WO2022100021A1 - 基于风机相互预警的虚拟激光雷达系统及方法 - Google Patents

基于风机相互预警的虚拟激光雷达系统及方法 Download PDF

Info

Publication number
WO2022100021A1
WO2022100021A1 PCT/CN2021/090214 CN2021090214W WO2022100021A1 WO 2022100021 A1 WO2022100021 A1 WO 2022100021A1 CN 2021090214 W CN2021090214 W CN 2021090214W WO 2022100021 A1 WO2022100021 A1 WO 2022100021A1
Authority
WO
WIPO (PCT)
Prior art keywords
wind
wind speed
data
neural network
relationship
Prior art date
Application number
PCT/CN2021/090214
Other languages
English (en)
French (fr)
Inventor
卜宏达
谢洪亮
王林鹏
葛亮
Original Assignee
远景能源有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 远景能源有限公司 filed Critical 远景能源有限公司
Publication of WO2022100021A1 publication Critical patent/WO2022100021A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/88—Lidar systems specially adapted for specific applications
    • G01S17/95—Lidar systems specially adapted for specific applications for meteorological use
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
    • G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • the invention relates to the technical field of wind turbines, in particular to a virtual laser radar system and method based on mutual early warning of wind turbines.
  • wind speed prediction has been a hot issue in recent years. Wind speed is affected by temperature, air pressure, terrain and other factors, and has strong randomness. According to the forecast time scale, wind speed forecast can be divided into long-term (months-years), medium-term (days-weeks), short-term (hours), and ultra-short-term (minutes) forecasts. These wind speed forecasts are mainly used for wind power forecasting of wind farms. The correct deployment plan can reduce the adverse impact of wind power on the grid.
  • wind speed prediction methods basically build a model for the original wind speed time series of the wind farm, and then predict the future wind speed, or convert the non-stationary wind speed sequence into a stationary sequence by means of difference, and then build the stationary sequence. Models and predictions. But for second-level wind speed prediction, relatively reliable prediction methods are rare. In "The Impact of Control Technology—2nd Edition" published by IEEE, the advance planning and control based on the second-level wind speed sequence in the future is proposed as a major challenge for future wind turbine control research.
  • the lidar measures the wind speed on a three-dimensional plane several hundred meters in front of the fan by shooting a laser to the front, so that the wind speed in the next few seconds can be predicted in advance. , but the actual use effect is not good due to the high price, great influence of weather, instability, and difficult calibration settings.
  • the purpose of the present invention is to provide a virtual laser radar system and method based on mutual early warning of wind turbines, so as to solve the problem that the existing physical laser radar has poor effect in the actual use of wind farms.
  • the present invention provides a virtual laser radar system based on mutual early warning of fans, including:
  • a data module is deployed in the wind field to obtain the front-to-back relationship of the fans arranged in the main wind direction in the wind field. Due to the poor use effect of the hardware laser radar, the present invention uses the software method to replace the laser radar to predict the gust of wind. Because the wind speed is transitive, the wind farm will be arranged at intervals in the main wind direction. If the front fan encounters a gust, the rear fan will also encounter the same gust after a period of time. Based on this physical The principle and observation of the wind farm operation data show that the wind speed between the fans is related and has a certain hysteresis characteristic, and the present invention uses a neural network to capture this relationship.
  • Wind speed transfer characteristic module to obtain the transfer characteristic of wind speed in the main wind direction in the area where the wind field is located;
  • the wind farm operation data module which obtains the actual operation data of the wind farm.
  • the neural network module captures the wind speed relationship between the fans according to the front-to-back relationship of the fans arranged at intervals in the main wind direction in the wind field, the transmission characteristics of the wind speed in the main wind direction, and the operation data of the wind field, so as to replace the physical lidar to predict the gust of wind.
  • the neural network module includes a recurrent neural network model and a graph neural network model;
  • the recurrent neural network model and the graph neural network model are formed according to the front-to-back relationship of the fans arranged at intervals in the main wind direction in the wind farm, and the transmission characteristics of the wind speed in the main wind direction;
  • the neural network module uses the actual operation data of the wind farm to train the cyclic neural network model and the graph neural network model, performs spatiotemporal prediction of wind speed, and uses the actual operation data of the wind farm in other time periods for verification.
  • the model is used to train the model for the actual operation of the wind farm for spatiotemporal prediction, and the data of other time periods are used for verification.
  • the forecast of wind speed is of great economic value in reducing fan fatigue and increasing power generation.
  • the actual operation data of the wind farm is second-level wind speed detection data
  • the neural network module captures the wind speed relationship between the fans, including: obtaining the correlation of the second-level wind speed detection data of each fan according to the second-level wind speed detection data, and obtaining the entire wind speed according to the correlation of the second-level wind speed detection data of each fan. Field flow field information.
  • the transmission characteristic of the wind speed in the main wind direction is the mutual early warning characteristic between wind turbines
  • the neural network module uses the time series deep learning algorithm to train the second-level wind speed detection data, establishes a specific wind field flow field model, and based on the specific wind field flow
  • the field model predicts the wind speed trend information of each wind turbine in the next tens of seconds in real time, and uses the mutual early warning feature between wind turbines to realize virtual lidar.
  • Each wind farm has massive second-level wind speed data.
  • the wind speed data of different planes are not isolated and irrelevant. In these massive data, their interrelations have described the flow field information of the entire wind farm. Therefore, combined with the data on the arrangement of wind farms, the time series deep learning algorithm can be used to train these wind speed data, so as to establish a flow field model of a specific wind field, and based on the flow field model, real-time prediction of each wind turbine in the next tens of seconds The wind speed trend information, and then use the mutual early warning between the wind turbines to realize the virtual lidar.
  • the temporal and spatial prediction of wind speed includes: in a specific technical field, in the traditional time series prediction problem, often use its own historical data to learn to find out the rules. , that is, using a recurrent neural network to capture the relationship before and after the time series.
  • a structural gate regression unit commonly used in recurrent neural networks.
  • Z t is the update gate, indicating whether the information at the last moment needs to be updated
  • r t is the reset gate, indicating whether the information at the last moment needs to be reset
  • h t is the hidden layer output of the gate regression unit, receiving the time information of h t-1 and h t ;
  • the contextual relationship on the time series is captured, and the future wind speed data is predicted through the wind speed historical data of the current wind turbine and other wind speed historical data.
  • X is the original node attribute
  • A is the adjacency matrix
  • Z is the transformed node attribute
  • I is the diagonal matrix.
  • the relationship between spatial nodes is captured through the graph neural network, and the law of data flow in the relationship between the spatial nodes is obtained.
  • the graph neural network has been used in traffic flow prediction, social network, protein structure prediction and other directions. Some people use it in the field of wind speed prediction, let alone in the scope of wind farms.
  • the cyclic neural network is used to capture the relationship in the time series of a single wind turbine
  • the graph neural network is used to capture the relationship between multiple wind turbines in the wind farm. inter-spatial relationship.
  • the wind speed sequence of the adjacent wind turbines we can obtain the obvious lag situation, that is, the wind speed of the wind turbine at the current time t, and the wind speed of the wind turbine at the time t+ ⁇ t. occurs, especially in the case of gusts, where ⁇ t depends on the distance d between the two fans and the current wind speed ws,
  • the wind speed of a certain fan at time t+ ⁇ t is predicted by the wind speed of the other fans at time t and the mutual positional relationship of the fans.
  • the wind speed of the first fan at time t is equal to that of the second fan at time t+ ⁇ t.
  • ⁇ t ⁇ d/ws where d is the distance between the two fans, and ws is the wind speed of the first fan at time t.
  • wind farm time series data is collected, data cleaning, standardization, division of training data sets and test data sets, construction of wind farm network structures, and training of deep learning neural networks are performed.
  • Network model, model validation, model deployment, fan control includes: collecting the operating wind speed data of each fan in the wind farm;
  • the training set is used to train the network weights, and the test set does not participate in any Training is only used to calculate the evaluation index to detect the performance of the model;
  • the wind farm network structure is divided into static network and dynamic network
  • the static network is the relationship matrix between fixed wind turbines in the training process
  • the dynamic network is the relationship matrix between the unfixed wind turbines in the training process
  • the relationship matrix It varies with the similarity of the current wind speed sequence of different fans
  • the deep learning neural network model input the normalized historical wind speed data and adjacency matrix of each wind turbine to the graph convolutional neural network to extract the spatial feature matrix, input the output result to the recurrent neural network model to extract the time feature matrix, and output the future of each wind turbine.
  • the wind speed value at the moment determine that the optimizer is Adam, and the loss function is the root mean square average error MSE;
  • model training set After the model training set converges, verify it on the test set, calculate the mean absolute error Mae index and draw the time series, and deploy the model at the wind farm station;
  • the wind speed of each wind turbine in the wind farm can be effectively predicted for a period of time in the future.
  • the most front-row fans in the incoming wind direction mainly rely on their own historical data for prediction.
  • the rear-row fans can also rely on the information provided by the front-row fans to make predictions with higher accuracy, especially when all fans are running normally. time to effectively replace lidar.
  • the wind speed trend prediction time based on the mutual early warning model of wind turbines is much longer than the range that can be predicted by the hardware lidar.
  • the range is about 80-200m in front of the wind wheel, and the wind speed prediction range based on deep learning can reach more than 300m.
  • the virtual lidar based on mutual early warning of wind turbines is based on the wind speed prediction sequence of deep learning, and is equipped with model predictive control algorithm and Markov decision model to calculate the comprehensive optimal power generation and component load of wind turbines.
  • the present invention also provides a virtual laser radar method based on mutual early warning of fans, including:
  • the wind farm deployment data module obtains the front-to-back relationship of the fans arranged at intervals in the main wind direction in the wind farm;
  • the wind speed transfer characteristics module obtains the transfer characteristics of the wind speed in the main wind direction in the area where the wind field is located;
  • the wind farm operation data module obtains the actual operation data of the wind farm.
  • the neural network module captures the wind speed relationship between the fans based on the front-to-back relationship of the fans arranged at intervals in the main wind direction in the wind field, the transmission characteristics of the wind speed in the main wind direction, and the operation data of the wind field, so as to replace the physical lidar to predict the wind gust.
  • the neural network module is used according to the front and rear relationship of the wind turbines arranged at intervals in the main wind direction in the wind field, the transmission characteristics of the wind speed in the main wind direction, and the wind field.
  • the operation data captures the wind speed relationship between the wind turbines to replace the physical lidar to predict wind gusts, and realizes a virtual lidar method based on the mutual early warning of wind turbines. Wind speed prediction has great economic value in reducing fan fatigue and increasing power generation.
  • the invention proposes a virtual laser radar system and method based on mutual early warning of wind turbines.
  • the wind speed of each machine point in the wind field for a period of time in the future can be predicted.
  • the predicted second-level wind speed is applied to the real-time control of wind turbines to overcome various limitations brought by hardware lidar.
  • the beneficial effects of the invention are as follows: cost saving, about 500,000 yuan per lidar can be saved; not affected by weather, high reliability, accuracy, and stable application; more comprehensive and detailed wind conditions are obtained, and the span is longer;
  • the software architecture is easy to deploy, does not require high computing resources, and has no deployment threshold; based on second-level wind speed time series prediction, it can protect wind turbines from damage from extreme loads, and effectively reduce tower fatigue loads, which will also reduce the cost of future wind turbine design. have important meaning.
  • FIG. 1 is a schematic diagram of a virtual laser radar method based on mutual early warning of fans according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a door regression unit of a virtual lidar system based on mutual early warning of fans according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of a neural network of a virtual lidar system diagram based on mutual early warning of fans according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a virtual lidar method based on mutual early warning of wind turbines according to an embodiment of the present invention.
  • the core idea of the present invention is to provide a virtual laser radar system and method based on mutual early warning of wind turbines, so as to solve the problem that the existing physical laser radar has poor effect in the actual use of wind farms.
  • the present invention provides a virtual laser radar system and method based on mutual early warning of wind turbines.
  • module to obtain the transmission characteristics of wind speed in the main wind direction in the area where the wind farm is located;
  • the wind farm operation data module to obtain the actual operation data of the wind farm;
  • the neural network module according to the front and rear relationship of the fans arranged at intervals in the main wind direction in the wind farm , the transfer characteristics of wind speed in the main wind direction, and the wind field operation data to capture the wind speed relationship between wind turbines to replace physical lidar to predict wind gusts.
  • the present invention uses a software method instead of lidar to predict wind gusts. Because the wind speed is transitive, the wind farm will be arranged at intervals in the main wind direction. If the front fan encounters a gust, the rear fan will also encounter the same gust after a period of time. Based on this physical The principle and observation of the wind farm operation data show that the wind speed between the fans is related and has a certain hysteresis characteristic, and the present invention uses a neural network to capture this relationship.
  • This embodiment provides a virtual lidar system based on mutual early warning of wind turbines.
  • FIG. 1 it includes: a wind farm deployment data module to obtain the front-to-back relationship of wind turbines arranged at intervals in the main wind direction in the wind farm; a wind speed transfer characteristic module , to obtain the transfer characteristics of the wind speed in the main wind direction in the area where the wind farm is located; the wind farm operation data module, to obtain the actual operation data of the wind farm; and the neural network module, according to the front and rear relationship of the fans arranged in the main wind direction in the wind farm, The transfer characteristics of wind speed in the main wind direction, as well as the wind field operation data, capture the wind speed relationship between wind turbines to replace physical lidar to predict wind gusts.
  • the neural network module includes a recurrent neural network model and a graph neural network model; The front and back relationship of the upwardly spaced fans and the transmission characteristics of the wind speed in the main wind direction are formed; the neural network module uses the actual operation data of the wind farm to train the recurrent neural network model and the graph neural network model, and performs the spatiotemporal prediction of wind speed. The actual operation data of the wind farm in other time periods are verified. Based on the cyclic neural network and the graph neural network, the model is used to train the model for the actual operation of the wind farm for spatiotemporal prediction, and the data of other time periods are used for verification. The forecast of wind speed is of great economic value in reducing fan fatigue and increasing power generation.
  • the actual operation data of the wind farm is second-level wind speed detection data;
  • the neural network module capturing the wind speed relationship between the wind turbines includes: according to the second-level wind speed detection. Data, obtain the correlation of the second-level wind speed detection data of each fan, and obtain the flow field information of the entire wind field according to the correlation of the second-level wind speed detection data of each fan.
  • the transmission characteristic of the wind speed in the main wind direction is the mutual early warning characteristic between wind turbines;
  • the neural network modules are arranged at intervals in the main wind direction in combination with the wind farm It uses time series deep learning algorithm to train the second-level wind speed detection data, establishes a specific wind field flow field model, and predicts the wind speed trend of each fan in the next tens of seconds in real time based on the specific wind field flow field model. information, and use the mutual early warning feature between wind turbines to realize virtual lidar.
  • Each wind farm has massive second-level wind speed data. The wind speed data of different planes are not isolated and irrelevant.
  • the time series deep learning algorithm can be used to train these wind speed data, so as to establish a flow field model of a specific wind field, and based on the flow field model, real-time prediction of each wind turbine in the next tens of seconds The wind speed trend information, and then use the mutual early warning between the wind turbines to realize the virtual lidar.
  • the spatiotemporal prediction of wind speed includes: Learning to find patterns, that is, using recurrent neural networks to capture the relationship before and after the time series.
  • Learning to find patterns that is, using recurrent neural networks to capture the relationship before and after the time series.
  • a structural gate regression unit commonly used in recurrent neural networks is shown in Figure 2.
  • Z t is the update gate, indicating whether the information at the last moment needs to be updated
  • r t is the reset gate, indicating whether the information at the last moment needs to be reset
  • h t is the hidden layer output of the gate regression unit, receiving the time information of h t-1 and h t ;
  • the contextual relationship on the time series is captured, and the future wind speed data is predicted through the wind speed historical data of the current wind turbine and other wind speed historical data.
  • the relationship between spatial nodes is captured through the graph neural network, and the law of data flow in the relationship between the spatial nodes is obtained.
  • the graph neural network has been used in traffic flow prediction, social network, protein structure prediction and other directions. Some people use it in the field of wind speed prediction, let alone in the scope of wind farms.
  • the cyclic neural network is used to capture the relationship in the time series of a single wind turbine
  • the graph neural network is used to capture the relationship between multiple wind turbines in the wind farm. spatial relationship. Not only use the historical data of the wind turbine itself, but also use the historical data of other wind turbines in the wind farm.
  • the wind speed of a fan at time t+ ⁇ t is predicted by the wind speed of the other fans at time t and the mutual positional relationship of the fans.
  • the wind speed of the first fan at time t is equal to the wind speed of the second fan at time t+ ⁇ t, ⁇ t ⁇ d/ws, where d is the distance between the two fans, and ws is the first fan.
  • the wind speed of a fan at time t is the above method provides a solid basis for predicting wind gusts in advance for downwind wind turbines. At the same time, all the wind turbines in the entire wind farm form a network, and the future wind speed of a certain wind turbine can be effectively predicted through the historical wind speed sequence and mutual position relationship of the remaining wind turbines.
  • wind farm time series data is collected, data cleaning, standardization, division of training data sets and test data sets, construction of wind farm network structures, and training of deep learning neural networks are performed.
  • Models, model validation, model deployment, fan control As shown in Figure 4, it specifically includes: collecting the operating wind speed data of each wind turbine in the wind farm; aligning the actual operating data of the wind speed by time, cleaning the abnormal data according to the operating state of the wind turbine to standardize the data, and subtracting the mean and dividing by the standard deviation, To ensure that the standardized data has a mean of 0 and a variance of 1:
  • the training set is used to train the network weights, and the test set does not participate in any Training is only used to calculate the evaluation index to detect the performance of the model; build a wind farm network structure, the wind farm network structure is divided into static network and dynamic network, the static network is the training process to fix the relationship matrix between fans, and the dynamic network is the training process.
  • the relationship matrix between the wind turbines is not fixed, and the relationship matrix changes with the similarity of the current wind speed time series of different wind turbines; training the deep learning neural network model, the graph convolutional neural network inputs the normalized historical wind speed data of each wind turbine and the adjacency matrix as Extract the spatial feature matrix, input the output result into the recurrent neural network model to extract the time feature matrix, and output the wind speed value of each fan in the future; determine the optimizer as Adam, and the loss function as the root mean square average error MSE; the model training set converges After that, it is verified in the test set, the average absolute error Mae index is calculated and the time sequence is drawn, and the model is deployed at the wind farm station; the virtual lidar is used to control the wind turbine.
  • the present invention can effectively predict the wind speed of each fan in the wind field for a period of time in the future through the "virtual laser radar" method.
  • the most front-row fans in the incoming wind direction mainly rely on their own historical data for prediction.
  • the rear-row fans can also rely on the information provided by the front-row fans to make predictions with higher accuracy, especially when all fans are running normally. time to effectively replace lidar.
  • the wind speed trend prediction time based on the mutual early warning model of wind turbines is much longer than the range that can be predicted by hardware lidar.
  • the wind speed is about 80-200m in front of the wind wheel, and the wind speed prediction range based on deep learning can reach more than 300m.
  • the virtual lidar based on mutual early warning of wind turbines is based on the wind speed prediction sequence of deep learning, and is equipped with model predictive control algorithm and Markov decision model to calculate the comprehensive optimal power generation and component load of wind turbines.
  • the neural network module is used according to the front and rear relationship of the wind turbines arranged at intervals in the main wind direction in the wind field, the transmission characteristics of the wind speed in the main wind direction, and the wind field.
  • the operation data captures the wind speed relationship between the wind turbines to replace the physical lidar to predict wind gusts, and realizes a virtual lidar method based on the mutual early warning of wind turbines. Wind speed prediction has great economic value in reducing fan fatigue and increasing power generation.
  • the invention proposes a virtual laser radar system and method based on mutual early warning of wind turbines.
  • the wind speed of each machine point in the wind field for a period of time in the future can be predicted.
  • the predicted second-level wind speed is applied to the real-time control of wind turbines to overcome various limitations brought by hardware lidar.
  • the present invention can realize wind speed prediction without using any physical laser radar.
  • the existing virtual laser radar is mostly based on the physical laser radar installed on the front fan to predict the wind speed ahead, and then the function of virtual laser radar can be realized.
  • Virtual LiDAR does not need to install any physical LiDAR. It uses historical wind speed data to predict the wind speed that the front fans will face, and uses the original wind speed detection equipment on the fans to detect the real-time wind speed of the fans. Combine the two to train the model, and then use the model.
  • the beneficial effects of the invention are as follows: cost saving, about 500,000 yuan per lidar can be saved; not affected by weather, high reliability, accuracy, and stable application; more comprehensive and detailed wind conditions are obtained, and the span is longer;
  • the software architecture is easy to deploy, does not require high computing resources, and has no deployment threshold; based on second-level wind speed time series prediction, it can protect wind turbines from damage from extreme loads, and effectively reduce tower fatigue loads, which will also reduce the cost of future wind turbine design. have important meaning.
  • the above embodiments describe in detail the different configurations of the virtual lidar system and method based on mutual early warning of wind turbines.
  • the present invention includes, but is not limited to, the configurations listed in the above embodiments. Contents that are transformed on the basis of the provided configuration all belong to the protection scope of the present invention. Those skilled in the art can draw inferences from the contents of the foregoing embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Electromagnetism (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Wind Motors (AREA)
  • Optical Radar Systems And Details Thereof (AREA)

Abstract

本发明提供了一种基于风机相互预警的虚拟激光雷达系统及方法，包括：风场部署数据模块，获取风场内在主风向上间隔排布的风机的前后关系；风速传递特性模块，获取风场所在的地区风速在主风向上的传递特性；风场运行数据模块，获取风场实际运行数据；以及神经网络模块，根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。

Description

基于风机相互预警的虚拟激光雷达系统及方法 技术领域
本发明涉及风力发电机技术领域，特别涉及一种基于风机相互预警的虚拟激光雷达系统及方法。
背景技术
随着能源供需的日益紧张和人类对环境气候变化问题的日益关注，众多欧美国家已将可再生清洁能源发展提高到国家战略高度。而风能作为清洁环保、可再生能源的代表，因其蕴含量大、分布范围广、开发前景好等优点，欧美发达国家均投入巨大资源在风电技术的研究与应用上。
在风电技术开发中，风速预测一直是近几年的热点问题。风速受温度、气压、地形等多种因素的影响，具有很强的随机性。风速预测按照预测时间尺度可分为长期(months～years)、中期(days～weeks)、短期(hours)、超短期(minutes)预测，这些风速预测主要为风电场的风功率预测服务，从而制定正确的调配计划，减轻风电对电网的不利影响。
目前国内外风速预测的方法虽然多种多样，但多集中在分钟级以上时间尺度的预测，比如卡尔曼滤波方法、时间序列分析方法、神经网络方法和模糊逻辑方法等。这些风速预测方法基本上是通过对风电场的原始风速时间序列建立模型，然后对未来风速进行预测，或者通过差分等手段，把非平稳的风速序列先转换为平稳序列，再对平稳序列进行建模和预测。但是对于秒级风速预测，相对可靠的预测方法并不多见。在IEEE发布的“The Impact of Control Technology—2nd Edition”中，基于未来秒级风速时序的提前规划与控制被当做未来风机控制研究的重大挑战提了出来。
近些年，出现了使用激光雷达进行秒级风速预测的方法，激光雷达通过向前方打出激光，测出风机前几百米处立体位面的风速情况，从而可以提前预测未来几秒的风速情况，但是由于价格昂贵、受天气影响大、不稳 定、标定难设置等原因，实际使用效果不佳。
综上所述，物理激光雷达在风场实际使用中效果很差，难以应用，且(1)物理激光雷达价格昂贵；(2)物理激光雷达受天气影响大；(3)物理激光雷达预报风速时间短，大约在10秒左右；(4)物理激光雷达标定难设置。
发明内容
本发明的目的在于提供一种基于风机相互预警的虚拟激光雷达系统及方法，以解决现有的物理激光雷达在风场实际使用中效果很差的问题。
为解决上述技术问题，本发明提供一种基于风机相互预警的虚拟激光雷达系统，包括：
风场部署数据模块，获取风场内在主风向上间隔排布的风机的前后关系；由于硬件激光雷达的使用效果不佳，本发明使用软件方法来代替激光雷达以预测阵风。因为风速具有传递性，风场部署时会在主风向上有前后关系地间隔排布，如果前排风机遇到了阵风(gust)，后排风机过一段时间也会遇到相同阵风，基于该物理原理以及对风场运行数据的观察可知，风机间的风速是相关的且具有一定滞后特性，本发明采用神经网络来捕获这种关系。
风速传递特性模块，获取风场所在的地区风速在主风向上的传递特性；
风场运行数据模块，获取风场实际运行数据；以及
神经网络模块，根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，所述神经网络模块包括循环神经网络模型和图神经网络模型；
循环神经网络模型和图神经网络模型为根据风场内在主风向上间隔排布的风机的前后关系、以及风速在主风向上的传递特性形成；
所述神经网络模块利用风场实际运行数据训练循环神经网络模型和图神经网络模型，进行风速时空预测，并使用其它时间段的风场实际运行数 据进行验证。基于循环神经网络和图神经网络，利用风场实际运行数据训练模型进行时空预测，并使用其它时间段数据进行验证，结果表明本发明实用性超过物理激光雷达，可以在实际风场中有效进行秒级风速预测，对减少风机疲劳及提升发电量有巨大经济价值。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，所述风场实际运行数据为秒级风速检测数据；
所述神经网络模块捕获风机间的风速关系包括：根据秒级风速检测数据，获取各个风机的秒级风速检测数据的相关性，并根据各个风机的秒级风速检测数据的相关性，获取整个风场的流场信息。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，所述风速在主风向上的传递特性为风机间的相互预警特性；
所述神经网络模块结合风场内在主风向上间隔排布的风机的前后关系，利用时间序列深度学习算法对秒级风速检测数据进行训练，建立特定风场流场模型，并基于特定风场流场模型实时预测每台风机在未来数十秒的风速趋势信息，并利用风机间的相互预警特性实现虚拟激光雷达。
每个风场都有海量的秒级风速数据，不同机位的风速数据并不是孤立互不相关的，在这些海量数据中，其相互关系已经描述了整个风电场的流场信息。因此结合风电场的机位排布数据，可以利用时间序列深度学习算法对这些风速数据进行训练，从而建立特定风场的流场模型，并基于流场模型实时预测每台风机在未来几十秒的风速趋势信息，进而利用风机间的相互预警来实现虚拟激光雷达。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，进行风速时空预测包括：在具体的技术领域，在传统时序预测问题上，常常使用自身的历史数据进行学习以寻找出规律，即利用循环神经网络捕捉时间序列前后的关系。例如循环神经网络中常用的一种结构门回归单元。
使用本风机风速历史数据和其他风机风速历史数据进行深度学习，利用循环神经网络门回归单元捕捉时间序列前后关系，包括：
z t＝σ(W zx t+U zh t-1)
r t＝σ(W tx t+U th t-1)
Figure PCTCN2021090214-appb-000001
Figure PCTCN2021090214-appb-000002
Z t是更新门，表示上一个时刻信息是否需要更新；
r t是重置门，表示上一个时刻信息是否需要重置；
Figure PCTCN2021090214-appb-000003
是候选输出，接收x t、h t-1的时刻信息；
h t是门回归单元的隐层输出，接收h t-1、h t的时刻信息；
通过门回归单元的训练捕捉时间序列上的前后关系，通过本风机风速历史数据和其他风机风速历史数据预测未来风速数据。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，进行风速时空预测还包括：在空间结构问题上，近两年出现了图神经网络GNN的方法，即对于图G＝(V，E)，V是图中的边集，E为图中的节点集，E中每一个节点的属性都受到相连节点属性的影响；
X是原节点属性，A是邻接矩阵，Z是变换后节点属性，I是对角矩阵，对于一个两层的图神经网络，激活函数分别采用ReLU和Softmax；则
Figure PCTCN2021090214-appb-000004
Figure PCTCN2021090214-appb-000005
通过图神经网络捕捉空间节点之间的相互关系，获取空间节点之间的相互关系中数据流动的规律；目前，图神经网络已经用于交通流预测、社交网络、蛋白质结构预测等方向，还未有人在风速预测领域使用，更未在风电场范围内使用。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，不仅利用了循环神经网络来捕捉单个风机时间序列上的关系，还利用了图神经网络来捕捉风场内多台风机之间空间结构上的关系。不仅使用风机自身历史数据，还使用风场内其它风机的历史数据，通过观察临近风机的风速时序，获取明显的滞后情况，即上风向风机当前t时刻的风速，在下风向风机t+Δt时刻也会出现，尤其是阵风情况下，其中Δt取决于两台风机之间的距离d与当前风速ws，
通过其余风机的t时刻风速及风机相互位置关系预测某台风机t+Δt时刻的风速，在某一风向上，除去风速衰减，第一风机在t时刻的风速等于第二风机在t+Δt时刻的风速，Δt≈d/ws，其中，d为两台风机之间的距离，ws为第一风机在t时刻的风速。上述方法为下风向的风机提供了提前预测 阵风的坚实基础。同时整个风场的所有风机构成了一个网络，通过其余风机的历史风速时序及相互位置关系可以有效预测某台风机未来风速。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，收集风场时序数据，进行数据清理、标准化、划分训练数据集和测试数据集、构建风场网络结构、训练深度学习神经网络模型，模型验证、模型部署、风机控制。具体包括：收集风场各台风机运行风速数据；
将风速实际运行数据按时间对齐，根据风机运行状态清洗异常数据，以使数据标准化，采用减去均值除以标准差，以保证标准化的数据均值为0，方差为1：
Figure PCTCN2021090214-appb-000006
按照时间先后划分训练集与测试集，其中训练集的数据占风速实际运行数据的80％，测试集的数据占风速实际运行数据的20％，训练集用来训练网络权重，测试集不参与任何训练，只用于计算评价指标检测模型性能；
构建风场网络结构，所述风场网络结构分为静态网络和动态网络，静态网络为训练过程固定风机之间的关系矩阵，动态网络为训练过程中不固定风机间的关系矩阵，该关系矩阵随着不同风机当前风速时序的相似性而变化；
训练深度学习神经网络模型，图卷积神经网络输入标准化后的各风机历史风速数据及邻接矩阵以提取空间特征矩阵，将输出结果输入到循环神经网络模型以提取时间特征矩阵，并输出各风机未来时刻的风速值；确定优化器为Adam，损失函数为均方根平均误差MSE；
模型训练集收敛后，在测试集进行验证，计算平均绝对误差Mae指标并画时序，将模型部署在风场站端；
利用虚拟激光雷达进行风机控制。
通过“虚拟激光雷达”方法，可以有效预测风场内各风机未来一段时间的风速。原理上，来风向上最前排风机主要依靠自身历史数据进行预测，后排风机除了自身历史规律外，还可以依靠前排风机提供的信息进行预测，精度更高，尤其是所有风机都正常运行的时候，以有效代替激光雷达。
可选的，在所述的基于风机相互预警的虚拟激光雷达系统中，基于风 机相互预警模型的风速趋势预测时间长度远比硬件激光雷达可预测的范围更长，通常的机舱式激光雷达风速测量范围在风轮前80～200m左右，而基于深度学习的风速预测范围可达300m以上。对于风机来说，风速时序预测范围越长，越适合其作为单个智能体做出未来的控制行为决策。基于风机相互预警的虚拟激光雷达基于深度学习的风速预测时序，并配以模型预测控制算法和马尔科夫决策模型，计算风电机组的发电量和部件载荷综合最优。
本发明还提供一种基于风机相互预警的虚拟激光雷达方法，包括：
风场部署数据模块获取风场内在主风向上间隔排布的风机的前后关系；
风速传递特性模块获取风场所在的地区风速在主风向上的传递特性；
风场运行数据模块获取风场实际运行数据；以及
神经网络模块根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。
在本发明提供的基于风机相互预警的虚拟激光雷达系统及方法中，通过神经网络模块根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风，实现了一种基于风机相互预警的虚拟激光雷达方法，该方法实用性超过激光雷达，可以在实际风场中有效进行秒级风速预测，对减少风机疲劳及提升发电量有巨大经济价值。
本发明提出了一种基于风机相互预警的虚拟激光雷达系统及方法，通过对历史风速数据的深度学习，对未来一段时间的风场各机位点风速进行预测。同时，将预测的秒级风速应用于风机实时控制，克服硬件激光雷达带来的种种限制。
本发明的有益效果在于：节约成本，可省去每台激光雷达约50万元成本；不受天气影响，可靠性高，准确，稳定应用；得到风况更全面，更详细，跨度更长；该软件架构部署方便，对计算资源要求不高，无部署门槛；基于秒级风速时序预测，可以保护风机不受极限载荷破坏，并且有效降低塔筒疲劳载荷，对未来风电机组设计的成本降低同样有重要意义。
附图说明
图1是本发明一实施例基于风机相互预警的虚拟激光雷达方法示意图；
图2是本发明一实施例基于风机相互预警的虚拟激光雷达系统门回归单元示意图；
图3是本发明一实施例基于风机相互预警的虚拟激光雷达系统图神经网络示意图；
图4是本发明一实施例基于风机相互预警的虚拟激光雷达方法示意图。
具体实施方式
以下结合附图和具体实施例对本发明提出的基于风机相互预警的虚拟激光雷达系统及方法作进一步详细说明。根据下面说明和权利要求书，本发明的优点和特征将更清楚。需说明的是，附图均采用非常简化的形式且均使用非精准的比例，仅用以方便、明晰地辅助说明本发明实施例的目的。
另外，除非另行说明，本发明的不同实施例中的特征可以相互组合。例如，可以用第二实施例中的某特征替换第一实施例中相对应或功能相同或相似的特征，所得到的实施例同样落入本申请的公开范围或记载范围。
本发明的核心思想在于提供一种基于风机相互预警的虚拟激光雷达系统及方法，以解决现有的物理激光雷达在风场实际使用中效果很差的问题。
为实现上述思想，本发明提供了一种基于风机相互预警的虚拟激光雷达系统及方法，包括：风场部署数据模块，获取风场内在主风向上间隔排布的风机的前后关系；风速传递特性模块，获取风场所在的地区风速在主风向上的传递特性；风场运行数据模块，获取风场实际运行数据；以及神经网络模块，根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。
由于硬件激光雷达的使用效果不佳，本发明使用软件方法来代替激光雷达以预测阵风。因为风速具有传递性，风场部署时会在主风向上有前后 关系地间隔排布，如果前排风机遇到了阵风(gust)，后排风机过一段时间也会遇到相同阵风，基于该物理原理以及对风场运行数据的观察可知，风机间的风速是相关的且具有一定滞后特性，本发明采用神经网络来捕获这种关系。
本实施例提供一种基于风机相互预警的虚拟激光雷达系统，如图1所示，包括：风场部署数据模块，获取风场内在主风向上间隔排布的风机的前后关系；风速传递特性模块，获取风场所在的地区风速在主风向上的传递特性；风场运行数据模块，获取风场实际运行数据；以及神经网络模块，根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。
具体的，在所述的基于风机相互预警的虚拟激光雷达系统中，所述神经网络模块包括循环神经网络模型和图神经网络模型；循环神经网络模型和图神经网络模型为根据风场内在主风向上间隔排布的风机的前后关系、以及风速在主风向上的传递特性形成；所述神经网络模块利用风场实际运行数据训练循环神经网络模型和图神经网络模型，进行风速时空预测，并使用其它时间段的风场实际运行数据进行验证。基于循环神经网络和图神经网络，利用风场实际运行数据训练模型进行时空预测，并使用其它时间段数据进行验证，结果表明本发明实用性超过物理激光雷达，可以在实际风场中有效进行秒级风速预测，对减少风机疲劳及提升发电量有巨大经济价值。
进一步的，在所述的基于风机相互预警的虚拟激光雷达系统中，所述风场实际运行数据为秒级风速检测数据；所述神经网络模块捕获风机间的风速关系包括：根据秒级风速检测数据，获取各个风机的秒级风速检测数据的相关性，并根据各个风机的秒级风速检测数据的相关性，获取整个风场的流场信息。
另外，在所述的基于风机相互预警的虚拟激光雷达系统中，所述风速在主风向上的传递特性为风机间的相互预警特性；所述神经网络模块结合风场内在主风向上间隔排布的风机的前后关系，利用时间序列深度学习算 法对秒级风速检测数据进行训练，建立特定风场流场模型，并基于特定风场流场模型实时预测每台风机在未来数十秒的风速趋势信息，并利用风机间的相互预警特性实现虚拟激光雷达。每个风场都有海量的秒级风速数据，不同机位的风速数据并不是孤立互不相关的，在这些海量数据中，其相互关系已经描述了整个风电场的流场信息。因此结合风电场的机位排布数据，可以利用时间序列深度学习算法对这些风速数据进行训练，从而建立特定风场的流场模型，并基于流场模型实时预测每台风机在未来几十秒的风速趋势信息，进而利用风机间的相互预警来实现虚拟激光雷达。
在本发明的一个实施例中，在所述的基于风机相互预警的虚拟激光雷达系统中，进行风速时空预测包括：在具体的技术领域，在传统时序预测问题上，常常使用自身的历史数据进行学习以寻找出规律，即利用循环神经网络捕捉时间序列前后的关系。例如循环神经网络中常用的一种结构门回归单元，如图2所示。使用本风机风速历史数据和其他风机风速历史数据进行深度学习，利用循环神经网络门回归单元捕捉时间序列前后关系，包括：
z t＝σ(W zx t+U zh t-1)
r t＝σ(W tx t+U th t-1)
Figure PCTCN2021090214-appb-000007
Figure PCTCN2021090214-appb-000008
Z t是更新门，表示上一个时刻信息是否需要更新；
r t是重置门，表示上一个时刻信息是否需要重置；
Figure PCTCN2021090214-appb-000009
是候选输出，接收x t、h t-1的时刻信息；
h t是门回归单元的隐层输出，接收h t-1、h t的时刻信息；
通过门回归单元的训练捕捉时间序列上的前后关系，通过本风机风速历史数据和其他风机风速历史数据预测未来风速数据。
在本发明的另一个实施例中，在所述的基于风机相互预警的虚拟激光雷达系统中，进行风速时空预测还包括：在空间结构问题上，近两年出现了图神经网络GNN的方法，如图3所示，即对于图G＝(V，E)，V是图中的边集，E为图中的节点集，E中每一个节点的属性都受到相连节点属性的影响；X是原节点属性，A是邻接矩阵，Z是变换后节点属性，I是对角 矩阵，对于一个两层的图神经网络，激活函数分别采用ReLU和Softmax；则
Figure PCTCN2021090214-appb-000010
Figure PCTCN2021090214-appb-000011
通过图神经网络捕捉空间节点之间的相互关系，获取空间节点之间的相互关系中数据流动的规律；目前，图神经网络已经用于交通流预测、社交网络、蛋白质结构预测等方向，还未有人在风速预测领域使用，更未在风电场范围内使用。
具体的，在所述的基于风机相互预警的虚拟激光雷达系统中，不仅利用了循环神经网络来捕捉单个风机时间序列上的关系，还利用了图神经网络来捕捉风场内多台风机之间空间结构上的关系。不仅使用风机自身历史数据，还使用风场内其它风机的历史数据，通过观察临近风机的风速时序，获取明显的滞后情况，即上风向风机当前t时刻的风速，在下风向风机t+Δt时刻也会出现，尤其是阵风情况下，其中Δt取决于两台风机之间的距离d与当前风速ws，通过其余风机的t时刻风速及风机相互位置关系预测某台风机t+Δt时刻的风速，在某一风向上，除去风速衰减，第一风机在t时刻的风速等于第二风机在t+Δt时刻的风速，Δt≈d/ws，其中，d为两台风机之间的距离，ws为第一风机在t时刻的风速。上述方法为下风向的风机提供了提前预测阵风的坚实基础。同时整个风场的所有风机构成了一个网络，通过其余风机的历史风速时序及相互位置关系可以有效预测某台风机未来风速。
进一步的，在所述的基于风机相互预警的虚拟激光雷达系统中，收集风场时序数据，进行数据清理、标准化、划分训练数据集和测试数据集、构建风场网络结构、训练深度学习神经网络模型，模型验证、模型部署、风机控制。如图4所示，具体包括：收集风场各台风机运行风速数据；将风速实际运行数据按时间对齐，根据风机运行状态清洗异常数据，以使数据标准化，采用减去均值除以标准差，以保证标准化的数据均值为0，方差为1：
Figure PCTCN2021090214-appb-000012
按照时间先后划分训练集与测试集，其中训练集的数据占风速实际运行数据的80％，测试集的数据占风速实际运行数据的20％，训练集用来训练网络权重，测试集不参与任何训练，只用于计算评价指标检测模型性能；构建风场网络结构，所述风场网络结构分为静态网络和动态网络，静态网络为训练过程固定风机之间的关系矩阵，动态网络为训练过程中不固定风机间的关系矩阵，该关系矩阵随着不同风机当前风速时序的相似性而变化；训练深度学习神经网络模型，图卷积神经网络输入标准化后的各风机历史风速数据及邻接矩阵以提取空间特征矩阵，将输出结果输入到循环神经网络模型以提取时间特征矩阵，并输出各风机未来时刻的风速值；确定优化器为Adam，损失函数为均方根平均误差MSE；模型训练集收敛后，在测试集进行验证，计算平均绝对误差Mae指标并画时序，将模型部署在风场站端；利用虚拟激光雷达进行风机控制。
本发明通过“虚拟激光雷达”方法，可以有效预测风场内各风机未来一段时间的风速。原理上，来风向上最前排风机主要依靠自身历史数据进行预测，后排风机除了自身历史规律外，还可以依靠前排风机提供的信息进行预测，精度更高，尤其是所有风机都正常运行的时候，以有效代替激光雷达。
另外，在所述的基于风机相互预警的虚拟激光雷达系统中，基于风机相互预警模型的风速趋势预测时间长度远比硬件激光雷达可预测的范围更长，通常的机舱式激光雷达风速测量范围在风轮前80～200m左右，而基于深度学习的风速预测范围可达300m以上。对于风机来说，风速时序预测范围越长，越适合其作为单个智能体做出未来的控制行为决策。基于风机相互预警的虚拟激光雷达基于深度学习的风速预测时序，并配以模型预测控制算法和马尔科夫决策模型，计算风电机组的发电量和部件载荷综合最优。
在本发明提供的基于风机相互预警的虚拟激光雷达系统及方法中，通过神经网络模块根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风，实现了一种基于风机相互预警的虚拟激光雷 达方法，该方法实用性超过激光雷达，可以在实际风场中有效进行秒级风速预测，对减少风机疲劳及提升发电量有巨大经济价值。
本发明提出了一种基于风机相互预警的虚拟激光雷达系统及方法，通过对历史风速数据的深度学习，对未来一段时间的风场各机位点风速进行预测。同时，将预测的秒级风速应用于风机实时控制，克服硬件激光雷达带来的种种限制。
本发明可以实现不采用任何物理激光雷达来实现风速预测，现有的虚拟激光雷达多基于前排风机安装的物理激光雷达来预测前方风速后，才能实现虚拟激光雷达的功能，而本发明中的虚拟激光雷达无需安装任何物理激光雷达，通过风速历史数据预测前排风机即将面对的风速，并采用风机上原有的风速检测设备，检测风机实时风速，将两者结合训练模型，然后采用该模型来预测前排风机的未来风速，避免了任何物理激光雷达的安装，就可以避免物理激光雷达的全部缺点，而现有技术只要采用一台物理激光雷达，就难免会遇到价格昂贵、受天气影响大、预报风速时间短、标定难设置的难题，本发明的重大意义在于避免任何物理激光雷达的安装。
本发明的有益效果在于：节约成本，可省去每台激光雷达约50万元成本；不受天气影响，可靠性高，准确，稳定应用；得到风况更全面，更详细，跨度更长；该软件架构部署方便，对计算资源要求不高，无部署门槛；基于秒级风速时序预测，可以保护风机不受极限载荷破坏，并且有效降低塔筒疲劳载荷，对未来风电机组设计的成本降低同样有重要意义。
综上，上述实施例对基于风机相互预警的虚拟激光雷达系统及方法的不同构型进行了详细说明，当然，本发明包括但不局限于上述实施中所列举的构型，任何在上述实施例提供的构型基础上进行变换的内容，均属于本发明所保护的范围。本领域技术人员可以根据上述实施例的内容举一反三。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。对于实施例公开的系统而言，由于与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
上述描述仅是对本发明较佳实施例的描述，并非对本发明范围的任何限定，本发明领域的普通技术人员根据上述揭示内容做的任何变更、修饰，均属于权利要求书的保护范围。

Claims (10)

  1. 一种基于风机相互预警的虚拟激光雷达系统，其特征在于，包括：
    风场部署数据模块，被配置为获取风场内在主风向上间隔排布的风机的前后关系；
    风速传递特性模块，被配置为获取风场所在的地区风速在主风向上的传递特性；
    风场运行数据模块，被配置为获取风场实际运行数据；以及
    神经网络模块，被配置为根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。
  2. 如权利要求1所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，所述神经网络模块包括循环神经网络模型和图神经网络模型；
    循环神经网络模型和图神经网络模型为根据风场内在主风向上间隔排布的风机的前后关系、以及风速在主风向上的传递特性形成；
    所述神经网络模块利用风场实际运行数据训练循环神经网络模型和图神经网络模型，进行风速时空预测，并使用其它时间段的风场实际运行数据进行验证。
  3. 如权利要求2所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，所述风场实际运行数据为秒级风速检测数据；
    所述神经网络模块捕获风机间的风速关系包括：根据秒级风速检测数据，获取各个风机的秒级风速检测数据的相关性，并根据各个风机的秒级风速检测数据的相关性，获取整个风场的流场信息。
  4. 如权利要求3所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，所述风速在主风向上的传递特性为风机间的相互预警特性；
    所述神经网络模块结合风场内在主风向上间隔排布的风机的前后关系，利用时间序列深度学习算法对秒级风速检测数据进行训练，建立特定风场流场模型，并基于特定风场流场模型实时预测每台风机在未来数十秒的风速趋势信息，并利用风机间的相互预警特性实现虚拟激光雷达。
  5. 如权利要求2所述的基于风机相互预警的虚拟激光雷达系统，其特 征在于，进行风速时空预测包括：
    使用本风机风速历史数据和其他风机风速历史数据进行深度学习，利用循环神经网络门回归单元捕捉时间序列前后关系，包括：
    z t＝σ(W zx t+U zh t-1)
    r t＝σ(W tx t+U th t-1)
    Figure PCTCN2021090214-appb-100001
    Figure PCTCN2021090214-appb-100002
    Z t是更新门，表示上一个时刻信息是否需要更新；
    r t是重置门，表示上一个时刻信息是否需要重置；
    Figure PCTCN2021090214-appb-100003
    是候选输出，接收x t、h t-1的时刻信息；
    h t是门回归单元的隐层输出，接收h t-1、h t的时刻信息；
    通过门回归单元的训练捕捉时间序列上的前后关系，通过本风机风速历史数据和其他风机风速历史数据预测未来风速数据。
  6. 如权利要求2所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，进行风速时空预测还包括：
    通过图神经网络捕捉空间节点之间的相互关系，获取空间节点之间的相互关系中数据流动的规律；
    G＝(V，E)，V是图中的边集，E为图中的节点集，E中每一个节点的属性受到相连节点属性的影响；
    X是原节点属性，A是邻接矩阵，Z是变换后节点属性，I是对角矩阵，对于一个两层的图神经网络，激活函数分别采用ReLU和Softmax；
    Figure PCTCN2021090214-appb-100004
    Figure PCTCN2021090214-appb-100005
  7. 如权利要求1所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，通过其余风机的t时刻风速及风机相互位置关系预测某台风机t+Δt时刻的风速，在某一风向上，除去风速衰减，第一风机在t时刻的风速等于第二风机在t+Δt时刻的风速，Δt≈d/ws，其中，d为两台风机之间的距离，ws为第一风机在t时刻的风速。
  8. 如权利要求1所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，将风速实际运行数据按时间对齐，根据风机运行状态清洗异常数据，以使数据标准化，采用减去均值除以标准差，以保证标准化的数据均 值为0，方差为1：
    Figure PCTCN2021090214-appb-100006
    按照时间先后划分训练集与测试集，其中训练集的数据占风速实际运行数据的80％，测试集的数据占风速实际运行数据的20％，训练集用来训练网络权重，测试集用于计算评价指标检测模型性能；
    构建风场网络结构，所述风场网络结构分为静态网络和动态网络，静态网络为训练过程固定风机之间的关系矩阵，动态网络为训练过程中不固定风机间的关系矩阵，该关系矩阵随着不同风机当前风速时序的相似性而变化；
    训练深度学习神经网络模型，图卷积神经网络输入标准化后的各风机历史风速数据及邻接矩阵以提取空间特征矩阵，将输出结果输入到循环神经网络模型以提取时间特征矩阵，并输出各风机未来时刻的风速值；确定优化器为Adam，损失函数为均方根平均误差MSE；
    模型训练集收敛后，在测试集进行验证，计算平均绝对误差Mae指标并画时序，将模型部署在风场站端；
    利用虚拟激光雷达进行风机控制。
  9. 如权利要求1所述的基于风机相互预警的虚拟激光雷达系统，其特征在于，基于风机相互预警的虚拟激光雷达基于深度学习的风速预测时序，并配以模型预测控制算法和马尔科夫决策模型，计算风电机组的发电量和部件载荷综合最优。
  10. 一种基于风机相互预警的虚拟激光雷达方法，其特征在于，包括：
    风场部署数据模块获取风场内在主风向上间隔排布的风机的前后关系；
    风速传递特性模块获取风场所在的地区风速在主风向上的传递特性；
    风场运行数据模块获取风场实际运行数据；以及
    神经网络模块根据风场内在主风向上间隔排布的风机的前后关系、风速在主风向上的传递特性、以及风场运行数据，捕获风机间的风速关系，以代替物理激光雷达预测阵风。
PCT/CN2021/090214 2020-11-16 2021-04-27 基于风机相互预警的虚拟激光雷达系统及方法 WO2022100021A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011274875.2A CN112215305A (zh) 2020-11-16 2020-11-16 基于风机相互预警的虚拟激光雷达系统及方法
CN202011274875.2 2020-11-16

Publications (1)

Publication Number Publication Date
WO2022100021A1 true WO2022100021A1 (zh) 2022-05-19

Family

ID=74057093

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/090214 WO2022100021A1 (zh) 2020-11-16 2021-04-27 基于风机相互预警的虚拟激光雷达系统及方法

Country Status (2)

Country Link
CN (1) CN112215305A (zh)
WO (1) WO2022100021A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115905997A (zh) * 2022-10-28 2023-04-04 安徽省国家电投和新电力技术研究有限公司 基于预测偏差优化的风电机组气象灾害预警方法及系统
CN115983494A (zh) * 2023-02-10 2023-04-18 广东工业大学 一种新建小样本风电场的短期风电功率预测方法及系统
CN115977874A (zh) * 2023-01-09 2023-04-18 中电投新疆能源化工集团木垒新能源有限公司 基于激光测风雷达的风电机组偏航自适应校准方法及系统
CN116292097A (zh) * 2023-05-17 2023-06-23 安徽省国家电投和新电力技术研究有限公司 基于激光雷达智能感知的风机组控制方法及系统
CN116451608A (zh) * 2023-04-18 2023-07-18 中广核风电有限公司 一种复杂地形的混合风功率预测方法及装置
CN117233869A (zh) * 2023-11-15 2023-12-15 南京信息工程大学 一种基于GRU-BiTCN的站点短期风速预测方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215305A (zh) * 2020-11-16 2021-01-12 远景能源有限公司 基于风机相互预警的虚拟激光雷达系统及方法
CN113411549B (zh) * 2021-06-11 2022-09-06 上海兴容信息技术有限公司 一种判断目标门店的业务是否正常的方法
CN115455368B (zh) * 2022-11-09 2023-02-03 中国人民解放军国防科技大学 一种基于湍流强度和平均风速的对流层低层阵风计算方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120185414A1 (en) * 2010-12-15 2012-07-19 Vaisala, Inc. Systems and methods for wind forecasting and grid management
US20140336934A1 (en) * 2013-05-07 2014-11-13 Atomic Energy Council - Institute Of Nuclear Energy Research Ensemble wind power forecasting platform system and operational method thereof
CN110889535A (zh) * 2019-10-28 2020-03-17 国网江西省电力有限公司电力科学研究院 一种基于卷积循环神经网络的风电场内多点位风速预测方法
CN111784041A (zh) * 2020-06-28 2020-10-16 中国电力科学研究院有限公司 一种基于图卷积神经网络的风电功率预测方法及系统
CN112215305A (zh) * 2020-11-16 2021-01-12 远景能源有限公司 基于风机相互预警的虚拟激光雷达系统及方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6975925B1 (en) * 2002-03-19 2005-12-13 Windlynx Systems, B.V. Forecasting an energy output of a wind farm
US20200225655A1 (en) * 2016-05-09 2020-07-16 Strong Force Iot Portfolio 2016, Llc Methods, systems, kits and apparatuses for monitoring and managing industrial settings in an industrial internet of things data collection environment
CN108717195B (zh) * 2018-05-24 2020-12-25 远景能源有限公司 一种相干多普勒测风激光雷达系统及其控制方法
CN109063930A (zh) * 2018-08-30 2018-12-21 上海电力学院 一种基于聚类分析的动态风电场总功率预测方法
CN109784563B (zh) * 2019-01-18 2023-05-23 南方电网科学研究院有限责任公司 一种基于虚拟测风塔技术的超短期功率预测方法
CN110735769A (zh) * 2019-09-18 2020-01-31 西安察柏科技咨询有限公司 一种预测风机故障的方法及装置、系统
CN111709490B (zh) * 2020-06-24 2022-03-18 河北工业大学 一种基于gru神经网络的风机健康状态评估方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120185414A1 (en) * 2010-12-15 2012-07-19 Vaisala, Inc. Systems and methods for wind forecasting and grid management
US20140336934A1 (en) * 2013-05-07 2014-11-13 Atomic Energy Council - Institute Of Nuclear Energy Research Ensemble wind power forecasting platform system and operational method thereof
CN110889535A (zh) * 2019-10-28 2020-03-17 国网江西省电力有限公司电力科学研究院 一种基于卷积循环神经网络的风电场内多点位风速预测方法
CN111784041A (zh) * 2020-06-28 2020-10-16 中国电力科学研究院有限公司 一种基于图卷积神经网络的风电功率预测方法及系统
CN112215305A (zh) * 2020-11-16 2021-01-12 远景能源有限公司 基于风机相互预警的虚拟激光雷达系统及方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WANG CHEN, KOU PENG: "Wind Speed Forecasts of Multiple Wind Turbines in a Wind Farm Based on Integration Model Built by Convolutional Neural Network and Simple Recurrent Unit", TRANSACTIONS OF CHINA ELECTROTECHNICAL SOCIETY, vol. 35, no. 13, 10 July 2020 (2020-07-10), pages 2723 - 2735, XP055932155, ISSN: 1000-6753, DOI: 10.19595/j.cnki.1000-6753.tces.191715 *

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115905997A (zh) * 2022-10-28 2023-04-04 安徽省国家电投和新电力技术研究有限公司 基于预测偏差优化的风电机组气象灾害预警方法及系统
CN115977874A (zh) * 2023-01-09 2023-04-18 中电投新疆能源化工集团木垒新能源有限公司 基于激光测风雷达的风电机组偏航自适应校准方法及系统
CN115977874B (zh) * 2023-01-09 2024-03-19 中电投新疆能源化工集团木垒新能源有限公司 基于激光测风雷达的风电机组偏航自适应校准方法及系统
CN115983494A (zh) * 2023-02-10 2023-04-18 广东工业大学 一种新建小样本风电场的短期风电功率预测方法及系统
CN115983494B (zh) * 2023-02-10 2023-09-12 广东工业大学 一种新建小样本风电场的短期风电功率预测方法及系统
CN116451608A (zh) * 2023-04-18 2023-07-18 中广核风电有限公司 一种复杂地形的混合风功率预测方法及装置
CN116292097A (zh) * 2023-05-17 2023-06-23 安徽省国家电投和新电力技术研究有限公司 基于激光雷达智能感知的风机组控制方法及系统
CN116292097B (zh) * 2023-05-17 2023-08-18 安徽省国家电投和新电力技术研究有限公司 基于激光雷达智能感知的风机组控制方法及系统
CN117233869A (zh) * 2023-11-15 2023-12-15 南京信息工程大学 一种基于GRU-BiTCN的站点短期风速预测方法
CN117233869B (zh) * 2023-11-15 2024-02-23 南京信息工程大学 一种基于GRU-BiTCN的站点短期风速预测方法

Also Published As

Publication number Publication date
CN112215305A (zh) 2021-01-12

Similar Documents

Publication Publication Date Title
WO2022100021A1 (zh) 基于风机相互预警的虚拟激光雷达系统及方法
Li et al. Short-term wind power prediction based on extreme learning machine with error correction
Zheng et al. Short-term wind power forecasting using a double-stage hierarchical ANFIS approach for energy management in microgrids
WO2023004838A1 (zh) 一种风电出力区间预测方法
Wang et al. Dynamic spatio-temporal correlation and hierarchical directed graph structure based ultra-short-term wind farm cluster power forecasting method
CN108985965A (zh) 一种结合神经网络和参数估计的光伏功率区间预测方法
Cui et al. An algorithm for forecasting day-ahead wind power via novel long short-term memory and wind power ramp events
Xu et al. Correlation based neuro-fuzzy Wiener type wind power forecasting model by using special separate signals
Pandit et al. Accounting for environmental conditions in data-driven wind turbine power models
CN105512766A (zh) 一种风电场功率预测方法
Wang et al. Wind power curve modeling with asymmetric error distribution
CN106611243A (zh) 一种基于garch模型的风速预测残差修正方法
CN110991701A (zh) 一种基于数据融合的风电场风机风速预测方法及系统
CN106600055A (zh) 一种基于自激励门限自回归模型的风速预测方法
Lai et al. Sub-region division based short-term regional distributed PV power forecasting method considering spatio-temporal correlations
Yu et al. Ultra-short-term wind power subsection forecasting method based on extreme weather
Zhou et al. Modeling of wind turbine power curve based on Gaussian process
Sun et al. A multi-model-integration-based prediction methodology for the spatiotemporal distribution of vulnerabilities in integrated energy systems under the multi-type, imbalanced, and dependent input data scenarios
Ding et al. Artificial intelligence based abnormal detection system and method for wind power equipment
Hu et al. Operational reliability evaluation method based on big data technology
CN109829572B (zh) 雷电气候下的光伏发电功率预测方法
CN115640737A (zh) 一种面向异常天气状态的风电功率预测方法及系统
Zhao et al. Short-term Wind Power Prediction Method Based on GCN-LSTM
Li et al. A gray rbf model improved by genetic algorithm for electrical power forecasting
Huang et al. Theory-guided Output Feedback Neural Network (Tg-OFNN) for Short-term Wind Power Forecasting

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21890556

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21890556

Country of ref document: EP

Kind code of ref document: A1