CN-121998451-A - Wind power plant wind speed prediction method, device, equipment and medium based on deep learning

CN121998451ACN 121998451 ACN121998451 ACN 121998451ACN-121998451-A

Abstract

The application discloses a wind power plant wind speed prediction method, device, equipment and medium based on deep learning, and relates to the technical field of intelligent operation and maintenance of wind power plants. The method comprises the steps of preprocessing data, constructing a three-dimensional input tensor, inputting a wind speed prediction model to perform wind speed prediction to obtain a wind speed prediction result of a prediction period, encoding a model of the wind speed prediction model to obtain an initial node state matrix based on the three-dimensional input tensor, constructing dynamic physical construction modules to obtain a dynamic physical information graph based on the three-dimensional input tensor and static data of a fan, performing iterative calculation by an iterative graph information transfer module to obtain the node state matrix at each moment in the prediction period, and decoding by a decoding module to obtain the wind speed prediction result. The high-precision wind turbine stage ultra-short-term wind speed prediction provided by the embodiment of the application can provide key technical support for high-order intelligent cooperative control of a wind power plant.

Inventors

XIA YAPING
REN YUFENG
Shi Gengfu
TANG XUEZHEN
CHANG YUFANG
QUAN RUI
HUANG WENCONG
WAN HANG

Assignees

湖北工业大学

Dates

Publication Date: 20260508
Application Date: 20260104

Claims (10)

1. The wind power plant wind speed prediction method based on deep learning is characterized by comprising the following steps of: acquiring meteorological data and fan operation data in a preset time window before the current moment, and acquiring fan static data of each fan; Preprocessing the fan operation data and the meteorological data, and constructing a three-dimensional input tensor based on the preprocessed data; inputting the three-dimensional input tensor and the fan static data into a trained wind speed prediction model to perform wind speed prediction, so as to obtain a wind speed prediction result of a prediction period; the wind speed prediction model comprises an encoding module, a dynamic physical construction module, an iteration diagram information transmission module and a decoding module; the encoding module is used for obtaining an initial node state matrix at the current moment based on the three-dimensional input tensor encoding; The dynamic physical construction module is used for constructing dynamic physical information graphs at all moments based on the three-dimensional input tensor and the static data of the fan; The iteration map information transfer module is used for carrying out iterative computation based on the initial node state matrix and the dynamic physical information map to obtain a node state matrix at each moment in the prediction period; The decoding module is used for decoding based on the node state matrix at each moment to obtain a wind speed prediction result of the prediction period.
2. The deep learning-based wind farm wind speed prediction method of claim 1, wherein the preprocessing of the wind turbine operating data and the meteorological data comprises: uniformly aligning the fan operation data with the time axis of the meteorological data to obtain an aligned data sequence, and resampling the aligned data sequence to a preset time resolution; detecting and removing abnormal values based on the resampled data sequence, and filling the missing values through interpolation calculation to obtain a data sequence after data cleaning; Converting each variable in the data sequence after data cleaning into an input format preset by the wind speed prediction model to obtain preprocessed time sequence data; the constructing based on the preprocessed data to obtain the three-dimensional input tensor comprises the following steps: Extracting node characteristics related to a single fan based on the preprocessed data, and obtaining global characteristics related to the whole wind power plant; And performing sequence splicing on the node characteristics of each fan and the global characteristics to construct and obtain the three-dimensional input tensor.
3. A method of wind farm wind speed prediction based on deep learning according to claim 2, wherein the training process of the wind speed prediction model comprises: Acquiring historical data of a historical period, wherein the historical data comprises historical meteorological data and historical fan operation data, and executing the preprocessing step based on the historical data to obtain preprocessed time series data; Intercepting a data sequence of a first period from the preprocessed time series data as sample input, intercepting a data sequence of a second period from the preprocessed time series data, extracting to obtain a wind speed sequence of the second period, and taking the wind speed sequence as a sample label; executing the construction on the sample input to obtain a three-dimensional input tensor, and inputting the corresponding three-dimensional input tensor into the wind speed prediction model to obtain a wind speed prediction result of a second period output by the wind speed prediction model; constructing a loss function based on the wind speed prediction result and the corresponding sample label, and transmitting based on the numerical value of the loss function to adjust the parameters of the wind speed prediction model until the model converges to obtain a trained wind speed prediction model.
4. A method of predicting wind speed of a wind farm based on deep learning according to any of claims 1-3, wherein the obtaining an initial node state matrix at a current time based on the three-dimensional input tensor code comprises: Decomposing the three-dimensional input tensor into a node characteristic sequence and a global characteristic sequence; Processing the global feature sequence through time pooling operation to obtain a global physical state vector; Inputting the global physical state vector into a gating network to generate a gating vector; And performing gating modulation operation on the node characteristic sequence based on the gating vector, processing the node characteristic sequence subjected to the gating modulation operation through a transducer encoder so as to capture a complex time sequence dependency relationship, extracting an output sequence of the last time step, and constructing to obtain the initial node state matrix.
5. The deep learning-based wind farm wind speed prediction method according to claim 1, wherein the dynamic physical information graph comprises an adjacency matrix and an edge feature matrix, and the process of constructing the adjacency matrix comprises: analyzing based on the three-dimensional input tensor to obtain real-time parameters at corresponding moments, wherein the real-time parameters comprise global wind direction, global turbulence intensity, inflow wind speed of each fan and yaw angle; Determining the position coordinates of each fan and the relative position relation between every two fans based on the fan static data, and determining the upstream fan corresponding to each downstream fan based on the relative position relation and the global wind direction at the corresponding moment; Calculating inflow wind speed loss caused by the influence of each upstream fan on each downstream fan, and forming a loss matrix based on the inflow wind speed loss of a fan pair formed by each downstream fan and a corresponding upstream fan; normalizing the numerical value in the defect matrix to obtain the adjacency weight of the influence degree of each upstream fan on the downstream fan, and obtaining the adjacency weight matrix; the process of constructing the edge feature matrix includes: And determining the corresponding directed edges of each fan pair, extracting the multidimensional physical characteristics of each directed edge, constructing to obtain edge characteristic vectors, and constructing to obtain the edge characteristic matrix based on the edge characteristic vectors of each directed edge.
6. The method for predicting wind speed of a wind farm based on deep learning according to claim 5, wherein the constructing obtains dynamic physical information diagrams at each moment, and the method comprises: And deducing based on the meteorological data to obtain a forecast wind condition parameter of each moment in a forecast period, and respectively updating the adjacent matrix and the edge feature matrix of the corresponding moment based on the forecast wind condition parameter of each moment to obtain the dynamic physical information graph of each moment.
7. The method for predicting wind speed of a wind farm based on deep learning according to claim 5 or 6, wherein the iterative calculation is performed based on the initial node state matrix and the dynamic physical information map to obtain a node state matrix at each moment in the prediction period, and the method comprises the following steps: extracting a node embedded vector of each fan based on the initial node state matrix, and calculating the scaling dot product attention of each fan pair based on the node embedded vector of each fan; Extracting adjacent weight and edge feature vector of each fan pair in the dynamic physical information graph at corresponding time; Calculating the attention score of each fan pair based on the attention of the scaling dot product, the adjacency weight and the edge feature vector; according to all fan pairs corresponding to the downstream fans, carrying out weighted fusion on the adjacent weight, the edge feature vector and the node embedded vector of the upstream fan of each fan pair based on the value of the attention score, and updating based on the weighted fusion result to obtain the node embedded vector of the downstream fan at the next moment; Obtaining a node state matrix at a corresponding moment based on the node embedded vector of each fan at the next moment; And transferring the updated node state matrix to the step of extracting the node embedded vector of each fan until the node state matrix of each moment in the prediction period is updated.
8. Wind power plant wind speed prediction device based on deep learning, characterized by comprising: The data acquisition module is used for acquiring meteorological data and fan operation data in a preset time window before the current moment and acquiring fan static data of each fan; the data processing module is used for preprocessing the fan operation data and the meteorological data and constructing a three-dimensional input tensor based on the preprocessed data; The prediction module is used for inputting the three-dimensional input tensor and the fan static data into a trained wind speed prediction model to perform wind speed prediction, so as to obtain a wind speed prediction result of a prediction period; the wind speed prediction model comprises an encoding module, a dynamic physical construction module, an iteration diagram information transmission module and a decoding module; the encoding module is used for obtaining an initial node state matrix at the current moment based on the three-dimensional input tensor encoding; The dynamic physical construction module is used for constructing dynamic physical information graphs at all moments based on the three-dimensional input tensor and the static data of the fan; The iteration map information transfer module is used for carrying out iterative computation based on the initial node state matrix and the dynamic physical information map to obtain a node state matrix at each moment in the prediction period; The decoding module is used for decoding based on the node state matrix at each moment to obtain a wind speed prediction result of the prediction period.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any one of claims 1 to 7 when the program is executed by the processor.
10. A non-transitory computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Description

Wind power plant wind speed prediction method, device, equipment and medium based on deep learning Technical Field The application relates to the technical field of intelligent operation and maintenance of wind power plants, in particular to a wind power plant wind speed prediction method, device, equipment and medium based on deep learning. Background Wind power generation is a key renewable energy technology and is increasingly important in the global energy system. However, the natural intermittence and volatility of wind energy presents challenges for stable operation of the power system. In order to realize stable control and optimal scheduling of wind power, a high-precision ultra-short-term wind speed prediction technology is important. Wind speed prediction methods in the related art can be broadly divided into three types including a physical model, a statistical model, and an artificial intelligence model. The physical model, for example, a method based on numerical weather prediction (Numerical Weather Prediction, NWP), predicts weather changes in a large scale range by solving an atmospheric physical equation set, and the method has longer prediction capability, but needs to design a complex data fusion and correction model to process data errors from different NWP sources, and is difficult to meet the requirements of high-precision and minute ultra-short-term prediction in a wind farm (hundred meters) only based on original NWP data. Statistical models, such as autoregressive integral moving average (ARIMA) models, while simple and easy to implement, generally require the assumption that the time series data satisfies linear or stationary conditions, and it is difficult to effectively capture strong nonlinear, non-stationary characteristics in the wind speed sequence. Furthermore, artificial intelligence models can use Graph Neural Networks (GNNs) to capture the spatial dependencies between fans, but existing artificial intelligence models still have the following problems: The static diagram introduces the concept of spatial proximity, but the core physical mechanism of interaction between the static diagram and the dominant fans, namely the wake effect, is dynamically coupled, the propagation path and the influence range of the wake are changed severely along with the real-time wind direction, and the static diagram cannot capture the physical causal relationship which is driven by the wind direction and has directionality and time variability. And secondly, the explicit modeling of the physical rule is missing. The artificial intelligent model in the related art relies on the model to self-inductive learn the correlation among fans from mass data, but the model structure itself is not explicitly embedded with aerodynamic prior knowledge about how wake flows are formed, deflected and attenuated, which not only requires huge data volume, but also the model learning may be only statistical correlation, not real physical causality. When a wind farm encounters new conditions that are not covered in training data, the prediction reliability and generalization capability of the model will be challenged due to the lack of strong constraints of the physical laws. In summary, the related art has a bottleneck in the accuracy of the ultra-short-term collaborative prediction of multiple units of the wind farm due to either insufficient spatial-temporal resolution, or over-simplified model assumptions, or failure to dynamically and explicitly integrate the spatial physical mechanism into the model architecture. Disclosure of Invention The application provides a wind power plant wind speed prediction method, device, equipment and medium based on deep learning, which are used for solving the defects of the related technology, and the technical scheme is as follows: in a first aspect, the present application provides a wind farm wind speed prediction method based on deep learning, including: acquiring meteorological data and fan operation data in a preset time window before the current moment, and acquiring fan static data of each fan; Preprocessing the fan operation data and the meteorological data, and constructing a three-dimensional input tensor based on the preprocessed data; inputting the three-dimensional input tensor and the fan static data into a trained wind speed prediction model to perform wind speed prediction, so as to obtain a wind speed prediction result of a prediction period; the wind speed prediction model comprises an encoding module, a dynamic physical construction module, an iteration diagram information transmission module and a decoding module; the encoding module is used for obtaining an initial node state matrix at the current moment based on the three-dimensional input tensor encoding; The dynamic physical construction module is used for constructing dynamic physical information graphs at all moments based on the three-dimensional input tensor and the static data of the fan; The it