wind turbine generator prototype, and the prototype test meets the IEC load test standard specification requirement of the wind turbine generator.

Further, the result recording period of the predicted load is a PLC execution period.

Further, the obtained predicted load is used for load control to reduce the unit specific working condition operation load; or the load of the wind turbine generator is controlled to be out of limit to protect the wind turbine generator; or for evaluating the service life of the wind turbine; in S102, the load prediction model obtained by deep learning and the control or evaluation algorithm are embedded in the controller model a to obtain a controller model B.

The invention also provides a wind turbine generator applying the intelligent method for online load prediction of the wind turbine generator, wherein the controller model B is implanted into the PLC of the wind turbine generator.

By adopting the technical scheme, the invention at least has the following advantages:

1. according to the method, the controller model after deep learning is implanted into the wind turbine PLC, extra load measuring equipment does not need to be added in batches, extra load sensors do not need to be added, the load can be predicted in real time by the controller through real-time fitting of the controller model, and the cost is low.

2. The load prediction result can be used for load control to reduce the running load of the unit under a specific working condition, can be used for controlling the load of the wind generation unit to be out of limit to protect the wind generation unit, and can be used for evaluating the service life of the wind generation unit; and can also be used as a basis for comparing theoretical load with actual load.

Drawings

The foregoing is only an overview of the technical solutions of the present invention, and in order to make the technical solutions of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and the detailed description.

Fig. 1 is a schematic diagram of an intelligent method for online prediction of wind turbine load according to an embodiment of the present invention.

Detailed Description

As shown in fig. 1, the invention provides an intelligent method for online prediction of wind turbine load, which comprises the following steps:

s101, performing combined simulation on the fan model and the controller model A under each wind condition, and obtaining measurable data output by the controller model A and load data output by the fan model.

Specifically, the controller is modeled as a Bladed external controller interface format and wind condition data is prepared, including all wind conditions that can be generated as required by the IEC standard specification design, such as high wind, low wind, turbulence, and the like. And performing joint simulation of the fan model and the controller model under each wind condition by using software Bladed. Acquiring measurable data output by a controller model A and load data output by a fan model through simulation; the measurable data output by the controller model A at least comprises variables of left and right acceleration of an engine room, front and back acceleration of the engine room, rotating speed, wind direction, torque, pitch angle, propeller adjusting speed and historical values of the variables; the load data output by the fan model at least comprises the components of tower bottom and tower top loads.

And S102, carrying out deep learning on the measurable data and the load data obtained in the S101 to predict the load, and embedding a load prediction model obtained by the deep learning into the controller model A to obtain a controller model B.

Specifically, deep learning employs a non-linear autoregressive neural network NARX with external input; the input items of the learning are measurable data output by the controller model A, and the output items of the learning are load data output by the fan model. And obtaining the load prediction model through the deep learning training. And then embedding the load prediction model after deep learning into a controller model A to form a controller model B.

S103, implanting the controller model B into a PLC of the wind turbine generator, and obtaining the real-time predicted load on line.

In actual use, the controller model B can be used for the first time and used for the later period, when the controller model B is used for the first time, the controller model B is generally implanted into a PLC of a wind turbine model machine, the wind turbine model machine is generally provided with load measuring equipment, actual load can be obtained through the load measuring equipment, and then the predicted load is calibrated and corrected according to the actual load. Or if the difference between the predicted load and the measured load is too large, the model needs to be trained and corrected through field data.

When the controller model B is used in the later period, the molded controller model B can be implanted into a wind turbine generator which is produced in batches, so that real-time predicted load can be obtained on line, and the result recording period of the predicted load is the PLC execution period. The obtained predicted load is used for load control to reduce the running load of the unit under a specific working condition; or the load of the wind turbine generator is controlled to be out of limit to protect the wind turbine generator; or for evaluating the service life of the wind turbine; in practical application, a corresponding control or evaluation algorithm needs to be added into the controller model B at the same time, and the on-line load prediction application is performed on the wind turbine generator by combining a load prediction model obtained through deep learning.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the present invention in any way, and it will be apparent to those skilled in the art that the above description of the present invention can be applied to various modifications, equivalent variations or modifications without departing from the spirit and scope of the present invention.