Ultra-short-term wind power prediction method and system for complex climatic environment

Abstract

The invention discloses an ultra-short-term wind power prediction method and system for a complex climatic environment, and relates to the technical field of wind power, and the method comprises the steps: extracting the historical environment and power data of a target scene, and carrying out the analysis to obtain a typical environment condition set; main environmental components are obtained through principal component analysis, and key environmental factors are extracted to construct a basic prediction model library; predicting an ultra-short-term environment condition based on the environment state transition matrix, and configuring a model calling parameter; and according to the calling parameters and the basic model library, configuring a wind power prediction model and performing prediction. Therefore, the technical effects of improving the environment adaptability and the prediction accuracy and improving the response delay are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power, and particularly to a very short-term wind power prediction method and system for complex climate environments. Background Art

[0002] With the rapid development of the wind power industry, the grid connection of wind power has brought significant impacts on the stability and economy of the power system. Currently, common wind power prediction methods, including numerical weather prediction models, physical models, and statistical learning models, can provide prediction results to a certain extent. However, in the face of complex climate environments such as sudden weather changes and wind speed fluctuations, there are technical problems such as poor environmental adaptability, low prediction accuracy, and high response latency, which affect the operation efficiency of wind farms and the dispatching ability of the power system. Summary of the Invention

[0003] The present invention provides a very short-term wind power prediction method and system for complex climate environments to solve the technical problems of poor environmental adaptability, low prediction accuracy, and high response latency in the prior art, and to achieve the technical effects of improving environmental adaptability and prediction accuracy and improving response latency.

[0004] In a first aspect, the present invention provides a very short-term wind power prediction method for complex climate environments, wherein the method includes:

[0005] Extract the historical environmental data and historical power data of the target scenario, and parse to obtain the set of typical environmental conditions of the target scenario, wherein the historical environmental data and historical power data are stored in an associated manner.

[0006] Traverse the set of typical environmental conditions for principal component analysis to obtain the main environmental component composition of each typical environmental condition.

[0007] According to the main environmental component composition of each of the set of typical environmental conditions, extract the set of key environmental factors of the scenario, and construct and train the corresponding basic prediction model according to the set of key environmental factors of the scenario, and store it as a basic model library.

[0008] Based on a pre-constructed environmental state transition matrix, obtain the prediction result of the very short-term environmental condition, and configure the model call parameters according to the prediction result of the very short-term environmental condition.

[0009] Configure the wind power prediction model according to the model call parameters and the basic model library, and perform power prediction.

[0010] In a second aspect, the present invention further provides a very short-term wind power prediction system for complex climate environments, wherein the system includes:

[0011] A historical data extraction and analysis module is used to extract the historical environmental data and historical power data of the target scenario, and parse and obtain the typical environmental condition set of the target scenario. Among them, the historical environmental data and historical power data are stored in an associated manner.

[0012] An environmental component analysis module is used to traverse the typical environmental condition set for principal component analysis to obtain the main environmental component composition of each typical environmental condition.

[0013] A basic model construction module is used to extract the set of key environmental factors of the scenario according to the main environmental component composition of each typical environmental condition set, and construct and train the corresponding basic prediction model according to the set of key environmental factors of the scenario, and store it as a basic model library.

[0014] A call parameter configuration module is used to obtain the ultra-short-term environmental condition prediction result based on the pre-constructed environmental state transition matrix, and configure the model call parameters according to the ultra-short-term environmental condition prediction result.

[0015] A prediction execution module is used to configure the wind power prediction model according to the model call parameters and the basic model library, and perform power prediction.

[0016] The present invention discloses an ultra-short-term wind power prediction method and system for complex climate environments, including: extracting the historical environmental data and historical power data of the target scenario, and parsing and obtaining the typical environmental condition set of the target scenario. Among them, the historical environmental data and historical power data are stored in an associated manner; traversing the typical environmental condition set for principal component analysis to obtain the main environmental component composition of each typical environmental condition; extracting the set of key environmental factors of the scenario according to the main environmental component composition of each typical environmental condition set, and constructing and training the corresponding basic prediction model according to the set of key environmental factors of the scenario, and storing it as a basic model library; obtaining the ultra-short-term environmental condition prediction result based on the pre-constructed environmental state transition matrix, and configuring the model call parameters according to the ultra-short-term environmental condition prediction result; configuring the wind power prediction model according to the model call parameters and the basic model library, and performing power prediction. The ultra-short-term wind power prediction method and system for complex climate environments disclosed by the present invention solve the technical problems of poor environmental adaptability, low prediction accuracy, and high response delay, and achieve the technical effects of improving the environmental adaptability and prediction accuracy and improving the response delay. Description of the Drawings

[0017] Figure 1 It is a schematic flow chart of the ultra-short-term wind power prediction method for complex climate environments of the present invention;

[0018] Figure 2 It is a schematic structural diagram of the ultra-short-term wind power prediction system for complex climate environments of the present invention.

[0019] Explanation of the accompanying drawings: historical data extraction and analysis module 11, environmental component analysis module 12, basic model construction module 13, calling parameter configuration module 14, prediction execution module 15. DETAILED DESCRIPTION

[0020] The technical solution provided in the embodiments of the present invention is to solve the technical problems of poor environmental adaptability, low prediction accuracy and high response delay in the prior art. The overall idea adopted is as follows:

[0021] First, extract the historical environmental data and historical power data of the target scenario and store them in association. Next, parse and obtain the typical environmental condition set of the target scenario; then, perform principal component analysis on these typical environmental condition sets to extract the main environmental component composition of each typical environmental condition. Based on these main environmental component compositions, further extract the scene key environmental factor set, and build the corresponding basic prediction model based on these environmental factor sets, and store it in the basic model library. Subsequently, based on the pre-built environmental state migration matrix, obtain the prediction results of the ultra-short-term environmental conditions, and configure the model call parameters according to these prediction results. Finally, configure the wind power prediction model according to the configured model call parameters and the basic model library, and perform power prediction.

[0022] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0023] Embodiment 1

[0024] Figure 1 The present invention is a flow chart of an ultra-short-term wind power prediction method for a complex climate environment, wherein the method comprises:

[0025] The historical environmental data and the historical power data of the target scene are extracted, and a typical environmental condition set of the target scene is obtained by parsing, wherein the historical environmental data and the historical power data are associated and stored.

[0026] Specifically, first, historical environmental data (such as temperature, humidity, light intensity, etc.) and historical power data (i.e., the output power of wind power equipment) are collected through a data interface or API interaction for the target scenario. The above data has timestamps to ensure the timeliness and accuracy of the data and is stored in a database for subsequent processing.

[0027] Specifically, since the historical environmental data and historical power data need to be analyzed in association, it is preferable to use a relational database management system for data storage to ensure that the historical environmental data and historical power data can be associated through a common key value (such as a timestamp).

[0028] In some embodiments, the historical environmental data and historical power data of the target scenario are extracted, and a typical environmental condition set of the target scenario is parsed and obtained, including:

[0029] A multi-dimensional parameter space is established, and the historical environmental data is mapped into a sample point set of the multi-dimensional parameter space;

[0030] Based on the clustering analysis of the sample point set in the multi-dimensional parameter space, according to the clustering analysis result, the clustering center of each clustering cluster is extracted and defined as a typical environmental condition, and is stored as the typical environmental condition set.

[0031] Specifically, first, according to the characteristics of the target scenario, the dimensions of the multi-dimensional parameter space are defined. Optionally, the dimensions include environmental parameters such as environmental temperature, humidity, wind speed, and light intensity. Then, the historical environmental data of the target scenario is mapped into the multi-dimensional parameter space to form a sample point set. Each sample point in this sample point set represents the environmental parameters at a historical time point. Next, a suitable data analysis method, such as a clustering method based on density or distance, is used to perform clustering analysis on the sample point set to group similar sample points into the same category, thereby identifying the main environmental patterns in the target scenario. Among them, the clustering analysis result is multiple clustering clusters, and each clustering cluster corresponds to a main environmental pattern in the target scenario.

[0032] Further, according to the clustering analysis result, the clustering center of each clustering cluster is calculated, such as the average value or median value of this type of sample points in the multi-dimensional parameter space. This clustering center can be regarded as the typical environmental condition of this type of main environmental pattern. Furthermore, the clustering centers of each clustering cluster are collected, and all the clustering centers are defined as multiple typical environmental conditions and stored as the typical environmental condition set of the target scenario.

[0033] Through the above method, representative typical environmental conditions can be extracted from the historical environmental data of the target scenario, thereby providing data support and a basic reference for subsequent optimization decisions, prediction modeling, or control strategy design.

[0034] Traverse the set of typical environmental conditions for principal component analysis to obtain the main environmental component composition of each typical environmental condition.

[0035] Specifically, different typical environmental conditions correspond to different environmental index situations. Then, principal component analysis can be performed on each typical environmental condition in combination with historical data to obtain the environmental factors that have a major impact on wind power and their corresponding contribution degrees in each typical environmental condition, providing a basis for the construction of subsequent prediction models.

[0036] Specifically, principal component analysis can extract key environmental components from complex environmental data. These components can reflect the variability of environmental data to the greatest extent and can be used as inputs for prediction models to improve the accuracy and efficiency of ultra-short-term wind power prediction.

[0037] In addition, obtaining the main environmental component composition of each typical environmental condition through principal component analysis helps to reduce the input dimension of the model, reduce the computational complexity, and retain the most important information, thereby enhancing the wind power prediction efficiency.

[0038] In some embodiments, traversing the set of typical environmental conditions for principal component analysis to obtain the main environmental component composition of each typical environmental condition includes:

[0039] Interact with wind power big data to determine the set of key environmental factors;

[0040] Combined with the set of key environmental factors, traverse the set of typical environmental conditions, and perform principal component analysis on each of the typical environmental conditions respectively to obtain the corresponding environmental component composition;

[0041] Serialize the environmental component composition based on the principal component contribution degree, and select the environmental components that meet the preset cumulative principal component contribution degree as the main environmental component composition.

[0042] Specifically, first, by accessing wind power big data, obtain all environmental factors that have a significant impact on wind power generation efficiency and equipment operation from the expert system or wind power knowledge base to form a set of key environmental factors; then, for each typical environmental condition in the set of typical environmental conditions, extract the data related to the set of key environmental factors, and construct a corresponding data matrix according to the extraction results to perform principal component analysis on each typical environmental condition, decompose several principal components and calculate the contribution rate of each principal component, and record the principal components and their weights of each typical environmental condition to form the corresponding environmental component composition; furthermore, sort the contribution rates of each principal component from high to low to form a principal component contribution degree sequence, and calculate the cumulative value of the principal component contribution rate from the top of the sequence until the cumulative contribution rate reaches the preset threshold (such as 90%), and use the environmental factors corresponding to the principal components within this range as the main environmental components, and correspondingly output the main environmental component composition.

[0043] Specifically, the main environmental component composition includes the main environmental components of each typical environmental condition and their contribution rates. Among them, the contribution rate of the main environmental component group is the re-normalized contribution rate. Through the above steps, the core environmental features of typical environmental conditions can be efficiently extracted, the analysis dimension can be reduced, and thus the pertinence and reliability of data processing can be improved.

[0044] According to the main environmental component composition of each of the typical environmental condition sets, a set of key environmental factors for the scenario is extracted, and a corresponding basic prediction model is constructed and trained based on the set of key environmental factors for the scenario, and stored as a basic model library.

[0045] Specifically, the set of key environmental factors for the scenario refers to all the key environmental factors of the target scenario. In other words, the key environmental factors included in each main environmental component composition are subsets of the set of key environmental factors for the scenario.

[0046] Specifically, the basic prediction model is a machine learning or regression analysis model with a relatively simple model structure and parameters. It is constructed and trained based on the set of key environmental factors for the scenario. The obtained corresponding basic prediction model has good prediction performance in the corresponding key environmental factor dimension, while its performance in other key environmental factor dimensions is relatively low. Through the above construction method with a tendency, the obtained basic prediction model has high prediction efficiency, and the basic model library that aggregates multiple dimensions can provide more comprehensive prediction potential.

[0047] In some embodiments, according to the main environmental component composition of each of the typical environmental condition sets, a set of key environmental factors for the scenario is extracted, and a corresponding basic prediction model is constructed and trained based on the set of key environmental factors for the scenario, and stored as a basic model library, including:

[0048] Obtain the union of the main environmental component compositions corresponding to the typical environmental condition sets, and output it as the set of scenario environmental factors;

[0049] According to the occurrence frequency of the scenario environmental factors in the set of scenario environmental factors, serialize the set of scenario environmental factors, and retain the first N scenario environmental factors whose cumulative occurrence frequency meets the frequency threshold, and output it as the set of key environmental factors for the scenario, where N is a positive integer greater than or equal to 2;

[0050] Based on the set of key environmental factors for the scenario, combine the historical environmental data and the historical power data, construct and train the corresponding basic prediction model, and store it as a basic model library.

[0051] Specifically, first, the union of each main environmental component in the typical environmental conditions is merged to form a set of scenario environmental factors, ensuring that all relevant environmental factors are covered. Then, the occurrence frequencies of the scenario environmental factors in different typical environmental conditions are counted, and the set of scenario environmental factors is sorted in descending order of frequency. Next, based on the descending order result, the cumulative occurrence frequencies of the scenario environmental factors are calculated, and the top N scenario environmental factors that meet the preset frequency threshold are selected as the set of key scenario environmental factors, where N is a positive integer greater than or equal to 2.

[0052] Specifically, the feature data related to the set of key scenario environmental factors and the corresponding historical power data are extracted from the historical environmental data as the target variables, and the data is preprocessed (such as denoising, normalization, etc.) to obtain model training data to ensure the effectiveness of model training. Then, for each key scenario environmental factor, a basic prediction model is constructed. Exemplarily, the training input of the basic prediction model corresponding to each key scenario environmental factor is the data in the historical environmental data where this key environmental factor is a single variable or an approximately single variable, and the training input is the relevant power output value.

[0053] Exemplarily, the basic prediction model is constructed based on a statistical model, a machine learning model, or a deep learning model, including support vector machine (SVM), random forest (RF), long short-term memory network (LSTM), etc.

[0054] Specifically, the trained multiple basic prediction models are stored according to the corresponding key environmental factors, and their applicable ranges and prediction capabilities are marked, such as information on the training data range, model performance evaluation metrics, usage limitations, etc., to generate a basic model library.

[0055] Optionally, multiple basic prediction models with different model scales and structures are constructed and trained for each key scenario environmental factor to improve the diversity of the basic model library.

[0056] Through the above method steps, key environmental factors can be effectively extracted and an efficient basic prediction model library can be generated, providing a model basis for the dynamic power prediction and environmental analysis of the target scenario.

[0057] Based on the pre-constructed environmental state transition matrix, the ultra-short-term environmental condition prediction results are obtained, and the model call parameters are configured according to the ultra-short-term environmental condition prediction results.

[0058] Specifically, the environmental state transition matrix is a probability matrix that describes the probability of transitioning from one environmental state to another. It is used to reflect the environmental change characteristics of the target scenario and can predict the change trend of the environmental state. By using the state transition matrix, it is possible to combine the current environmental state to predict the possible environmental state at the next time step, thereby providing a basis for setting the model call parameters and improving the adaptability of the subsequent prediction model to the ultra-short-term environment of the target scenario.

[0059] In some embodiments, based on a pre-constructed environmental state transition matrix, an ultra-short-term environmental condition prediction result is obtained, and model call parameters are configured according to the ultra-short-term environmental condition prediction result, including:

[0060] Perform interval discretization analysis on the historical environmental data according to the set of key environmental factors of the scenario, and statistically analyze the results to construct the environmental state transition matrix;

[0061] Obtain the real-time environmental data of the target scenario through real-time monitoring, and parse the real-time environmental data to determine the real-time environmental state distribution;

[0062] Combine the real-time environmental state distribution with the environmental state transition matrix to predict the environmental state distribution at a preset future time step, and output it as the ultra-short-term environmental condition prediction result;

[0063] Based on the ultra-short-term environmental condition prediction result, extract the predicted environmental state and the corresponding prediction probability, and combine the main environmental component composition of each typical environmental condition, and calculate the call severity coefficient of each basic model by weighted calculation to obtain the model call parameters.

[0064] Specifically, first, according to the set of key environmental factors of the scenario, the historical environmental data is segmented according to a preset interval to generate a discretized data set of the environmental state distribution interval, and the discretized data set is statistically analyzed to analyze the transition probability of the environmental state from the current state to other states. Then, analyze the environmental state distribution at different time steps to construct the environmental state transition matrix, and each element of this environmental state transition matrix represents the probability value of transitioning from one state to another.

[0065] Specifically, activate the real-time sensor to collect the environmental data of the target scenario, and extract the feature data related to the set of key environmental factors of the scenario; then, classify the real-time environmental data according to the discretization interval standard of the environmental state transition matrix to determine the current real-time environmental state distribution.

[0066] Further, using the real-time environmental state distribution as the initial state at the current moment, combined with the environmental state transition matrix, calculate the distribution probability of the future environmental state at a preset time step to obtain the ultra-short-term environmental condition prediction result, where the ultra-short-term environmental condition prediction result includes each possible future environmental state and its corresponding prediction probability.

[0067] Specifically, based on the ultra-short-term environmental condition prediction result, extract the predicted main environmental states and their corresponding probability distributions. Then, combine the main environmental component compositions of the typical environmental conditions corresponding to each predicted main environmental state to calculate the call severity coefficient of each basic model, which reflects the comprehensive probability proportion of the main environmental components in the prediction result. Next, normalize the call severity coefficient of each basic model as the model call parameter to ensure flexible call of different models according to their relevance in the target scenario prediction.

[0068] Configure the wind power prediction model according to the model call parameter and the basic model library, and perform power prediction.

[0069] In some embodiments, initializing the wind power prediction model according to the model call parameter and the basic model library and performing power prediction includes:

[0070] Parse the historical environmental data to obtain the historical environmental component composition;

[0071] Based on the historical environmental component composition, perform basic model calls in the basic model library to obtain a reference basic model group;

[0072] Input the historical environmental data into the reference basic model group to obtain a basic prediction reference data set, and use the basic prediction reference data set and the historical power data as training samples to construct and train an integrated mapping model;

[0073] Based on the model call parameter, perform repeatable calls in the basic model library to obtain a target basic model group, and integrate the target basic model group and the integrated mapping model to obtain the wind power prediction model.

[0074] Specifically, first, use the set of key environmental factors of the scenario to extract and analyze the historical environmental data to generate the historical environmental component composition, which includes the key environmental factors corresponding to the historical data and their main component contribution degrees. Then, screen and match the basic model group according to the historical environmental component composition, and output it as the reference basic model group. In other words, the reference basic model group consists of models associated with the historical environmental components in the basic model library.

[0075] Specifically, the historical environmental data is input into the reference basic model group to obtain the corresponding predicted output data, forming a basic prediction reference data set. Then, the basic prediction reference data set is matched with the historical power data to form a training sample data set. The training sample data set includes the composition of key environmental factors and the correlation information between the basic model prediction output and the actual power value. Through this training sample data set, an integrated mapping model can be trained. The integrated mapping model takes into account the interaction relationships of multiple key environmental factors and has the ability to adaptively fuse the prediction outputs of multiple basic prediction models, thereby optimizing the overall performance of power prediction.

[0076] Further, according to the model call parameters generated based on the real-time environmental conditions, the target basic model group is dynamically selected from the basic model library. The target basic model group is composed of multiple types of basic models in corresponding proportions of the corresponding model call parameters, thereby ensuring that the subsequent generated wind power prediction model has good real-time performance and adaptability.

[0077] Specifically, the target basic model group and the integrated mapping model are dynamically combined to construct the final wind power prediction model. By combining the diversity of the basic model group and the optimization ability of the integrated mapping model, high-precision power prediction can be achieved. Among them, the output result of the target basic model group is the output of the integrated mapping model.

[0078] In some implementation manners, using the basic prediction reference data set and the historical power data as training samples, constructing and training an integrated mapping model includes:

[0079] Obtain the reference model call parameters corresponding to the reference basic model group, and establish the correlation relationships among the reference model call parameters, the basic prediction output data set, and the historical power data;

[0080] Construct the integrated mapping model based on machine learning, and use the model call parameters and the basic prediction output data set as sample inputs and the historical power data as sample outputs to perform supervised training of the integrated mapping model;

[0081] Verify the model performance of the integrated mapping model. If the integrated mapping model performs well under all the reference basic model groups in the historical environmental data, output the integrated mapping model.

[0082] Specifically, first, obtain the calling parameters corresponding to the reference basic model group, that is, the calling weights of the reference basic model group for different types of models; then, establish the mapping relationship between the reference model calling parameters, the basic prediction output data set, and the historical power data to form an associated feature data set; next, select a regression model (such as random forest regression, gradient boosting decision tree, or deep learning model) as the implementation method of the integrated mapping model, use the combined feature data of the reference model calling parameters and the basic prediction output data set as the training input, and use the historical power data as the true target value of the integrated mapping model for supervised learning, and dynamically adjust the model hyperparameters to optimize the mapping performance of the integrated mapping model.

[0083] Specifically, use the historical environmental data and the prediction output of the reference basic model group for verification. Among them, the evaluation indicators for verification include mean square error MSE, mean absolute error MAE, prediction accuracy, etc.; when the integrated mapping model performs well in all historical environmental data and the reference basic model group, that is, when the set performance index threshold is met, then output this model as the final integrated mapping model.

[0084] Through the above process, cover all scenarios of the reference basic model group, ensure the generalization ability and stability of the model, and further enable the integrated mapping model to effectively integrate the outputs of multiple basic prediction models, improving the power prediction accuracy and robustness in complex environments.

[0085] In summary, the ultra-short-term wind power prediction method for complex climate environments provided by the present invention has the following technical effects:

[0086] By extracting the historical environmental data and historical power data of the target scenario, and parsing to obtain the typical environmental condition set of the target scenario, where the historical environmental data and historical power data are associated and stored; traversing the typical environmental condition set for principal component analysis to obtain the main environmental component composition of each typical environmental condition; according to the main environmental component composition of each typical environmental condition in the typical environmental condition set, extract the key environmental factor set of the scenario, and construct and train the corresponding basic prediction model according to the key environmental factor set of the scenario, and store it as the basic model library; based on the pre-constructed environmental state transition matrix, obtain the ultra-short-term environmental condition prediction result, and configure the model calling parameters according to the ultra-short-term environmental condition prediction result; configure the wind power prediction model according to the model calling parameters and the basic model library, and perform power prediction, so as to achieve the technical effects of improving environmental adaptability and prediction accuracy and improving response latency.

[0087] Embodiment 2

[0088] Figure 2 It is a schematic structural diagram of the ultra-short-term wind power prediction system for complex climate environments of the present invention. For example, Figure 1The flow diagram of the ultra-short-term wind power prediction method for complex climate environments of the present invention can be implemented through a structure as shown in Figure 2 below.

[0089] Based on the same concept as the ultra-short-term wind power prediction method for complex climate environments in the above embodiments, the ultra-short-term wind power prediction system provided by the present invention includes:

[0090] A historical data extraction and analysis module 11, configured to extract historical environmental data and historical power data of a target scenario, and analyze and obtain a set of typical environmental conditions of the target scenario, wherein the historical environmental data and the historical power data are stored in an associated manner.

[0091] An environmental component analysis module 12, configured to traverse the set of typical environmental conditions for principal component analysis to obtain the composition of the main environmental components of each typical environmental condition.

[0092] A basic model construction module 13, configured to extract a set of key environmental factors of the scenario according to the composition of each main environmental component of the set of typical environmental conditions, and construct and train a corresponding basic prediction model according to the set of key environmental factors of the scenario, and store it as a basic model library.

[0093] A calling parameter configuration module 14, configured to obtain a prediction result of the ultra-short-term environmental condition based on a pre-constructed environmental state transition matrix, and configure model calling parameters according to the prediction result of the ultra-short-term environmental condition.

[0094] A prediction execution module 15, configured to configure a wind power prediction model according to the model calling parameters and the basic model library, and perform power prediction.

[0095] Among them, the historical data extraction and analysis module 11 includes:

[0096] A multi-dimensional parameter space construction unit, configured to establish a multi-dimensional parameter space and map the historical environmental data into a set of sample points in the multi-dimensional parameter space.

[0097] A clustering analysis unit, configured to perform clustering analysis on the set of sample points in the multi-dimensional parameter space, and according to the clustering analysis result, extract the clustering center of each clustering cluster as a typical environmental condition, and store it as the set of typical environmental conditions.

[0098] Among them, the environmental component analysis module 12 includes:

[0099] A key environmental factor identification unit, configured to interact with wind power big data to determine a set of key environmental factors.

[0100] A typical environmental condition analysis unit is used to traverse the set of typical environmental conditions in combination with the set of key environmental factors, and perform principal component analysis on each of the typical environmental conditions respectively to obtain the corresponding environmental component composition.

[0101] An environmental component selection unit is used to serialize the environmental component composition based on the principal component contribution degree, and select the environmental components that meet the preset cumulative principal component contribution degree as the main environmental component composition.

[0102] Among them, the basic model construction module 13 includes:

[0103] A scene environmental factor set construction unit is used to obtain the union of the multiple main environmental component compositions corresponding to the set of typical environmental conditions, and output it as the scene environmental factor set.

[0104] A scene key environmental factor set determination unit is used to serialize the scene environmental factor set according to the occurrence frequency of the scene environmental factors in the scene environmental factor set, and retain the first N scene environmental factors whose cumulative occurrence frequency meets the frequency threshold, and output it as the scene key environmental factor set, where N is a positive integer greater than or equal to 2.

[0105] A basic prediction model construction and training unit is used to construct and train the corresponding basic prediction model based on the scene key environmental factor set, in combination with the historical environmental data and the historical power data, and store it in the basic model library.

[0106] Among them, the call parameter configuration module 14 includes:

[0107] An environmental state transition matrix construction unit is used to perform interval discretization analysis on the historical environmental data according to the scene key environmental factor set, and statistically analyze the results to construct the environmental state transition matrix.

[0108] A real-time environmental state monitoring unit is used to monitor and obtain the real-time environmental data of the target scene in real time, and analyze the real-time environmental data to determine the real-time environmental state distribution.

[0109] A very short-term environmental condition prediction unit is used to predict the environmental state distribution in the future preset time step in combination with the real-time environmental state distribution and the environmental state transition matrix, and output it as the very short-term environmental condition prediction result.

[0110] A model call parameter calculation unit is used to extract the predicted environmental state and the corresponding predicted probability based on the very short-term environmental condition prediction result, and combine the main environmental component composition of each of the typical environmental conditions to calculate the call weight coefficient of each basic model by weighting, and obtain the model call parameter.

[0111] Among them, the prediction execution module 15 includes:

[0112] A historical environment component analysis unit for analyzing the historical environment data to obtain the composition of historical environment components.

[0113] A basic model calling unit for calling a basic model in the basic model library according to the composition of the historical environment components to obtain a reference basic model group.

[0114] An integrated mapping construction unit for inputting the historical environment data into the reference basic model group to obtain a basic prediction reference data set, and using the basic prediction reference data set and the historical power data as training samples to construct and train an integrated mapping model.

[0115] A wind power prediction model integration unit for repeatedly calling in the basic model library based on the model calling parameters to obtain a target basic model group, and integrating the target basic model group and the integrated mapping model to obtain the wind power prediction model.

[0116] Further, the execution steps of the integrated mapping construction unit in the prediction execution module 15 further include:

[0117] Obtaining the reference model calling parameters corresponding to the reference basic model group, and establishing the association relationship between the reference model calling parameters, the basic prediction output data set and the historical power data.

[0118] Constructing the integrated mapping model based on machine learning, and using the model calling parameters and the basic prediction output data set as sample inputs and the historical power data as sample outputs to perform supervised training on the integrated mapping model.

[0119] Verifying the model performance of the integrated mapping model. If the integrated mapping model performs well under all the reference basic model groups in the historical environment data, output the integrated mapping model.

[0120] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the foregoing Embodiment 1 are equally applicable to the ultra-short-term wind power prediction system for complex climate environments described in Embodiment 2. For the sake of brevity of the specification, no further elaboration is made here.

[0121] It should be understood that the embodiments and the above descriptions disclosed in the present invention enable those skilled in the art to implement the present invention using the present invention. At the same time, the present invention is not limited to the above-mentioned part of the embodiments. It should be understood that those of ordinary skill in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An ultra-short-term wind power forecasting method for complex climate environments, characterized in that: The method comprises: Extracting historical environmental data and historical power data of a target scene, and parsing to obtain a typical environmental condition set of the target scene, wherein the historical environmental data and the historical power data are associated and stored; Traversing the typical environmental condition set to perform principal component analysis to obtain the main environmental component composition of each typical environmental condition; Extracting a scene key environmental factor set according to each of the main environmental components of the typical environmental condition set, and constructing a basic prediction model corresponding to the training according to the scene key environmental factor set, and storing it in a basic model library; Based on the pre-built environmental state migration matrix, obtain the ultra-short-term environmental condition prediction result, and configure the model calling parameters according to the ultra-short-term environmental condition prediction result; According to the model calling parameters and the basic model library, a wind power prediction model is configured and power prediction is performed.

2. The ultra-short-term wind power prediction method for complex climate environment according to claim 1, characterized in that: Extract the historical environmental data and historical power data of the target scene, and parse and obtain the typical environmental condition set of the target scene, including: Establishing a multidimensional parameter space, and mapping the historical environmental data to a sample point set of the multidimensional parameter space; A cluster analysis is performed on the sample point set in the multi-dimensional parameter space. According to the cluster analysis result, a cluster center of each cluster is extracted and defined as a typical environmental condition, and the cluster center is stored as the typical environmental condition set.

3. The ultra-short-term wind power prediction method for complex climate environment according to claim 2, characterized in that: The typical environmental condition set is traversed to perform principal component analysis to obtain the main environmental component composition of each typical environmental condition, including: Interactive wind power big data to determine the key environmental factor set; In combination with the key environmental factor set, the typical environmental condition set is traversed, and principal component analysis is performed on each of the typical environmental conditions to obtain the corresponding environmental component composition; The environmental component compositions are serialized based on the principal component contribution, and the environmental components satisfying the preset cumulative principal component contribution are selected as the main environmental component compositions.

4. The ultra-short-term wind power prediction method for complex climate environments according to claim 3, characterized in that: According to each of the main environmental components of the typical environmental condition set, a scene key environmental factor set is extracted, and a basic prediction model corresponding to the training is constructed according to the scene key environmental factor set, and stored in a basic model library, including: Obtaining a union of a plurality of main environmental components corresponding to the typical environmental condition set, and outputting the union as a scene environmental factor set; According to the occurrence frequency of the scene environment factors in the scene environment factor set, the scene environment factor set is serialized, and the first N scene environment factors whose cumulative occurrence frequency meets the frequency threshold are retained, and the output is the scene key environment factor set, where N is a positive integer greater than or equal to 2; Based on the set of key environmental factors of the scenario, combined with the historical environmental data and the historical power data, a corresponding basic prediction model is constructed and trained and stored in a basic model library.

5. The ultra-short-term wind power prediction method for complex climate environment according to claim 4, characterized in that: Based on the pre-built environmental state transition matrix, an ultra-short-term environmental condition prediction result is obtained, and model call parameters are configured according to the ultra-short-term environmental condition prediction result, including: According to the set of key environmental factors of the scenario, the historical environmental data is discretized between partitions, and statistical analysis results are performed to construct the environmental state migration matrix; Real-time monitoring to obtain real-time environmental data of the target scene, and analyzing the real-time environmental data to determine the real-time environmental state distribution; Combining the real-time environmental state distribution with the environmental state migration matrix, predicting the environmental state distribution at a preset time step in the future, and outputting the ultra-short-term environmental state prediction result; Based on the ultra-short-term environmental condition prediction results, the predicted environmental state and the corresponding prediction probability are extracted, and combined with the main environmental component composition of each typical environmental condition, the call weight coefficient of each basic model is obtained by weighted calculation to obtain the model call parameters.

6. The ultra-short-term wind power prediction method for complex climate environment according to claim 5, characterized in that: Initializing the wind power prediction model according to the model calling parameters and the basic model library, and performing power prediction, including: Analyze the historical environmental data to obtain the historical environmental component composition; According to the historical environmental component composition, a basic model is called in the basic model library to obtain a reference basic model group; Inputting the historical environmental data into the reference basic model group, obtaining a basic prediction reference data set, and using the basic prediction reference data set and the historical power data as training samples to construct and train an integrated mapping model; Based on the model calling parameters, the basic model library is repeatedly called to obtain a target basic model group, and the target basic model group is integrated with the integrated mapping model to obtain the wind power prediction model.

7. The ultra-short-term wind power prediction method for complex climate environment according to claim 6, characterized in that: Using the basic prediction reference data set and the historical power data as training samples, constructing and training an integrated mapping model, including: Acquire reference model calling parameters corresponding to the reference basic model group, and establish an association relationship among the reference model calling parameters, the basic prediction output data set and the historical power data; Constructing the integrated mapping model based on machine learning, and using the model calling parameters and the basic prediction output data set as sample inputs and the historical power data as sample outputs to perform supervised training of the integrated mapping model; The model performance of the integrated mapping model is verified, and if the integrated mapping model performs well under all the reference basic model groups under the historical environmental data, the integrated mapping model is output.

8. Ultra-short-term wind power prediction system for complex climate environment, characterized by: The system is used to execute the ultra-short-term wind power prediction method for complex climate environments according to any one of claims 1 to 7, and the system comprises: A historical data extraction and analysis module, used to extract historical environmental data and historical power data of a target scene, and parse and obtain a typical environmental condition set of the target scene, wherein the historical environmental data and the historical power data are associated and stored; An environmental component analysis module is used to traverse the typical environmental condition set to perform principal component analysis and obtain the main environmental component composition of each typical environmental condition; A basic model building module, used to extract a scene key environmental factor set according to each of the main environmental components of the typical environmental condition set, and to build a basic prediction model corresponding to the training according to the scene key environmental factor set, and store it in a basic model library; A calling parameter configuration module is used to obtain an ultra-short-term environmental condition prediction result based on a pre-built environmental state migration matrix, and configure a model calling parameter according to the ultra-short-term environmental condition prediction result; The prediction execution module is used to configure the wind power prediction model and perform power prediction according to the model calling parameters and the basic model library.

Images