Offshore wind power prediction management system

Abstract

The application discloses a kind of offshore wind power prediction management systems, it is related to electric power system technical field.The system, including main server wind power prediction sub-server, wind farm monitoring module, large-scale data weather forecast module and interactive terminal;The main server includes wind farm data analysis sub-server, scale reduction cluster sub-server and relational database;The wind farm data analysis sub-server is used to provide wind farm data required for the wind power prediction of the wind power prediction sub-server;The scale reduction cluster sub-server provides the environmental data required for the wind power prediction of the wind power prediction sub-server;Wind farm monitoring module is used to obtain the unit data of offshore wind turbine unit, and the large-scale data weather forecast module is used to obtain wind speed, temperature, wind direction, atmospheric stability and wake effect data.The application can provide weather change basis for offshore wind power prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular to a sea wind power prediction management system. BACKGROUND

[0002] The wind power prediction system is a high-tech system with a complex infrastructure structure for providing a 15min-4h and 24h within the next day in 15min time resolution of wind farm output power prediction information infrastructure. The current rapid development of wind power has made the wind power prediction system a necessary system for wind farms. The sea wind resource is abundant, the wind turbine model is large, and the power generation is in the megawatt level. Accurate sea wind power prediction is beneficial to the safe operation of the power grid and higher economic benefits.

[0003] For the prediction of sea wind power, the traditional prediction model does not calculate the output power difference caused by the change of wind direction and atmospheric conditions. The power generation of the wind farm not only depends on the meteorological conditions such as wind speed and temperature, but also is related to the complex atmospheric interaction at different time and space scales inside and outside the wind farm. Compared with onshore wind power, sea wind farms are concentrated in a region, resulting in a higher density of wind farms. Therefore, the change of atmospheric conditions is very important for the prediction of sea wind power. SUMMARY

[0004] The present application aims to provide a sea wind power prediction management system to provide weather changes for sea wind power prediction.

[0005] To achieve the above-mentioned purpose, the embodiment of the present application provides a sea wind power prediction management system, which comprises a main server, a wind power prediction sub-server, a wind farm monitoring module, a large-scale data weather forecast module and an interactive terminal; the main server comprises a wind farm data analysis sub-server, a downscaling cluster sub-server and a relational database;

[0006] The main server is provided with an independent interface between each wind farm, and the wind farm data analysis sub-server and the relational database exchange data through the communication interface; the relational database communicates with the interactive terminal to perform real-time prediction and display of wind power;

[0007] The wind farm data analysis sub-server is in communication connection with the wind power prediction sub-server, and the wind farm data analysis sub-server is used to provide the wind farm data required by the wind power prediction sub-server for wind power prediction;

[0008] The downscaling cluster sub-server is in communication connection with the wind power prediction sub-server, and the downscaling cluster sub-server provides the environmental data required by the wind power prediction sub-server for wind power prediction;

[0009] The wind farm monitoring module is used to obtain unit data of the offshore wind turbine generator, and the large-scale data weather forecast module is used to obtain wind speed, temperature, wind direction, atmospheric stability and wake effect data.

[0010] Preferably, the wind power prediction sub-server comprises a plurality of wind farm CPUs corresponding to different wind farms, and the memory of the corresponding wind farm CPU is allocated according to the size of the wind farm.

[0011] Preferably, the offshore wind power prediction management system further comprises a meteorological mast module for measuring wind power data of the offshore wind farm and sending the wind power data to the main server.

[0012] Preferably, the offshore wind power prediction management system further comprises a configuration module for controlling the display of template drawing, telemetry, remote signaling and remote pulse on the interactive terminal.

[0013] Preferably, the interactive terminal adopts a B / S-based test device display interface.

[0014] Preferably, the offshore wind power prediction management system further comprises a security protection module, and the security protection module comprises a preset scanning port and a real-time scanning port, and is used for security management and real-time monitoring of the virtual machine of the offshore wind power prediction management system.

[0015] Preferably, the offshore wind power prediction management system further comprises a centralized control management module, and the centralized control management module is used for daily operation and maintenance management, wind farm increase and decrease management control, wind farm fault processing and wind farm configuration management of the offshore wind power prediction management system.

[0016] Preferably, the operation method of the wind power prediction sub-server comprises: obtaining wind power data of the offshore wind farm, and dividing the wind power data into atmospheric stability measurement parameters and turbulence intensity and wind shear measurement parameters to obtain classified wind power data; obtaining a historical wind power sequence, and dividing the historical wind power sequence into fluctuation sections to obtain a classified historical wind power sequence; training a network prediction model according to the classified wind power data and the classified historical wind power sequence, using the trained network prediction model to perform ultra-short-term power prediction to obtain preliminary prediction power of a to-be-predicted period; performing fluctuation type identification on the preliminary prediction power to obtain a corresponding error correction strategy and an error prediction result; and adding the preliminary prediction power and the error prediction result to obtain final power prediction.

[0017] Compared with the prior art, the present application has the following beneficial effects:

[0018] The offshore wind power prediction and management system provided by this invention includes a main server, a wind power prediction sub-server, a wind farm monitoring module, a large-scale data weather forecasting module, and an interactive terminal. The main server includes a wind farm data analysis sub-server, a downscaling cluster sub-server, and a relational database. The main server has independent interfaces with each wind farm. The wind farm data analysis sub-server and the relational database exchange data through a communication interface. The relational database communicates with the interactive terminal to perform real-time wind power prediction and display. The wind farm data analysis sub-server is communicatively connected to the wind power prediction sub-server and provides the wind farm data required for wind power prediction. The downscaling cluster sub-server is also communicatively connected to the wind power prediction sub-server and provides the environmental data required for wind power prediction. The wind farm monitoring module obtains unit data of offshore wind turbine generators, and the large-scale data weather forecasting module obtains data on wind speed, temperature, wind direction, atmospheric stability, and wake effect. This invention can provide weather change data for offshore wind power forecasting. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the structure of an offshore wind power prediction and management system provided in a certain embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram of the structure of a computer terminal device provided in a certain embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0024] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0025] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0026] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0027] This embodiment takes ESX / ESXi virtualization technology as an example. The resource virtualization hierarchy includes: on top of the hardware, ESX / ESXi (a type of server resource pooling) is first deployed to pool the various resources of the physical machine (CPU, memory, external storage, network, etc.). On this basis, it is possible to quickly generate multiple virtual machines with different operating systems and deploy different business application systems on each virtual machine to achieve the goal of resource aggregation.

[0028] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure of an offshore wind power prediction and management system provided in a certain embodiment of the present invention. In this embodiment, the offshore wind power prediction and management system includes a main server 10, a wind power prediction sub-server 50, a wind farm monitoring module 60, a large-scale data weather forecast module 70, and an interactive terminal 80; the main server 10 includes a wind farm data analysis sub-server 20, a downscaling cluster sub-server 30, and a relational database 40;

[0029] The main server 10 has an independent interface with each wind farm. The wind farm data analysis sub-server 20 and the relational database 40 exchange data through a communication interface. The relational database 40 communicates with the interactive terminal 80 to perform real-time prediction and display of wind power.

[0030] The wind farm data analysis sub-server 20 is communicatively connected to the wind power prediction sub-server 50, and the wind farm data analysis sub-server 20 is used to provide the wind farm data required for wind power prediction to the wind power prediction sub-server 50.

[0031] The downscaling cluster sub-server 30 is communicatively connected to the wind power prediction sub-server 50, and the downscaling cluster sub-server 30 provides the wind power prediction sub-server 50 with the environmental data required for wind power prediction.

[0032] The wind farm monitoring module 60 is used to obtain unit data of offshore wind turbine generators, and the large-scale data weather forecast module 70 is used to obtain data on wind speed, temperature, wind direction, atmospheric stability and wake effect.

[0033] In this embodiment, the master server can be understood as a CPU that integrates and manages the data of all other sub-servers, the relational database is a data storage module that can retrieve data information in real time and quickly, and the wind power prediction sub-server can be used to obtain information such as environmental data.

[0034] In one embodiment, server resource pooling technology is used to virtualize the server, storage, and network resources required by the offshore wind power prediction and management system. The main server 10 has independent interfaces with each wind farm, and it houses a wind farm data analysis sub-server 20, a downscaling cluster sub-server 30, a relational database 40, a wind power prediction sub-server 50, and an interactive terminal 80. The wind farm data analysis sub-server 20 and the relational database 40 exchange and analyze data through communication interfaces. The relational database 40 communicates with the interactive terminal 80 to perform real-time wind power prediction and display. The wind farm data analysis sub-server 20 communicates with the wind power prediction sub-server 50 and provides the wind farm data required for the final wind power prediction. The downscaling cluster sub-server 30 and the analysis sub-server communicate with the wind power prediction sub-server 50 and provide the environmental data required for the final wind power prediction. The wind power prediction sub-server 50 implements improved principal component analysis (PCA). The system employs a short-term offshore WPP algorithm based on PCA (Programmable Optimization Analysis). The data source subsystem includes monitoring modules 60 for each wind farm or dike meters and a large-scale data weather forecasting module 70. The monitoring modules 60 or dike meters are used to obtain data from multiple offshore wind turbine generators, while the large-scale data weather forecasting module 70 is used to obtain data on wind speed, temperature, wind direction, atmospheric stability, and wake effect.

[0035] In one embodiment, the wind power prediction sub-server 50 includes several wind farm CPUs corresponding to different wind farms, and memory is allocated to the corresponding wind farm CPUs according to the different wind farm scales.

[0036] In one embodiment, the offshore wind power prediction and management system further includes a meteorological mast module for measuring wind power data from the offshore wind farm and sending it to the main server 10.

[0037] In one embodiment, the offshore wind power prediction and management system further includes a configuration module for controlling the display of template drawing, telemetry, teleindication, and telepulse on the interactive terminal 80.

[0038] In one embodiment, the interactive terminal 80 adopts a B / S-based packaging and testing device display interface.

[0039] In one embodiment, the offshore wind power prediction and management system further includes a security protection module, which includes a preset scanning port and a real-time scanning port for security management and real-time monitoring of the virtual machine of the offshore wind power prediction and management system.

[0040] In one embodiment, the offshore wind power prediction and management system further includes a centralized control management module, which is used to perform daily operation and maintenance management, wind farm increase and decrease management and control, wind farm fault handling and wind farm configuration management on the offshore wind power prediction and management system.

[0041] In this embodiment of the invention, the offshore wind power prediction and management system also includes a distributed bus system. Built upon the latest middleware technology, a thorough understanding of existing real-time middleware implementations, and specialized and rational modifications adapted to the field of power automation, it is a general-purpose, stable, and high-performance network service management platform specifically applied to the field of new energy power automation. The network communication middleware provides general component-based network management functions, including distributed component management, access request proxy, and communication bus functions. Specific power applications are implemented as one or more service components for completing the offshore wind power prediction and management function. These service components can run on any host node in the system network, and the collaborative work of all service components constitutes a complete and reliable SCADA system.

[0042] During operation, when the system's software and hardware environment or user functional requirements change, users can easily adapt to the new requirements by refactoring components or modifying or upgrading a small number of components, thus ensuring the system's maintainability.

[0043] The real-time database system, built upon efficient memory management and indexing mechanisms, is an object-oriented, distributed, high-capacity, high-performance, and open real-time relational database management system. It ensures real-time response capabilities by employing technologies such as timers, memory pools, and shared memory, and guarantees fast query capabilities through an improved hash algorithm. It meets the real-time, consistency, predictability, and high-throughput requirements of SCADA systems and Guoneng Rixin's new energy business data.

[0044] The openness of the real-time database management subsystem of the offshore wind power forecasting integrated management platform is mainly reflected in two aspects: First, the real-time database data structure supports complete user customization, such as supporting CIM modeling as well as non-CIM modeling; Second, the real-time database management system provides data access clients with dynamic binding interfaces, static binding interfaces based on CIS specifications, and comprehensive support for SQL language and ODBC specifications.

[0045] The real-time database management subsystem is completely open to the application layer. Database schema, number of databases, database size, number of tables, table size, and table structure definitions all support user-configurable settings. It supports dynamic table creation, dynamic addition of fields, and dynamic addition / removal of records during runtime. It supports one-to-many and many-to-one relational data modeling and fast querying of relational data. To adapt to the needs of power system fault data management and dynamic phasor data management, the real-time database management system also supports the configuration and existence of two-dimensional data fields. The real-time database management system provides fast, structure-based static data binding and API queries, as well as flexible dynamic data post-binding and API queries. It provides standard SQL language queries based on a subset of ISO / IEC 9075:1999. It provides query / answer-based question-and-answer data query services and publication / subscription-based streaming data query services.

[0046] The front-end system includes: a) Front-end system structure: The front-end system is the interface for various real-time data to enter the offshore wind power massive data access and monitoring platform, mainly comprising a front-end acquisition system and a front-end forwarding system. The front-end system collects real-time operating information of the power equipment through communication with remote devices, provides the collected real-time data to application services, and implements remote control and adjustment functions of the power equipment according to the instructions issued by the application services. The front-end system adopts a layered design, consisting of an equipment layer, a framework layer, a protocol interpretation layer, and a data processing layer. The framework layer is the core of the front-end system, implementing channel monitoring, channel start / stop, channel redundancy, and channel management. In addition to channel redundancy, the framework layer also implements front-end node redundancy, ensuring data reliability and stability. The data processing layer mainly implements functions such as range conversion, coefficient conversion, validity checking, smoothing processing for telemetry measurements, and extreme value processing for telemetry signals. b) Support for multiple devices: The equipment layer mainly shields various communication media, providing a unified data read and write interface to the framework. The design separates the communication protocol and communication medium at the device layer, greatly facilitating the expansion and customization of new protocols. c) It supports multiple protocols; in the front-end system, communication protocols are the most diverse and variable. The front-end system primarily handles various different protocols at the protocol interpretation layer, greatly facilitating the expansion and customization of new protocols by abstracting the common interfaces and data access methods of different protocols.

[0047] The front-end system supports standard industrial Modbus protocols, power standard protocols 101, 103, and 104, as well as custom distributed cloud protocols. It also supports the development of custom protocols. Protocol development only needs to focus on protocol layer issues and has no impact on the communication framework or device layer.

[0048] The data access and monitoring system includes the following functions: a) Data processing: It can handle remote signaling, telemetry, remote pulse, and SOE data updates. It can handle linear and unit changes in telemetry quantities, as well as polarity changes in remote signaling quantities. Data supports quality codes and manual data input. b) Alarm functions: It supports various alarms, such as remote signaling change alarms, telemetry over-limit alarms, system start / exit events, and channel connectivity events. c) Control functions: It supports remote control and adjustment operations and interlock checks. d) Calculation functions: It supports user-defined formulas for mathematical operations on real-time data. The formula definitions support commonly used mathematical functions such as maximum, minimum, and average values.

[0049] A flexible and convenient configuration system that supports template drawing, curves, wind rose diagrams, and other functions. It supports the display of telemetry, telesignal, and telepulse measurements, and supports remote control and remote adjustment commands.

[0050] The functions of the historical storage and statistics system include: a) Historical storage: Supports periodic storage of telemetry, teleindication, and telepulse measurements; the default storage period is 5 minutes, which can be freely modified by the user, with a minimum storage period of 1 second; supports historical storage of alarms and events. b) Historical statistics: Supports automatic statistical functions by day, week, month, quarter, and year; supports statistics for user-defined time periods.

[0051] System performance metrics: Supports data collection refresh within 3 seconds and control execution within 3 seconds; supports up to 1 million data points (currently supports 500,000 points, with support for millions of points to be added later).

[0052] Offshore wind power forecasting methods often employ covariance combination forecasting. In the equal-weighted average combination forecasting method, various models have the same impact and contribution to the forecast results. However, in reality, the accuracy of different forecasting models varies, with some models showing significantly higher accuracy than others. A simple equal-weighted average combination method can reduce the role these models should play in the combined forecasting model. Therefore, non-equal-weighted combination forecasting methods often achieve better results. The covariance combination forecasting method is one such non-equal-weighted combination forecasting method.

[0053] In a preferred embodiment, the wind power prediction sub-server 50 includes multiple wind farms. Each wind farm CPU is assigned a memory allocation based on its size. For example, wind farm 1 has a power generation capacity of 50MW, so it is allocated 2GB of memory to 4 CPUs; wind farm 2 has a power generation capacity of 100MW, so it is allocated 6GB of memory to 8 CPUs; and wind farm N has a power generation capacity of 2625MW, so it is allocated 10.5GB of memory to 21 CPUs.

[0054] As a preferred embodiment, the interactive terminal 80 adopts a B / S-based packaging and testing system display interface and communicates in real time with multiple wind farms currently in operation.

[0055] As a preferred implementation, the integrated management platform for offshore wind power forecasting, built on a fog computing platform, can achieve unlimited scalability. When user data increases, there's no need to change the application architecture; simply adding more hardware devices is sufficient to support the application's growth. Regardless of the number of users, real-time application modifications can be made as easily as with a single user. End users are first distributed across different instances via a load balancing layer, distributing the load across multiple instances and thus enabling unlimited horizontal scaling of the application. On the fog computing platform, the integrated management platform for offshore wind power forecasting scales extremely rapidly, within minutes.

[0056] In a preferred embodiment, the offshore wind power forecasting integrated management platform also includes a unified security protection subsystem. This subsystem is implemented based on server resource pooling, rather than targeting a single virtual machine. This approach better addresses security issues between virtual machines and provides a unified solution to security problems posed by external network environments to all wind power forecasting system virtual machines running on the server resource pooling platform. The unified security protection subsystem includes: preset scanning ports, real-time scanning ports, communication packet filtering, clearing and repair, and status monitoring. It enables security management and real-time monitoring of each wind power forecasting system virtual machine based on a unified security solution using server resource pooling.

[0057] As a preferred implementation, it also includes a management sub-module for the centralized control terminal from engineering implementation to online service, which is used for the management and maintenance of the offshore wind power prediction integrated management platform on the fog computing platform, including: daily operation and maintenance management (ensuring the operation of wind measurement services for each wind farm 24 / 7); management and control of adding new wind farms (with the participation of engineering personnel); wind farm fault handling (with the participation of daily operation and maintenance personnel or R&D personnel); and wind farm holiday configuration management (responsible for daily operation and maintenance personnel).

[0058] Its advantages include: a reasonable division of roles and responsibilities among engineers, R&D personnel, and on-site wind farm maintenance personnel; a comprehensive management platform on top of the fog computing platform that enables 24 / 7 management and maintenance; daily maintenance personnel on the fog computing platform can receive technical support from R&D personnel at any time to solve more complex problems; and all historical data centralized on the fog computing platform can be provided to R&D personnel for in-depth data mining and continuous system optimization.

[0059] Offshore wind farms experience rapid atmospheric movements, affecting not only wind speed, wind direction, and wave height leading to varying sea surface roughness, but also the generation, propagation, and dissipation of wind turbine wakes. These processes influence short-term offshore wind power forecasts (from a few hours to a few days). Clearly, using physical methods to predict wind power results in low predictability as the wind power curve changes with wind conditions (wind speed, turbulence intensity, etc.), and requires significant computational resources. Statistical methods, on the other hand, utilize extensive historical data from wind farms to establish nonlinear relationships between meteorological variables such as wind speed and temperature and the wind farm's power generation per unit time. These methods predict future wind power output based on measured data, resulting in lower computational complexity. Commonly used statistical methods include support vector regression (SVR) models based on support vector machines, LightGBM (light gradient boosting machine), and Gaussian processes (GP). Recurrent neural networks (RNNs) can also be used to predict short-term wind speeds, offering higher accuracy than feedforward neural networks (FNNs). However, the training mechanism of a single neural network cannot accurately learn the behavior of wind signals, thus requiring optimization and improvement of the algorithm. Furthermore, a random search technique based on the clonal selection algorithm (CSA) can be implemented to train wavelet neural networks (WNNs) to improve prediction accuracy. Adding error feedback to an improved radial basis function neural network (RBFNN) can enhance the convergence and solution accuracy of wind speed or wind power prediction algorithms. However, these algorithms primarily consider meteorological factors such as wind speed, temperature, and pressure when predicting wind power, neglecting atmospheric stability (As) and wake disturbances within the wind farm. If we classify atmospheric stratification using turbulence intensity (Ti) and wind shear (SH) as measures of atmospheric stability, stable atmospheric stratification is characterized by a higher wind shear and lower turbulence intensity. The more stable the atmospheric stratification, the greater the wake effect loss of the wind farm. Furthermore, in the absence of wake effects, there can be a power output difference of up to 15% between stable and unstable atmospheric stratifications. Therefore, to simplify the problem, offshore wind power prediction must be conducted under a neutral atmospheric stratification (between stable and unstable).Under the same atmospheric stratification stability, given a wind speed range of 7-9 m / s, the output power of offshore wind farms can differ by up to 40% depending on the wind direction. This is attributed to the wake effect of different wind directions.

[0060] In one embodiment of the present invention, the operation method of the wind power prediction sub-server includes the following steps:

[0061] S110, acquire wind power data of offshore wind farms, and divide the wind power data into atmospheric stability measurement parameters and turbulence intensity and wind shear measurement parameters to obtain classified wind power data;

[0062] S120, Obtain the historical wind power sequence, and divide the historical wind power sequence into fluctuation segments to obtain the classified historical wind power sequence;

[0063] S130, The network prediction model is trained based on the classified wind power data and the classified historical wind power sequence. The trained network prediction model is used to perform ultra-short-term power prediction to obtain the preliminary predicted power for the period to be predicted.

[0064] S140, identify the fluctuation type of the preliminary predicted power to obtain the corresponding error correction strategy and error prediction results;

[0065] S150, the preliminary predicted power is added to the result of the error prediction to obtain the final power prediction.

[0066] When forecasting offshore wind power, the effects of atmospheric stability and wake effects are considered. While using features measuring atmospheric stability as input to the prediction model, a power-wind direction model is constructed based on wind direction and wake effect losses from wind farms. This model effectively quantifies wake effect losses from offshore wind farms in different wind directions and helps distinguish the impact of different atmospheric stratification stability on unidirectional wake effects. To fully utilize the input sequence information, an encoder-decoder framework with an attention mechanism is used in the algorithm model selection.

[0067] Atmospheric stability is measured using the Monin-Obukhov length, which reflects the relative magnitude of the work done by turbulent stress and the work done by buoyancy. The Monin-Obukhov length can be characterized by turbulence intensity and wind shear, both of which can be calculated from FINO1 meteorological mast measurements. T = A V90 / V 90 (1), S=V 90 / V 40 (2). Where T and S represent turbulence intensity and wind shear, respectively, and A V90 V represents the standard deviation of wind speed measurements at a height of 90m on the meteorological mast. 90and V 40 The wind speed values ​​were measured by a rotating cup anemometer at heights of 90m and 40m on the meteorological mast, respectively.

[0068] When a wind farm is considered as a whole, the impact of atmospheric stability on the wind farm is mainly reflected in wind speed fluctuations, which can be described by turbulence intensity and wind shear. To quantify wake effect losses and consider the influence of atmospheric temperature on the wake effect, it is generally necessary to construct a wake model for the wind farm. This embodiment employs two methods: the semi-empirical engineering model Jensen wake model, summarized from experiments; and models based on physical processes, such as computational fluid dynamics (CFD) models simulating wake aerodynamic characteristics, and the Fitch wake model, which is nested within numerical weather prediction (NWP) to calculate momentum and turbulent kinetic energy. Both wake models can effectively estimate wind speed losses caused by the wake effect, but both are limited to fixed wind directions.

[0069] Under the same wind speed, the output power of wind turbines changes continuously with wind direction and their location at the wind farm. Calculating the interaction between wind turbines based on wake characteristics requires a huge workload to construct wake models for wind farms by wind direction, and this process becomes increasingly complex with the increasing number of offshore wind turbines. Therefore, this embodiment ignores the spatial variable relationships between wind turbines and uses the power loss in each wind direction to reflect the wake effect. First, to verify the existence of a mapping relationship between wind direction and power loss in the wind farm, a dataset with wind speeds of 7-11 m / s is used to present the power loss in relation to wind direction. The power loss is evaluated using the power loss ratio, i.e.: η p =[1-P wake / P free (3), where η p Indicates the percentage of power loss; P wake P represents the output power of the wind turbine affected by the wake effect. free The power output of the wind turbine is calculated based on the wind direction when it operates in free wind (without wake interference). The average power output is then calculated by selecting the wind turbine operating in free wind direction based on the wind direction.

[0070] When forecasting offshore wind power, an ultra-short-term offshore WPP algorithm based on improved principal component analysis is implemented, considering the influence of atmospheric stability and wake effects. While using features measuring atmospheric stability as input to the prediction model, a power-wind direction model is constructed based on wind direction and wake effect losses from wind farms. This power-wind direction model is then combined with the ultra-short-term offshore WPP algorithm based on improved principal component analysis. This model can effectively quantify wake effect losses from offshore wind farms in different wind directions and helps distinguish the influence of different atmospheric stratification stability on the wake effect in the same direction. To fully utilize the input sequence information, an encoder-decoder framework with an attention mechanism is used in the algorithm model selection. The Monin-Obukhov length is used as the parameter for measuring atmospheric stability, reflecting the relative magnitude of turbulent stress work and buoyancy work. The Monin-Obukhov length characterizes turbulence intensity and wind shear, both of which can be calculated from FINO1 meteorological mast measurement data. T = A V90 / V 90 (1), S=V 90 / V 40 (2); where T and S represent turbulence intensity and wind shear, respectively, and A V90 V represents the standard deviation of wind speed measurements at a height of 90m on the meteorological mast. 90 and V 40 The wind speed values ​​were measured by a rotating cup anemometer at heights of 90m and 40m on the meteorological mast, respectively.

[0071] In a specific embodiment, the improved principal component analysis algorithm for ultra-short-term offshore WPP, which incorporates power and wind direction models, includes:

[0072] Step 1: Measure the wind power data of the offshore wind farm using the FINO1 meteorological mast and delete any missing or erroneous data. Divide the offshore wind farm data into parameters for atmospheric stability and parameters for turbulence intensity and wind shear. The atmospheric stability parameter uses the Monin-Obukhov length, which reflects the relative magnitude of the work done by turbulence stress and buoyancy. The turbulence intensity and wind shear parameters are characterized using the Monin-Obukhov length.

[0073] Step 2: Divide the historical power sequence into fluctuation segments and construct a neural network to classify the fluctuation process;

[0074] Step 3: Randomly initialize the weights of the improved principal component analysis (PCA), set the maximum number of iterations M = 50, and the current number of iterations m = 1;

[0075] Step 4: Construct multiple convolutional channels, input feature variables at different levels into the corresponding convolutional channels for feature extraction, and input the extracted temporal features into the LSTM layer according to time steps to calculate the output results of the improved principal component analysis;

[0076] Step 5: Calculate the network prediction error and use the Lookahead optimizer to back-optimize and improve the principal component analysis weights;

[0077] Step 6: If the maximum number of iterations (m>M) is reached, the iteration terminates and the parameters of the improved principal component analysis network are output; otherwise, let m = m+1 and go to step 4.

[0078] Step 7: Use the improved principal component analysis after training to perform ultra-short-term power prediction and obtain the preliminary predicted power for the period to be predicted.

[0079] Step 8: Identify the fluctuation type of the preliminary predicted power and adopt corresponding error correction strategies for different types of fluctuation processes;

[0080] Step 9: Add the preliminary predicted power to the error prediction result to complete the final power prediction.

[0081] The process of establishing the time series prediction mathematical model at the ultra-short-term scale used in the improved principal component analysis of the aforementioned power-wind-direction model for ultra-short-term offshore WPP algorithm is as follows:

[0082] 1. Time series exhibits certain dynamic time characteristics, meaning that the current sequence value is correlated with the sequence values ​​from several previous times, and this correlation increases as the time interval decreases. The ultra-short-term offshore wind power series, as a typical time series, can be predicted using the following model:

[0083] P(t)=f1(P(ta),P(t-2a),…)+E(t) (20), where a is the time interval for data acquisition, f1 is the time correlation function of the offshore wind power sequence, and E(t) is the prediction error at time t.

[0084] Due to the complexity of weather systems, offshore wind power sequences exhibit a certain degree of non-stationarity. Differential smoothing of the power sequence can reduce the f1 complexity and decrease prediction errors.

[0085] ΔP(t)=f2(ΔP(ta),ΔP(t-2a),…)+e(t) (21)

[0086] In the formula, ΔP(t) is the change in offshore wind power between time t and time ta, f2 is the time correlation function of the offshore wind power differential sequence, and e(t) is the minimum prediction error at time t.

[0087] 2. Preliminary prediction of offshore wind power considering multi-level characteristics:

[0088] Offshore wind power is essentially a manifestation of atmospheric kinetic energy, and its output is primarily affected by fluctuations in meteorological factors. When meteorological factors change rapidly, wind power output fluctuates frequently; when meteorological factors are stable, the fluctuations are smaller. Therefore, forecasting models should consider the fluctuations in meteorological factors. However, similar meteorological fluctuations can still lead to different changes in offshore wind power output. For example, the instantaneous output of a wind turbine is proportional to the cube of the wind speed. Under the same wind speed change, higher wind speeds obviously lead to more significant changes in wind power output than lower wind speeds. Furthermore, by incorporating historical power series, therefore:

[0089] ΔP'(t)=f(P,W,ΔP,V) (22)

[0090] In the formula, ΔP'(t) is the predicted value of the change in offshore wind power at time t, P and W are the power series and meteorological variables with several time steps, respectively, collectively referred to as power and meteorological variables; ΔP and V are the power difference series and meteorological fluctuation variables with several time steps, respectively, collectively referred to as fluctuation variables. After the inverse difference transformation, the predicted power at time t can be obtained.

[0091] 3. Prediction error correction model based on fluctuation process classification:

[0092] Power prediction is performed using improved principal component analysis, and the fluctuation processes are classified according to methods 1-2. During the statistical analysis of the prediction results, it was found that different fluctuation processes have different prediction errors. Therefore, this invention trains different error correction models using corresponding error samples for different fluctuation processes to enhance the adaptability of the error prediction model to specific fluctuation processes. Meanwhile, considering that the prediction errors for low-output and constant-output fluctuations are relatively small, no error correction is applied to them. The final power prediction result is expressed as follows:

[0093] P(t) 终 =P(ta)+f(P,W,ΔP,V)+g(X) (23)

[0094] In the formula, P(t) 终 Let t represent the predicted offshore wind power, g represent the error correction function after classification and matching, and X represent the feature variables that are strongly correlated with the initial prediction error.

[0095] 4. Improve principal component analysis:

[0096] Unlike traditional LSTM networks, Principal Component Analysis (PCA) is a recurrent neural network containing principal component variables and exhibiting convolutional properties. It autonomously extracts principal component variables—data features at each time step in the input event sequence—using convolutional structures to generate simple and effective deep features, improving the prediction accuracy of the recursive sequence learning module. It is suitable for unrestricted time series prediction problems affected by complex high-dimensional factors. Furthermore, it is not a simple concatenation of CNN and PCA, but rather extracts data features without disrupting the temporal structure of the sequence, ensuring the temporal structure of the extracted features. Additionally, it proposes the following three improvements to the results of PCA and convolution:

[0097] (1) For multi-convolutional channel structures with different levels of features: extract the temporal features and principal components of variables at different levels to ensure the flexibility and effectiveness of feature extraction. Concatenate the power and meteorological variables and fluctuation variables at a certain moment to form a three-dimensional tensor, which is the number of features, time step and number of channels, respectively, and input the corresponding convolutional channels;

[0098] (2) The Lookahead optimizer, a forward-looking improvement on the Adam optimizer, is used to achieve fast modeling and stable convergence. Iterative updates of two sets of weights are performed, and the search direction is selected by observing the fast weight sequence generated by another optimizer in advance. The Lookahead optimizer optimizes the network through the following steps:

[0099] Fast weights are generated by updating the model weights k times using the inner loop optimizer and storing the k sequence weights. The inner loop optimizer is the Adam optimizer, and the optimizer function is updated based on the number of times the optimizer is trained and the inner loop weights.

[0100] Slow weights are calculated using an exponentially weighted average algorithm based on the k sequence weights saved by the inner loop optimizer in each round.

[0101] Set the initial weights of the inner loop optimizer during the (j+1)th training iteration;

[0102] If the maximum number of training iterations is reached or the error meets a specific value, the training process is terminated, and the initial weights of the inner loop optimizer in the (j+1)th training iteration are used as the model weights; otherwise, j = j+1 is set, and the process switches to fast weights.

[0103] (3) Error correction considering feature selection: When wind power output fluctuates significantly, it may lead to a decrease in model prediction accuracy. Therefore, in order to balance model complexity and correction effect, a two-layer radial basis function neural network is used as the error correction model. The input factors of the error correction model will directly determine the correction accuracy. If all meteorological features are considered, the dimensionality of the input data will be too high, which may lead to low error correction efficiency. If the input factors are selected based on experience, the generalization ability of the model may decrease. In order to select appropriate input factors, the Xgboost algorithm is used to perform correlation analysis on predicted power, predicted power fluctuation, meteorological variables, meteorological fluctuation variables and predicted power error. The core idea of ​​the Xgboost algorithm is to use the characteristics of the tree model to quantify the correlation of each feature. The sum of the correlations of all features is 1. In order to ensure model efficiency and correction accuracy, the top 4 features with higher correlation coefficients are selected as the input of the error correction model.

[0104] In this embodiment, the power loss ratio with respect to wind direction is composed of eight approximately Gaussian distributed functions, with each peak approximately 45 degrees apart. In the direction of wind origin, the more upstream wind turbines there are, the greater the reduction in downstream wind speed. The power loss of the wind farm is consistent with the wake effect loss in each wind direction, and the power loss varies considerably within each approximately Gaussian distributed wind direction interval. In addition to wind direction and wind farm layout, two other factors affect the function distribution: firstly, data is scarce in certain wind directions or speed ranges, causing the mean of the power loss ratio to be discrete; secondly, the stability of the atmospheric stratification varies, resulting in different wake effect losses. Therefore, depending on wind direction and wind farm layout, the wake effect loss of the wind farm has a certain functional relationship with the wind direction; this embodiment uses the Pd model. Let Pi represent the power loss of the i-th (i = 1, 2, ..., n) data sample of the wind farm, then: Pi = W(vi, di; B) + C, (4); where: W(·) represents the wake effect loss of the wind farm; vi and di are the wind speed and wind direction respectively, B is the regression parameter; white noise C is the white noise with a mean of 0 and a variance of σ 2 If the wind farm's power loss Pi is a Gaussian random variable, then the power loss Pi is mainly caused by the wake effect, and is also affected by atmospheric stability. Unlike the theoretical wind power curve of a single turbine, the power loss (output) of a wind farm mainly depends on two variables: vi and di. In this process, there is no need to construct a wake model; the Pd parameter, derived from the wind speed and power loss in each wind direction, can characterize the wake effect loss corresponding to each wind speed under different wind directions. By inputting the Pd parameter along with turbulence intensity and wind shear into the wind power prediction model, the influence of atmospheric stability on the wake effect can be taken into account, thereby correcting the prediction error caused by it.

[0105] Preferably, the offshore wind power prediction model includes a gated recurrent unit (GRU): To address the long duration and gradient minimization issues encountered by traditional recurrent neural networks (RNNs) in sequence prediction, a Long Short-Term Memory (LSTM) network combined with a gated recurrent unit (GRU) can be used. Compared to LSTM, GRU has a simpler structure, significantly improves training efficiency, and has lower computational overhead. Therefore, considering the hardware's computational power and time cost, GRU is used in both the encoding and decoding stages. While GRU has one less gate than LSTM, it still implements the "selection" and "forgetting" mechanisms found in LSTM.

[0106] The following are the expressions for the gating and input / output of the GRU:

[0107] Rt=r(Wr*[h t-1 x t (5)

[0108] Zt=r(Wz)*[h t-1 x t (6)

[0109] Ht=tanh(WH*[Rt*h t-1 x t (7)

[0110] ht=(1-Zt)*h t-1 +Zt*Ht] (8)

[0111] Where Rt and Zt are the reset gate and update gate, respectively, and Ht, Xt, and ht represent the candidate hidden state at the current time step, the input at the current time step, and the hidden state passed to the next time step, respectively; similarly, h t-1 Represents the hidden state at the previous time step; r is the sigmoid activation function; tanh represents the tanh activation function; [*] indicates barren features; Wr, Wz, and WH are weight parameters that the model needs to learn.

[0112] The Encoder-Decoder framework, commonly used in neural machine translation (NMT) and question-answering systems, transforms the input sequence into a fixed-length vector during the encoding phase. The decoder then predicts the current output based on this vector and the output sequence from previous time steps. Therefore, the traditional Encoder-Decoder framework cannot fully utilize the features of the input sequence when solving sequence-to-sequence problems. To avoid information loss due to compression, an attention mechanism is employed. Normalization weights are assigned to the transformed vector sequence based on their impact on the output, thus effectively utilizing the information of the input sequence. Attention mechanisms are incorporated into both the encoding and decoding phases. In the encoding phase, the model learns the weight distribution of input features such as wind speed, temperature, turbulence intensity, and wind shear to achieve feature extraction and selection; the same principle applies in the decoding phase.

[0113] (1) Encoding: During the encoding stage, the GRU learns various meteorological variables, atmospheric stability characteristics (i.e., turbulence intensity and wind shear), and the Pd parameter, outputting the hidden state of the current node as one of the inputs to the decoder, and simultaneously passing it to the next node. Let Xi represent the input of the i-th (i = 1, 2, ..., n) node, Xi = (xi1, xi2, ..., xim). Then the hidden state of the GRU is:

[0114] h i =f(h i-1 ,Xi) (9)

[0115] h i-1 and h i Let h be the hidden states of the (i-1)th and ith nodes, respectively, and f(·) represent GRU. The attention mechanism enables the model to autonomously select the most important input features through learning. First, h... i-1 Calculate the attention score for each feature in Xi:

[0116] e ik =a(h i-1 x ik (10)

[0117] In the formula, x ik e represents the k-th feature input to the i-th node, such as wind speed, temperature, etc. ik It is x ik The attention score is represented by a(·), where a(·) represents the attention mechanism. In this example, the model parameters are the same as the GRU weight matrix, denoted by We in the encoding stage and Wd in the decoding stage, as shown below:

[0118] e ik =We(h i-1 x ik (11)

[0119] Secondly, the attention probability distribution, i.e., the weights of each feature, is calculated using the softmax function, as shown in the following formula:

[0120] γ ik =exp(e ik ) / Σexp(e ik ),k=1,2…m (12)

[0121] In the formula, γ ik Let γ be the weight of the k-th feature in the input of the i-th node. ik Satisfying Σγ ik If k = 1, k = 1, 2, ..., m, then the input of the i-th node can be represented as:

[0122] Xi=(γ i1 xi1,γ i2 xi2,…γ im xim) (13)

[0123] Therefore, after the input sequence is processed by the attention mechanism, the weights of each feature are not equal. For example, wind speed is the main feature for constructing the wind power curve, so its weight is naturally greater than that of temperature, thus effectively extracting information from the input sequence.

[0124] (2) Decoding: In the decoding stage, if a traditional Encoder-decoder framework is used, all wind power-related variables after encoding will be compressed into a fixed-length vector, resulting in low information utilization. The attention mechanism obtains ci by summing the n hidden states of the encoded output:

[0125] Ci=Σβ ij h j (14)

[0126] In the formula, β ij These are the weights of each hidden state, and their calculation method is the same as γ. ik The same applies, starting with the hidden state S i-1 Simultaneously with the encoder, output the hidden state h j Attention score is obtained:

[0127] d ij =a(S i-1 ,h j (15)

[0128] Similarly, the attention mechanism parameters in the decoding phase are the same as the GRU weight matrix, i.e.:

[0129] d ij =Wd(S i-1 ,h j (16)

[0130] Then β is calculated using the softmax function. ij :

[0131] β ij =exp(d ij ) / Σexp(d ij ),j=1,2,…n (17)

[0132] The hidden state S of the i-th node in the decoding phase i Calculated by the following formula:

[0133] S i =f(S) i-1 Y i-1 ,ci) (18)

[0134] In the formula Y i-1 This represents the wind power output of the previous node. Therefore, to predict the current wind power Y... i It is not only based on the input Y i-1 Also related to S at the current moment i Related to ci:

[0135] Y i =N(S) i Y i-1 ,ci) (19)

[0136] In the formula, N(·) is a 3-layer fully connected neural network.

[0137] In one embodiment, a complete operation and maintenance management mechanism is established at the fog terminal, including detailed personnel division, detailed classification of operation and maintenance management activities, and workflow definitions for each small-scale sub-activity within each category. Regarding personnel, this includes interface-level operation and maintenance personnel, development personnel, and operation and maintenance management managers. During the workflow, the entire fault handling process has a well-defined mechanism for classifying and handling the nature of the problem. When the fault handling level is low, such as configuration or hardware issues, it can be resolved by operation and maintenance personnel; when the complexity of the problem exceeds the capabilities of operation and maintenance personnel, such as code-level faults or performance issues, it is handled by development personnel. Simultaneously, process tracking is completed, accepted, and archived by a professional manager, thereby achieving rapid response to problem discovery, reasonable division of labor for problem resolution, professional and efficient problem-solving processes, quality control of task completion, and event control, reducing workload and lowering labor costs.

[0138] In one embodiment, the system uses Pd parameters obtained from the Pd model to predict offshore wind power, which can effectively quantify the wake effect loss of wind farms with different wind directions and speeds, thereby improving prediction accuracy. Considering atmospheric stability can not only correct the prediction power error caused by using the average wind speed, but also correct the wake effect loss under different atmospheric layer stability by combining the Pd model, providing a reliable basis for the grid connection and dispatch of offshore wind power, and realizing accurate prediction of offshore wind power.

[0139] Please see Figure 2 This invention provides a computer terminal device, including one or more processors and a memory. The memory is coupled to the processors and is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the operation method of the wind power prediction sub-server as described in any of the above embodiments.

[0140] The processor controls the overall operation of the computer terminal device to complete all or part of the steps of the aforementioned wind power forecasting sub-server operation method. The memory stores various types of data to support the operation of the computer terminal device. This data may include, for example, instructions for any application or method operating on the computer terminal device, as well as application-related data. The memory can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0141] In an exemplary embodiment, the computer terminal device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described wind power prediction sub-server operation method and achieve the same technical effect as the described method.

[0142] In another exemplary embodiment, a computer-readable storage medium including a computer program is also provided. When executed by a processor, the computer program implements the steps of the operation method of the wind power prediction sub-server in any of the above embodiments. For example, the computer-readable storage medium may be the memory including the computer program described above. The computer program may be executed by the processor of a computer terminal device to complete the operation method of the wind power prediction sub-server described above and achieve the same technical effects as the method described above.

[0143] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A marine wind power prediction and management system, characterized in that, It includes a main server, a wind power prediction sub-server, a wind farm monitoring module, a large-scale data weather forecasting module, and an interactive terminal; the main server includes a wind farm data analysis sub-server, a downscaling cluster sub-server, and a relational database. The main server has an independent interface with each wind farm, and the wind farm data analysis sub-server and the relational database exchange data through a communication interface; the relational database communicates with the interactive terminal to perform real-time prediction and display of wind power. The wind farm data analysis sub-server is communicatively connected to the wind power prediction sub-server, and the wind farm data analysis sub-server is used to provide the wind farm data required for wind power prediction to the wind power prediction sub-server. The downscaling cluster sub-server is communicatively connected to the wind power prediction sub-server, and the downscaling cluster sub-server provides the wind power prediction sub-server with the environmental data required for wind power prediction. The wind farm monitoring module is used to obtain unit data of offshore wind turbine generators, and the large-scale data weather forecasting module is used to obtain data on wind speed, temperature, wind direction, atmospheric stability and wake effect. The operation method of the wind power prediction sub-server includes: acquiring wind power data from offshore wind farms and dividing the wind power data into parameters for measuring atmospheric stability and parameters for measuring turbulence intensity and wind shear; acquiring historical wind power sequences and dividing the historical wind power sequences into fluctuation segments to obtain fluctuation process classifications; using a trained network prediction model to perform ultra-short-term power prediction to obtain preliminary predicted power for the period to be predicted; identifying the fluctuation type of the preliminary predicted power to obtain the corresponding error correction strategy and error prediction results; and adding the preliminary predicted power with the error prediction results to obtain the final power prediction.

2. The offshore wind power prediction and management system according to claim 1, characterized in that, The wind power prediction sub-server includes several wind farm CPUs corresponding to different wind farms, and memory is allocated to the corresponding wind farm CPUs according to the different wind farm scales.

3. The offshore wind power prediction and management system according to claim 1, characterized in that, It also includes a meteorological mast module, used to measure wind power data from offshore wind farms and send it to the main server.

4. The offshore wind power prediction and management system according to claim 1, characterized in that, It also includes a configuration module for controlling the display of template drawing, telemetry, teleindication, and telepulse on the interactive terminal.

5. The offshore wind power prediction and management system according to claim 1, characterized in that, The interactive terminal uses a B / S-based packaging and testing device display interface.

6. The offshore wind power prediction and management system according to claim 1, characterized in that, It also includes a security protection module, which includes a preset scanning port and a real-time scanning port, for security management and real-time monitoring of the virtual machine of the offshore wind power prediction and management system.

7. The offshore wind power prediction and management system according to claim 1, characterized in that, It also includes a centralized control management module, which is used for daily operation and maintenance management, wind farm increase and decrease management and control, wind farm fault handling and wind farm configuration management of the offshore wind power prediction and management system.

Images