JP2019218937A - Wind power generation system and method - Google Patents

Abstract

To provide a wind power generation system capable of improving power generation efficiency of a windmill for power generation and reducing fatigue loading.SOLUTION: The present invention relates to a wind power generation device comprising a nacelle connected to a tower, and blades connected to a first end portion of the nacelle. The wind power generation device also comprises: a Doppler rider which is mounted in the nacelle and measures a wind state in front of the wind power generation device; a learning unit which generates a wind state model for predicting the wind state at a position of the wind power generation device by inputting measurement of the detected wind state of the Doppler rider; and a predictor for predicting the wind state at a position of the wind power generation device on the basis of the wind state model received from the learning unit and the (latest) measurement of the Doppler rider. The learning device and the predictor are separated from each other.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a wind power generation system. In particular, the present invention relates to a wind power generator and a wind power prediction method for a wind power generator capable of performing a high-speed and high-accuracy prediction of a wind condition when a wind condition of a remote location in front of a rider reaches a windmill by a rider.

  Responses such as blade pitch angle control by controlling the operation of wind turbines using wind conditions measured by a Doppler lidar with a nacelle (LiDAR: Light Detection And Ranging, hereinafter referred to as lidar) Proposals to secure time are attracting attention. By measuring the wind condition on the windward side of the windmill with a rider installed on the windmill, the wind condition at the position of the windmill can be grasped and the characteristics of the windmill can be accurately evaluated. Control and torque control are expected to improve the efficiency of the wind turbine, reduce the fatigue load of the wind turbine blade, and improve the reliability and life of the wind turbine.

  For example, Patent Document 1 discloses a wind power generation device that measures a wind speed at a remote location in front of a wind power generation device using a front wind speed measurement device and controls the wind power generation device based on the measured wind speed.

JP 2013-148058 A

  The rider can measure the wind condition in front of the windmill, but the actual wind condition does not propagate as it arrives at the windmill from a distance. Especially near the windmill, the wind condition changes complicatedly due to the rotor itself and the windmill tower, and the manner of change is not uniform depending on the wind speed and time. Patent Literature 1 describes correction of a measurement position due to a change in attitude of a windmill, but does not consider changes in wind conditions from a distance to reach the windmill. Therefore, in the wind condition prediction system, when predicting the wind condition of the wind turbine position using the real-time measurement value in front of the wind turbine by the rider, the prediction processing time is required, so that the real-time prediction that follows the rider output and the wind turbine efficiency improvement based on the prediction, Insufficient reduction of wind turbine fatigue load.

  Accordingly, an object of the present invention is to provide a wind power generation system and a wind power generation system capable of performing a high-speed and high-accuracy prediction of a wind condition when a wind condition of a remote place in front of a windmill reaches a windmill by a rider.

  A wind power generator according to the present invention is preferably a wind power generator having a tower, a nacelle connected to the tower, and a blade connected to a first end of the nacelle, the wind power generator being mounted on the nacelle, A Doppler lidar that measures the wind condition in front of the device, and a learning device that generates a wind condition model for predicting the wind condition at the position of the wind power generator, using the measured value of the wind condition detected by the Doppler rider as an input, A predictor that predicts a wind condition at the position of the wind piezoelectric device based on a wind condition model received from the learning device and a (latest) measured value of a Doppler lidar; It is a separated wind power generator.

  In a prediction system based on machine learning, a learner and a predictor are separated, and the learner and the predictor are operated at required time intervals. The measurement value of the front wind condition by the rider is transmitted to the learning device at a relatively long cycle, for example, every 10 minutes, and the learning device uses the data to create a wind condition model for obtaining the wind turbine position wind condition from the front wind condition. This wind condition model is sent to the predictor at a relatively long cycle, for example, every 10 minutes. The predictor receives a measured value of the wind condition in front of the rider at a relatively short cycle, for example, every 0.25 s, and, based on the measured value and the wind condition model, sends a predicted value of the wind condition at the windmill position to the controller every 1 s. Send.

  According to the present invention, since the wind condition is predicted by using the learning unit that generates the wind condition model, the prediction of the wind condition when the wind condition at a remote place in front of the rider reaches the windmill by the rider with high accuracy. It can be carried out. In addition, a high-speed (real-time) wind turbine control can be assured by separating a learning unit that requires a relatively long time and a predictor that requires a small amount of complicated calculation and a short calculation time and operate them at different timings. it can.

It is a figure showing an example of a wind condition prediction system in a wind power generator. It is a figure showing an example of a wind condition prediction system in a wind power generator. It is a figure which shows the data of the wind condition prediction system in a wind power generator, and the flow of a process. FIG. 3 is a diagram illustrating a hardware configuration diagram of a learning device. FIG. 3 is a diagram illustrating a hardware configuration diagram of a predictor. FIG. 4 is a diagram illustrating a relationship between a time required for model generation and the number of data in a learning device. It is a figure showing an example of a wind condition prediction system in a wind power generator. It is a figure which shows the relationship between the wind speed in a wind power generator, and a measurement range.

  The embodiment for implementing the present invention proposes a method of predicting a wind speed immediately before a rotor from a distant wind speed measured by a rider by using machine learning. Here, a prediction system based on machine learning is examined separately for a learning device and a prediction device.

  A function to execute a calculation to generate a wind condition model at the windmill position from variables representing weather conditions such as wind conditions by learning from rider measurement values for a certain period of time and other sensors such as cup-type anemometer and arrow feather anemometer by learning Call it a learning device. A functional part that executes a calculation for predicting the wind condition at the windmill position using the latest rider measurement value based on the wind condition model is called a predictor. The learning device and the predictor are composed of an operation unit, a storage unit, and an interface unit (details will be described later). The learner by the rider collects a lot of data on the wind conditions and weather conditions measured by the rider and the wind conditions at the windmill to generate a wind condition model.In order to improve the model accuracy, that is, the accuracy of prediction, When the number of data is increased, the processing time increases, and a processing time of several seconds to several minutes, and in some cases, a time order is required.

  On the other hand, since changes in wind conditions are affected by weather conditions around the windmill, it is effective to improve the accuracy of model generation using data collected at a time as close as possible to the time when the model is used. . In generating the model, the number of data and the freshness of the data must be taken into account. The predictor predicts the wind turbine position and wind conditions using the latest data from the rider based on the model generated by the learning device, but the processing time for this prediction is less than 1 second and is relatively short. Can be processed.

  In the case of a horizontal type in which a laser beam used for measurement is emitted in a horizontal direction, a rider installed in a windmill can measure a wind condition several hundred meters ahead. For example, if a wind condition 300 m ahead is measured and the wind speed is 15 m / s toward the wind turbine, the wind turbine will reach the wind turbine 20 seconds later, and if it is 30 m / s, it will reach the wind turbine 10 seconds later. By setting the pitch angle, the yaw angle, and the like of the windmill in accordance with the wind condition within the time of the wind condition propagation, it is possible to improve the power generation efficiency of the windmill and reduce the fatigue load. Therefore, the model generation time in the learning device becomes a problem. The time required for controlling the pitch angle, the yaw angle, and the like must not be impaired by the generation of the wind condition model. On the other hand, wind condition models must maintain high prediction accuracy.

  As shown in FIG. 1, a wind power generator 1 including a Doppler lidar 2 as a wind condition measuring device (hereinafter, the Doppler lidar may be referred to as a rider and the wind power generator may be referred to as a windmill in the present specification). It is provided with a wind condition prediction system 10 by machine learning based on measured values, the state of the windmill itself, or data of various sensors 3 such as a cup-type anemometer or an arrow-type anemometer attached to the windmill.

  As shown in FIG. 2, the windmill includes at least one blade 24, a hub 23 to which the blade 24 is attached, a nacelle 20 to which the hub is attached, and a tower 22 that supports the nacelle 20. When the blade 24 receives the wind, the rotor constituted by the blade 24 and the hub 23 rotates, and power is generated by the generator installed in the nacelle 20. The rider 2 is installed in the nacelle 20 as a means for measuring a wind condition reaching the windmill.

  In FIG. 2, the downwind type in which the nacelle 20 is located on the windward side of the blade 24 with respect to the wind direction 44 is shown. It is applicable if the missing measurement values are supplemented.

  The wind condition prediction system 10 according to the present embodiment is configured by a learning unit 8 and a prediction unit 9 which are independent arithmetic units separated as shown in FIG. The data and processing flow when predicting the wind turbine position wind condition (wind speed at the location where the wind turbine is located) by the wind condition prediction system consisting of a learning device and a predictor from the wind condition in front of the wind turbine measured by Rider 2 An example is shown in FIG. The rider measurement values are separately sent to the learning device 8 via the Ethernet 11 and to the prediction device 9 via the CAN BUS 13 signal line. Rider measurement values include measurement values that are not used in the scope of the present invention, so the measurement values included in the explanatory variables are extracted therefrom, the inclination of the windmill tower due to the wind is corrected, the measurement data is normalized, and then learning is performed. The model generation by the calculator 8 and the calculation of the predicted value by the predictor 9 are performed.

  In the wind power generator, the pitch angle, yaw angle, generator torque, and the like of the wind turbine are controlled based on the wind condition and the wind turbine condition in order to improve the power generation efficiency and reduce the fatigue load. Wind power generators are generally designed to start generating power at a wind speed of 4 m / s and stop generating power at a wind speed of 15 m / s. The wind condition is measured by a sensor at the position of the windmill. However, since the wind condition is constantly changing, control of the pitch angle, the yaw angle, the generator torque and the like may not be able to follow the change in the wind condition in the measurement of the windmill position. For example, a gust (a phenomenon in which the wind speed changes suddenly and irregularly in a short period of time) gives a large fatigue load to the wind turbine, but if the effect of the gust appears on the wind turbine and then controls it, it may not be possible to avoid an unexpected fatigue load. is there. It is necessary to detect this before the gust reaches the windmill and adjust various settings of the windmill such as the pitch angle to the wind speed in anticipation of the response speed. By the way, it takes 1 second to adjust the pitch of the airflow generator by 10 degrees. By measuring the wind conditions ahead of the windmill using the rider, it is possible to predict the gust that will eventually reach the windmill.

  However, the actual wind conditions do not propagate as they arrive at the windmill from a distance. Especially near the windmill, the wind condition changes complicatedly due to the blade itself and the windmill tower, and the manner of change is not uniform depending on the wind speed and time. Therefore, using machine learning, we provide a method to predict the wind condition at the windmill position with high accuracy and high speed from the distant wind condition measured by the rider. The details will be described below.

  One embodiment of a wind condition prediction system based on machine learning employing a wind condition measuring device will be described with reference to FIG.

  A rider 2 for measuring the wind condition in front of the windmill is installed on the windmill 1 (specifically, the nacelle 20 of the windmill 1), and the measurement value, which is the wind condition data of the rider 2, is obtained by machine learning via the signal lines 11 and 13. It is sent to the wind condition prediction system 10 and predicts the wind condition at the wind turbine position from the wind condition in front of the wind turbine.

  The learning device 8 creates a wind speed model 12 from the wind condition data and sends it to the predictor 9. The predictor 7 calculates wind speed prediction data 14 based on the wind condition model 12 from the learning device 8 and the wind condition data 13 from the rider 2 and sends it to the controller 7.The controller 7 outputs from the windmill, In addition to inputting wind turbine information 5 such as pitch control and nacelle orientation, control information 6 such as wind turbine pitch angle control based on wind speed prediction data 14, wind speed of an anemometer or anemometer 3, and sensor information 4 such as wind is input. Send to windmill 1.

  FIG. 2 shows an example of a method of measuring the wind condition in front of the windmill by a rider installed near the windmill controller 7 in the case of a downwind type windmill. A laser beam is emitted from the rider in four directions (25, 26, 27, 28) in front of the windmill (upwind, opposite to the wind direction 44), and backscattered light from the aerosol in the atmosphere is detected. The backscattered light undergoes a Doppler shift reflecting the movement of the aerosol due to the wind, and the wind speed along the laser beam can be known from the shift amount. The wind speed can be determined for a plurality of ranges of measurement points 29 along each laser beam. Further, the laser beam is irradiated in a predetermined direction as shown in FIG. 2, and a wind speed change in a vertical plane 30 in the same range can be obtained in consideration of a difference in speed obtained from each direction. The number of ranges that can be measured by riders is about ten.

  The learning device generates a model that derives the R1 wind speed based on the R2 wind condition by using the explanatory variable as the rider measurement value in the range R2 and the target variable as the wind speed in the range R1 as shown in FIG. For example, the R2 range is the measurement point 300 m ahead of the windmill, and the R1 range is the windmill position. As the R2 wind condition serving as an explanatory variable, wind speed, wind direction, shear (vertical wind speed change), turbulence, temperature, humidity, and the like can be set alone or in any combination thereof. The target variable is a variable necessary for controlling the windmill, and the wind speed at the windmill position, the wind direction, a change in the wind speed in the vertical direction, and the like can be set alone or in combination. An algorithm corresponding to an arbitrary machine learning such as a support vector, a neural network, a decision tree, a random forest, and naive Bayes can be used as an algorithm for generating a wind speed model. At the time of learning, as shown at 31 in FIG. 2, a model is generated by acquiring a plurality of R2 wind conditions and R1 rider measurement values. When the wind condition model is generated, the measured value of the wind condition of R2 is input to the wind condition model generated as shown in FIG. 2 to obtain a predicted wind speed at the position of the windmill. For generation of the wind condition model, various sensors 3 such as a cup-type anemometer or a naval-type anemometer equipped with a nacelle in FIG. 1 can be used instead of the rider measurement value as the wind speed at the position of the windmill.

  Regardless of which algorithm is selected, a high-precision model generation requires a certain processing time. In order to obtain accuracy, it is necessary to handle a certain number or more of measurement data, and it is difficult to make the time required for model generation shorter than the cycle of real-time predicted value output required by the controller 7. That is, since the wind condition model generation process requires time, the wind in the R2 range reaches the windmill before the control by the controller 7 is completed. Therefore, the learning device 8 and the predictor 9 are separated as shown in FIG. 1, and the generation of the model and the calculation of the predicted value using the model are executed at different timings by the separated hardware.

  The configuration and analysis procedure will be described below while specifically showing the processing time and signal flow. The rider measurement value of the forward wind condition at R2, which is an explanatory variable (measurement at intervals of 0.25 s), and the target variable R1, that is, the wind condition at the windmill position, are output from the rider every 10 minutes at once, and are output via the signal line 11. Sent to the learning device 8. In 10 minutes, 600 sets of explanatory variables and objective variables are sent to the learning device 8 with the measured values as data. On the other hand, the measurement value of the R2 range (measurement at intervals of 0.25 s) is sent to the predictor 9 via the signal line 13 every 1 s (second) because the measurement value of each range of the rider is four. Become.

  FIG. 4 shows a hardware configuration of the learning device 8. The learning device 8 includes a processor (CPU) 41, a memory 42, a storage device 43, and a communication I / F 44. The processor 41 controls the learning device 8 to realize the arithmetic circuits shown in FIG. 3 for extracting measurement data, correcting inclination, normalizing measurement data, learning, and the like. The memory 42 is a work area of the processor 41, and is a non-temporary or temporary storage medium for storing various programs and data. The storage device 33 is composed of a HDD (Hard Disk Drive) and a flash memory, and stores various programs for executing each operation of measurement data extraction, inclination correction, measurement data normalization, learning, and measurement from the rider 2. Stores a value. The communication I / F 44 receives the measurement value from the rider 2 via the signal line (Ether) 11 and transmits the wind condition model to the predictor 9 via the signal line 12. Although the signal line 11 and the signal line 12 are shown as the same line in FIG. 4, actually, separate signal lines are preferable.

  FIG. 5 shows a hardware configuration diagram of the predictor 9. The predictor 9 has a processor (CPU) 51, a memory 52, a storage device 53, and a communication I / F 54. The processor 51 controls the predictor 9 and implements the arithmetic circuits shown in FIG. 3, such as measurement data extraction, inclination correction, measurement data normalization, and prediction. The memory 52 serves as a work area for the processor 51, and is a non-temporary or temporary storage medium for storing various programs and data. The storage device 53 is composed of an HDD (Hard Disk Drive) and a flash memory, and stores various programs for executing each calculation of measurement data extraction, inclination correction, and measurement data normal prediction. Store. The communication I / F 54 receives the measurement value from the rider 2 via the signal line 13, receives the wind condition model from the learning device 8 via the signal line 12, and transmits the wind condition prediction data via the signal line 14. Transmit to controller 7. Although the signal lines 12, 13, and 14 are shown as the same line in FIG. 5, actually, separate and independent signal lines are preferable. The signal line 13 is a deep battle separate from the signal line 11, and the measurement value from the rider 2 is transmitted to the learning device 8 and the prediction device 9 at different timings.

  The learning device 8 generates a wind condition model by the above-described machine learning algorithm. When the number of data to be used is 500 for the generation of the wind condition model, about 1.5 s is required as shown in the relationship between the number of data and the time required for model generation in FIG. 6, and for 600, about 2 s is required. I do. The model generated every 10 minutes by the learning device 8 is sent to the predictor 9 via the signal line 12 every 10 minutes. Aside from the learning device 8, the rider measurement value of the forward wind condition in R2 (measurement at intervals of 0.25 s) is output from the rider every 0.25 s as an explanatory variable, and sent to the predictor 9 via the signal line 13. The predictor averages the four data of the four points in the R2 range and sends a predicted value of the windmill position wind speed to the controller 7 every 1 s. The predictor 9 predicts the wind condition at the position of the wind turbine every 1 s based on the wind condition model sent from the learning device 8 every 10 minutes and the latest measurement value by the rider, and sends it to the controller 7.

  In addition to the predicted value from the predictor 9, the wind turbine controller 7 has sensor information such as wind speed and wind direction via a signal line 14, and output, rotation speed, pitch control, nacelle orientation, etc. via a signal line 5. Information is sent. The controller 7 integrates and processes these data, and sends control information of the windmill to the windmill via the signal line 6. The time interval specifically described in the present embodiment is an example, and an optimal numerical value setting can be applied as needed.

  The learning device requires a relatively long operation time to generate a wind condition model from the measurement data. On the other hand, the prediction of the predictor and the control of the controller based on the prediction require a strong real-time property (standby by another routine and delay of processing are not allowed). In the predictor, there are few complicated calculations, and the calculation time is short as described above, so that it does not substantially affect the real-time property of the controller.

  Unlike the configuration in which the learning device and the predictor are separated as described in the first embodiment, if the learning device and the predictor are configured with one hardware without being separated, the operation timing of the model generation and the predicted value output is reduced. Since the two different functions are not separated from each other, it is necessary to provide hardware or software for adjusting the timing of the two functions when applied to windmill control using lidar measurement. When the measurement value of the rider is lost, it is necessary to generate a model by complementing the missing data from the previous and subsequent ranges, and it becomes more difficult to adjust the operation timing. In the first embodiment, which is characterized by the separation of two functions, the configuration in which each function is executed by a different arithmetic circuit can improve the maintainability and the operation stability. It is desirable that the wind condition model is generated at a time close to the weather condition at the time of the forecast, but if the model is generated from a sufficient number of data including various weather conditions, it will not be the latest wind condition model but will have high accuracy. Can be predicted. The number of data is 100 or more, preferably 300 or more. Even if there is an unintentional stop of the learning device or an intentional stop such as maintenance, a large prediction error does not occur even when the model stored in the predictor is continuously used.

  As described above, according to the first embodiment, (i) the learner and the predictor are separated from each other, and the learner is operated at a timing different from that of the predictor. .

  (Ii) The hardware configuration required for the operation of each of the learning device and the prediction device can be selected (the learning device may have a relatively low operation speed and communication speed), and the cost of the system can be reduced.

  (Iii) Separating the learning device that performs more complicated processing than the predictor from the predictor and the controller, resulting from the complexity of the operation of the learning device and the lack of track record in the windmill control system. Even if a failure occurs, the function of the controller is not impaired, the stop of the control system can be avoided, and the maintainability and operation stability of the system can be improved.

  In the second embodiment, the same components as those in the first embodiment are denoted by the same reference numerals, and redundant description will be omitted. As described above, the processing of the predictor 9 is relatively simple and the operation time is short. Therefore, as shown in FIG. 7, it is possible to add a function of a predictor to the arithmetic unit and the storage device of the wind turbine controller 7. The wind turbine controller 7 generates the control signal 6 at time intervals in seconds, which coincides with the time interval of the predictor operation. Therefore, even if the predictor is incorporated in the controller and the same arithmetic unit is used, complicated adjustment of the timing is unnecessary. With the configuration in which the learning device and the predictor are separated and the predictor and the controller are driven by the same hardware, the control system of the windmill and the entire prediction system can be simplified.

  According to the second embodiment, a predictor whose processing is relatively simple is incorporated in a controller conventionally installed in a wind turbine for wind turbine control, thereby simplifying the configuration of a wind condition prediction system by machine learning. can do.

  In Examples 1 and 2, the measurement point range of the wind condition in front of the windmill, which is an explanatory variable, was one point of R2 in FIG. By adding the measured value of the R3 range and the measured values of the R4 and R5 ranges to the explanatory variables of the predictor 9, information on the direction of the change in the wind condition is added, and the change in the wind condition occurs rapidly. The prediction accuracy can be improved.

  Further, in an explanatory variable including measurement values in a plurality of ranges, a measurement position in at least one of the plurality of ranges is set shorter than a distance at which a damping effect near the windmill becomes significant. FIG. 8 shows the relationship between the range and the wind speed. Riders can measure wind conditions in multiple ranges almost simultaneously. This figure shows wind speeds of different ranges measured at substantially the same time, and wind speeds up to 400 m by size. It can be seen that the wind speed starts decreasing when the range is close to the windmill and becomes 120 m or less, and the decrease in wind speed becomes remarkable when the range becomes 100 m or less. This is called a damming effect, and in a region where this effect is remarkable, the influence on the wind condition is large. As a result, the wind condition distribution in this region is different from that in a region far from the windmill. Therefore, the R3 range is set at a position 130 m or less from the windmill where the damping effect is remarkable, and by adding it to the explanatory variables, information on the wind condition distribution in the vicinity of the windmill is added, and the accuracy of prediction is improved. be able to.

  However, when waiting for the measurement of the wind conditions in the range close to the wind turbine and making predictions using this, there may be a shortage of time to set the various control parameters of the wind turbine to the optimal values for the wind conditions described above. When setting the value of R3, it is necessary to consider the response speed of windmill control.

  According to the configuration of the first embodiment, the high-speed prediction, high accuracy, and high reliability are simultaneously satisfied in the wind condition prediction system installed in the wind turbine. However, to actually achieve these features, the rider's measurements must be sent to the predictor without loss. For example, when the measurement range R2 of the forward wind condition, which is an explanatory variable, is set to 200 m and the predictor outputs a predicted value every second, if one point of data loss occurs, two points before and after the correction are corrected If a value is used, the reliability of the explanatory variable will decrease before and after the data loss time. Assuming a wind speed of 20 m / s, it takes 10 s for a wind condition of 200 m to reach the windmill, which is not a time for control in view of the response time of windmill control. When a rider is mounted on an upwind type windmill, loss of rider data due to the blade cannot be avoided. It is good if it is possible to correct data loss by averaging, such as estimating the amount of power generation based on long-term average wind speed, but if the purpose of the present invention is to control windmills by short-term prediction, data loss is predicted. Can be a problem that impairs the accuracy in Therefore, the problem of data loss can be avoided by adding a downwind type windmill to the components of the present invention. This makes it possible to realize high-speed and high-precision prediction.

  DESCRIPTION OF SYMBOLS 1 ... Windmill for power generation, 2 ... Rider, 3 ... Conventional anemometer and anemometer attached to a windmill, 6 ... Transmission line sent to the windmill body of a controller, 7 ... Windmill controller, 8 ... Learning machine, 9: Predictor, 10: Wind speed prediction system by machine learning, 11: Transmission line for sending the lidar measured value to the learning device, 12: Signal line for sending the wind speed model generated by the learning device to the predictor, 13: Rider measurement value Transmission lines to be sent to the predictor, 20: nacelle, 22: tower, 23: hub, 24: blade.

Claims (15)

A nacelle connected to a tower, and a wind turbine having a blade connected to a first end of the nacelle,
A Doppler lidar mounted on the nacelle and measuring a wind condition in front of the wind power generator,
A learning device that generates a wind condition model for predicting a wind condition at the position of the wind power generation device, with a measurement value of the detected wind condition of the Doppler lidar being input,
A wind condition model received from the learning device, and a predictor that predicts a wind condition at the position of the wind piezoelectric device based on the measured value of the Doppler lidar,
A wind power generator, wherein the learning device and the predictor are separated.   The wind power generator according to claim 1, wherein the learning device and the predictor operate by independent calculation circuits.   The wind power generator according to claim 2, wherein the detection data of the Doppler lidar is transmitted to the learning device and the prediction device via independent signal lines.   The wind power generator according to claim 1, wherein the predictor and a controller for controlling a windmill of the wind power generator are operated by the same arithmetic circuit.   A plurality of measurement points of the wind condition in front of the windmill, which are explanatory variables in the predictor, wherein a measurement position of at least one measurement point is shorter than a distance at which the damping effect in the windmill becomes significant. 2. The wind power generator according to 1.   The wind power generator according to claim 2, wherein the wind power generator is a downwind type wind turbine. A nacelle connected to a tower, and a wind turbine having a blade connected to a first end of the nacelle,
A Doppler lidar for measuring a wind condition in front of the wind turbine mounted on the nacelle,
The wind condition of the front range of the wind power generator measured by the Doppler lidar is used as an explanatory variable, and the wind condition of the position of the wind power device is used as a target variable to predict the wind condition of the position of the wind power device. A learning device for generating a situation model,
A wind condition model received from the learning device, based on the measured value of the Doppler lidar, to predict the wind condition at the position of the wind piezoelectric device, a predictor configured with a different arithmetic unit from the learning device And having
A wind power generator, wherein the learning device and the predictor perform calculations at different independent timings.   8. The wind power generator according to claim 7, wherein the learning device generates a wind condition model with a cycle longer than a cycle predicted by the predictor. The wind condition of the front range of the wind power generator as the explanatory variable, the wind speed, wind direction, shear, turbulence, temperature, humidity alone or set as any combination thereof,
The wind power generator according to claim 7, wherein the target variable is set as a wind speed, a wind direction, and a vertical wind speed change in the wind power generator alone or as a combination thereof. A wind condition prediction method for a wind power generator,
By Doppler lidar, measure the wind condition in front of the wind turbine,
The measured value of the Doppler lidar as an explanatory variable, the wind condition of the position of the wind turbine as a target variable, to generate a wind condition model for predicting the wind condition of the position of the wind turbine,
Based on the wind condition model and the latest measurement value of the Doppler lidar, predict a wind condition at the position of the wind piezoelectric device,
The cycle for generating the wind condition model is longer than the period for predicting the wind condition at the position of the wind piezoelectric device, wherein the wind condition predicting method for the wind power generator is characterized in that:   The wind condition prediction method for a wind power generator according to claim 10, wherein the generation of the wind condition model and the prediction of the wind condition at the position of the wind piezoelectric device are executed at independent timings.   The input of the detection data of the Doppler lidar for the generation of the wind condition model and the input of the detection data of the Doppler lidar for the wind condition for prediction of the wind condition at the position of the wind piezoelectric device are independent and different. The wind condition prediction method for a wind turbine generator according to claim 10, wherein the timing is timing.   A plurality of measurement points of the wind condition in front of the windmill, which are explanatory variables in the predictor, wherein a measurement position of at least one measurement point is shorter than a distance at which the damping effect in the windmill becomes significant. 12. The wind condition prediction method for a wind power generator according to item 11.   11. The method according to claim 10, wherein a cycle of generating the wind condition model is longer than a cycle of predicting a wind condition at the position of the wind piezoelectric device. The wind condition of the front range of the wind power generator as the explanatory variable, the wind speed, wind direction, shear, turbulence, temperature, humidity alone or set as any combination thereof,
The wind condition prediction method for a wind power generator according to claim 14, wherein the target variable is set as a wind speed, a wind direction, and a vertical wind speed change in the wind power generator alone or in a combination thereof.

Images