A Wind Power Control System Based on Wind Speed Prediction

Abstract

The present application provides a wind power control system based on wind speed prediction, which relates to the field of intelligent scheduling control. It obtains the time series of real-time wind speed values and the historical data of wind speed values within a predetermined time period, encodes the real-time wind speed time series and historical subsequences respectively, extracts the global long-term dependence and local disturbance features, then dynamically fuses the real-time features and the semantic association of historical similar scenarios through decision anchor points, and then generates a high-precision wind speed prediction value based on the fusion. Based on the real-time deviation analysis of the prediction value and the cut-out wind speed, combined with the dynamic threshold adjustment mechanism, a hierarchical power limit strategy is triggered. This solution automatically captures the non-linear time series law of wind speed through deep learning, dynamically responds to the changes of actual working conditions, effectively improves the prediction accuracy, shortens the control delay, optimizes the power generation efficiency and ensures the stability of the grid voltage.

Description

Technical Field

[0001] This application relates to the field of intelligent dispatching control, and more particularly, in the embodiments of this application, it relates to a wind power control system based on wind speed prediction. Background Art

[0002] With the acceleration of the global energy transformation process, wind power, as an important part of clean energy, has been continuously increasing its proportion in the power system. However, wind power has strong randomness and volatility, and its output power is directly affected by wind speed changes. The instantaneous fluctuations of wind speed may cause the output power of the wind farm to fluctuate violently, thereby threatening the stability of the power grid and reducing the power generation efficiency. Among them, the existing patent CN116960978A introduces a method for predicting the power of offshore wind farms based on the combined decomposition and reconstruction of wind speed and power, which improves the prediction accuracy of offshore wind power through combined decomposition, fuzzy entropy reconstruction, LSTM modeling, and Bayesian optimization.

[0003] However, there are still many problems in the existing wind power control technologies. On the one hand, in the wind speed prediction link, in the past, only simple statistical models were relied on to process wind speed data, and the complex and variable characteristics of the wind speed time series were insufficiently mined. For example, traditional methods are difficult to effectively capture the fluctuation laws of wind speed at different time scales and cannot make full use of the complex correlations between real-time wind speed and historical wind speed data, resulting in low wind speed prediction accuracy. On the other hand, in the wind power control strategy, the traditional method lacks a dynamic adjustment mechanism and usually performs power control based on fixed thresholds, failing to make timely and accurate responses according to the changes in real-time wind speed.

[0004] Therefore, an optimized wind power control scheme based on wind speed prediction is expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide a wind power control system based on wind speed prediction. First, it obtains the time series of real-time wind speed values within a predetermined period collected by a wind speed sensor and obtains the historical data of wind speed values. Using deep learning-based data analysis and coding techniques, it encodes the real-time wind speed time series and historical subsequences respectively, extracts global long-term dependence and local perturbation features, and then dynamically fuses the real-time features and semantic associations of historical similar scenarios through decision-making anchor points. After that, it generates a high-precision wind speed prediction value based on the fusion, and based on the real-time deviation analysis between the prediction value and the cut-out wind speed, combines the dynamic threshold adjustment mechanism to trigger a hierarchical power limit strategy. This scheme automatically captures the non-linear time series law of wind speed through deep learning, dynamically responds to the changes in actual working conditions, effectively improves the prediction accuracy, shortens the control delay, optimizes the power generation efficiency, and ensures the stability of the grid voltage.

[0006] According to one aspect of the present application, there is provided a wind power control system based on wind speed prediction, which includes:

[0007] A real-time wind speed value acquisition module for obtaining a time series of real-time wind speed values within a predetermined period collected by a wind speed sensor;

[0008] A historical data acquisition module for wind speed values for obtaining historical data of wind speed values;

[0009] A query encoding module for query encoding the time series of the real-time wind speed values and the historical data of the wind speed values to obtain short-term wind speed prediction values. Among them, the query encoding module includes: a real-time and historical wind speed time series encoding unit for performing wind speed semantic query dynamic encoding based on temporal correlation features on the time series of the real-time wind speed values and the historical data of the wind speed values to obtain real-time wind speed semantic query dynamic response encoding features; a short-term wind speed prediction unit for obtaining short-term wind speed prediction values based on the real-time wind speed semantic query dynamic response encoding features;

[0010] A restricted power strategy determination module for calculating the difference between the short-term wind speed prediction value and the cut-out wind speed value, and determining whether to adopt a restricted power strategy based on the comparison between the difference and a preset threshold.

[0011] Compared with the prior art, the wind power control system based on wind speed prediction provided by the present application first obtains a time series of real-time wind speed values within a predetermined period collected by a wind speed sensor, and obtains historical data of wind speed values. It uses data analysis and encoding techniques based on deep learning to encode the real-time wind speed time series and the historical subsequence respectively, extracts global long-term dependence and local perturbation features, and then dynamically fuses real-time features with semantic associations of historical similar scenarios through decision anchor points. After that, it generates high-precision wind speed prediction values based on the fusion, and based on the real-time deviation analysis between the prediction value and the cut-out wind speed, combines a dynamic threshold adjustment mechanism to trigger a hierarchical power restriction strategy. This solution automatically captures the non-linear temporal law of wind speed through deep learning, dynamically responds to changes in actual working conditions, effectively improves prediction accuracy, shortens control delay, optimizes power generation efficiency, and ensures grid voltage stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. They are used together with the embodiments of the present application to explain the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1It is a system block diagram of a wind power control system based on wind speed prediction according to an embodiment of the present application.

[0014] Figure 2 It is a schematic diagram of data flow of a wind power control system based on wind speed prediction according to an embodiment of the present application.

[0015] Figure 3 It is a block diagram of a query coding module in a wind power control system based on wind speed prediction according to an embodiment of the present application.

[0016] Figure 4 It is a block diagram of a real-time historical wind speed time series coding unit in a wind power control system based on wind speed prediction according to an embodiment of the present application.

[0017] Figure 5 It is a block diagram of a real-time wind speed semantic query sub-unit in a wind power control system based on wind speed prediction according to an embodiment of the present application.

[0018] Figure 6 It is a block diagram of a splicing weight factor calculation secondary sub-unit in a wind power control system based on wind speed prediction according to an embodiment of the present application. Detailed implementation manners

[0019] Hereinafter, various exemplary embodiments, features and aspects of the present application will be described in detail with reference to the drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0020] The special term "exemplary" here means "serving as an example, an embodiment or illustrative". Any embodiment described as "exemplary" here does not have to be construed as being superior or better than other embodiments.

[0021] In addition, in order to better illustrate the present application, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present application can also be implemented without some specific details. In some instances, methods, means, elements and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0022] Furthermore, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, "a plurality" means two or more unless otherwise specifically defined.

[0023] In response to the above technical problems, the present application proposes a wind power control system based on wind speed prediction. Figure 1 4 is a system block diagram of a wind power control system based on wind speed prediction according to an embodiment of the present application. Figure 2 FIG. 1 is a schematic diagram of data flow of a wind power control system based on wind speed prediction according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the wind speed prediction-based wind power control system 100 of the embodiment of the present application, it includes: a real-time wind speed value acquisition module 110, which is used to obtain the time series of real-time wind speed values within a predetermined time period collected by the wind speed sensor; a wind speed value historical data acquisition module 120, which is used to obtain the historical data of the wind speed value; a query coding module 130, which is used to query and encode the time series of the real-time wind speed value and the historical data of the wind speed value to obtain a short-term prediction value of the wind speed; a power limiting strategy determination module 140, which is used to calculate the difference between the short-term prediction value of the wind speed and the cut-out wind speed value, and determine whether to adopt the power limiting strategy based on the comparison between the difference and the preset threshold.

[0024] In the above-mentioned wind power control system 100 based on wind speed prediction, the real-time wind speed value acquisition module 110 is used to obtain the time series of real-time wind speed values collected by the wind speed sensor within a predetermined time period. It should be understood that by continuously monitoring and recording the wind speed within a predetermined time period by the wind speed sensor, the system can obtain a detailed and continuous dynamic data set. These data reflect the actual operating status of the wind farm within a specific time range, including the wind speed changes from seconds to hours or even longer time periods. Specifically, the high-precision wind speed sensors deployed on site will continuously measure the wind speed in the environment and transmit these measurement results to the control system in the form of a time series. This continuous data collection method helps to capture instantaneous fluctuations in wind speed and long-term trends. For example, in a short period of time, gusts may cause severe jitters in the output power of a wind farm; on a longer time scale, seasonal changes may affect the overall distribution pattern of wind speed. By accurately capturing these short-term and long-term wind speed characteristics, the system can better understand the current wind speed pattern and its future changing trends. In addition, since wind power is highly random and volatile, its output power is directly affected by wind speed changes. Accurately grasping real-time wind speed information can optimize wind power control. Specifically, wind speed sensors are usually installed in different locations of wind farms to fully cover the entire area and provide multi-dimensional data support. The data collected by the sensors is transmitted to the central control system via wired or wireless communication technology, where the data will be further processed and analyzed.

[0025] In the above-mentioned wind power control system 100 based on wind speed prediction, the historical data acquisition module 120 of the wind speed value is used to obtain the historical data of the wind speed value. It should be understood that due to the significant randomness and volatility of the wind speed, it is difficult to capture all possible change patterns based solely on real-time data. The historical data records various changes in the wind speed over a long period of time, including the typical wind speed distribution under normal operating conditions and extreme wind speed events under abnormal conditions (such as storms or strong gusts). These data can help the system learn and adapt to various different wind speed scenarios, so as to make more accurate predictions and responses when facing similar situations in the future. For example, in a specific area, if strong winds often occur in autumn historically, the system can be prepared in advance based on this experience and take corresponding preventive measures to ensure the safe and stable operation of the power grid. Moreover, considering that the effectiveness of deep learning models highly depends on the quality and diversity of training data. Training only with real-time wind speed data often cannot cover all potential patterns, which may lead to overfitting of the model or insufficient generalization ability. Training in combination with historical data can enable the model to access a wider sample space, thereby enhancing its adaptability to unknown situations. Specifically, historical data can be used to generate a large number of training samples, covering the changes in wind speed at different time periods and under different meteorological conditions. These diverse samples help the model learn more complex feature representations, thereby improving the prediction accuracy. It is worth mentioning that many meteorological stations and research institutions publicly release detailed meteorological observation data, which includes long-term wind speed measurement results. These external resources can be used as supplementary data sources to enrich the training set of the system.

[0026] Figure 3 Block diagram of the query encoding module in the wind power control system based on wind speed prediction according to an embodiment of the present application. As Figure 3 shown, the query encoding module 130 is used to perform query encoding on the time series of the real-time wind speed value and the historical data of the wind speed value to obtain a short-term wind speed prediction value, including: a real-time historical wind speed time series encoding unit 131, which is used to perform wind speed semantic query dynamic encoding based on time series correlation features on the time series of the real-time wind speed value and the historical data of the wind speed value to obtain a real-time wind speed semantic query dynamic response encoding feature; a short-term wind speed prediction unit 132, which is used to obtain a short-term wind speed prediction value based on the real-time wind speed semantic query dynamic response encoding feature.

[0027] Figure 4 Block diagram of the real-time historical wind speed time series encoding unit in the wind power control system based on wind speed prediction according to an embodiment of the present application. As Figure 4As shown, in an embodiment of the present application, the real-time historical wind speed time series encoding unit 131 includes: a wind speed value historical data sequence segmentation subunit 1311, which is used to perform sequence segmentation on the historical data of the wind speed value based on the predetermined time period to obtain a set of historical wind speed value time series subsequences; a wind speed time series encoding subunit 1312, which is used to perform time series encoding on the time series of the real-time wind speed value and each historical wind speed value time series subsequence in the set of historical wind speed value time series subsequences to obtain a real-time wind speed time series associated implicit feature coding vector and a set of historical wind speed local time series associated implicit feature coding vectors; a real-time wind speed semantic query subunit 1313, which is used to perform wind speed time series feature decision anchor point semantic adaptive query encoding on the set of the real-time wind speed time series associated implicit feature coding vector and the historical wind speed local time series associated implicit feature coding vector to obtain a real-time wind speed semantic query dynamic response coding vector as the real-time wind speed semantic query dynamic response coding feature.

[0028] In an embodiment of the present application, the wind speed value historical data sequence segmentation subunit 1311 is used to perform sequence segmentation on the historical data of the wind speed value based on the predetermined time period to obtain a set of historical wind speed value time series subsequences. It should be understood that, considering that the historical wind speed data contains local time series features similar to the current real-time wind speed evolution mode (such as gust precursors, periodic fluctuations), if the historical data is directly modeled as a whole, the local key mode will be submerged by the global statistical features due to the difference in time span, and the dynamic change law of the real-time wind speed cannot be effectively matched. To this end, the present application performs sequence segmentation of the historical data with equal time windows based on the predetermined time period to generate a set of historical wind speed value time series subsequences comparable to the current real-time sequence. That is, the unsegmented complete historical sequence will mask the interference of local features on the global modeling (for example, strong wind shear events are averaged), resulting in the model being unable to distinguish between long-term laws and transient anomalies, while the local features of different time scales can be separated by sequence segmentation, providing fine-grained input for the subsequent identification of the potential correlation law between the current wind speed trend and similar historical scenes.

[0029] In the embodiment of the present application, the wind speed time series encoding subunit 1312 is configured to: perform time series encoding based on the LSTM model on the time series of the real-time wind speed value and each historical wind speed value time series subsequence in the set of historical wind speed value time series subsequences to obtain the real-time wind speed time series correlation implicit feature encoding vector and the set of historical wind speed local time series correlation implicit feature encoding vectors. It should be understood that considering that the time series of the real-time wind speed value is the global dynamics at the current moment, that is, it represents the overall trend of the current operating state of the wind farm (such as the monthly wind speed distribution law), it is necessary to capture its continuous impact on power output from a global perspective; while the scope of action of each historical wind speed value time series subsequence is the local pattern in the entire historical data within a specific time window, and it is necessary to extract high-frequency fluctuations (such as sudden drops in single-day wind speed) or periodic laws (such as weekly cycle patterns) through local encoding. However, traditional linear models are difficult to capture long-period evolution laws and short-term disturbance patterns simultaneously due to the lack of memory units and gating mechanisms, resulting in significant deviations in real-time prediction and historical pattern matching. Based on this, the present application uses time series encoding based on the LSTM model to perform global time series modeling on the real-time wind speed time series, captures long-term dependence relationships through the forget gate and input gate mechanisms, and generates the real-time wind speed time series correlation implicit feature encoding vector; at the same time, performs parallel LSTM encoding on each subsequence in the set of historical wind speed subsequences, and uses its cell state memory characteristics to extract local disturbance features, forming a set of historical wind speed local time series correlation implicit feature encoding vectors. In this way, through the non-linear time series modeling ability of LSTM, the original wind speed sequence is compressed into a feature vector with high information density, realizing multi-scale dynamic feature decoupling. The real-time encoding vector represents the macroscopic evolution trend of the current wind speed, and the set of historical wind speed local encoding vectors stores the microscopic disturbance patterns.

[0030] Figure 5 The block diagram of the real-time wind speed semantic query subunit in the wind power control system based on wind speed prediction according to the embodiment of the present application. As Figure 5As shown, in the embodiment of the present application, the real-time wind speed semantic query subunit 1313 includes: an anchoring coding matrix calculation secondary subunit 1313-1, configured to calculate the anchoring coding matrix of each historical wind speed local temporal correlation implicit feature coding vector in the set of the real-time wind speed temporal correlation implicit feature coding vector and the historical wind speed local temporal correlation implicit feature coding vector to obtain a set of real-time wind speed-historical wind speed local temporal feature response anchoring coding matrices; a splicing weight factor calculation secondary subunit 1313-2, configured to determine the decision anchor adaptive splicing weight factor of each real-time wind speed-historical wind speed local temporal feature response anchoring coding matrix in the set of the real-time wind speed-historical wind speed local temporal feature response anchoring coding matrices to obtain a set of real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing weight factors; a real-time wind speed semantic query feature fusion secondary subunit 1313-3, configured to fuse the set of the real-time wind speed-historical wind speed local temporal feature response anchoring coding matrices based on the set of the real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing weight factors to obtain the real-time wind speed semantic query dynamic response coding vector.

[0031] In the embodiment of the present application, the real-time wind speed semantic query subunit 1313 is configured to perform wind speed temporal feature decision anchor point semantic adaptive query coding on the set of the real-time wind speed temporal correlation implicit feature coding vector and the historical wind speed local temporal correlation implicit feature coding vector to obtain a real-time wind speed semantic query dynamic response coding vector as the real-time wind speed semantic query dynamic response coding feature. It should be understood that considering that in the wind power control scenario, the real-time wind speed temporal correlation implicit feature coding vector and the historical wind speed local temporal correlation implicit feature coding vector respectively carry complementary information of the global dynamic pattern and the multi-scale local perturbation. However, due to the lack of a dynamic feature fusion mechanism in the traditional method, it is difficult to coordinate the interaction relationship between the global and local features, resulting in prediction errors and control lags. For example, if the global and local dynamic alignment is not achieved, problems such as "long-term trend drowning short-term mutations" or "local noise interfering with global modeling" may occur, directly affecting the wind speed prediction accuracy and the adaptability of the power control strategy. Based on this, the present application performs wind speed temporal feature decision anchor point semantic adaptive query coding on the set of the real-time wind speed temporal correlation implicit feature coding vector and the historical wind speed local temporal correlation implicit feature coding vector to obtain a real-time wind speed semantic query dynamic response coding vector. In this way, through semantic response decision anchoring, using the global coding of the real-time wind speed as the anchor point, the semantic contribution intensity of the historical local features is dynamically quantified, and then an adaptive splicing factor sequence is generated through weight normalization to achieve global-local semantic alignment, dynamic feature screening, and multi-granularity feature fusion, and generate a real-time wind speed semantic query dynamic response coding vector that integrates global trend constraints and historical experience migration.

[0032] In an embodiment of the present application, the anchoring encoding matrix calculation secondary subunit 1313-1 is configured to: perform deep implicit feature fully connected encoding on the real-time wind speed time series associated implicit feature encoding vector to obtain a real-time wind speed time series associated deep implicit encoding vector; perform deep implicit feature fully connected encoding on each historical wind speed local time series associated implicit feature encoding vector in the set of historical wind speed local time series associated implicit feature encoding vectors to obtain a set of historical wind speed local time series associated deep implicit encoding vectors; perform wind speed time series semantic response decision anchoring on the real-time wind speed time series associated deep implicit encoding vector and each historical wind speed local time series associated deep implicit encoding vector in the set of historical wind speed local time series associated deep implicit encoding vectors respectively to obtain a set of real-time wind speed - historical wind speed local time series feature response anchoring encoding matrices.

[0033] Specifically, performing deep implicit feature fully connected encoding on the real-time wind speed time series associated implicit feature encoding vector to obtain a real-time wind speed time series associated deep implicit encoding vector, which is represented by the deep implicit feature fully connected encoding formula as:

[0034]

[0035] Wherein, is the real-time wind speed time series associated implicit feature encoding vector, is matrix multiplication, and are the real-time wind speed weight matrix and the real-time wind speed bias vector respectively, is the activation function, is the real-time wind speed time series associated deep implicit encoding vector. It should be understood that although the real-time wind speed time series associated implicit feature encoding vector has captured the global long-term dependence relationship through the LSTM model, its potential high-order non-linear features and multi-scale interaction patterns still need to be further decoupled. The dense connection characteristic of the fully connected neural network provides a suitable tool for this. By mapping the real-time wind speed time series associated implicit feature encoding vector to a higher-dimensional latent space, the system can break through the linear expression limitation of the original feature space and explicitly enhance the semantic representation ability of the features. This process not only learns the abstract expression of the global statistical law through the weight matrix, but also reveals the hidden complex interaction relationships (such as the coupling effect of wind speed fluctuations in different time windows) in the real-time wind speed sequence with the help of the non-linear activation function (such as sigmoid), thus constructing a more accurate feature basis for the subsequent dynamic semantic alignment with the historical local perturbation features. In this way, through deep implicit encoding, the global features are non-linearly compressed and reconstructed, enabling it to retain the key information of the long-term trend and provide more discriminative feature anchor points for the local feature matching with historical similar scenarios.

[0036] Specifically, perform deep implicit feature fully connected encoding on each historical wind speed local temporal correlation implicit feature encoding vector in the set of historical wind speed local temporal correlation implicit feature encoding vectors to obtain a set of historical wind speed local temporal correlation deep implicit encoding vectors, which is expressed by the formula:

[0037]

[0038]

[0039] Wherein, is the set of historical wind speed local temporal correlation implicit feature encoding vectors, , , and are respectively the 1st, 2nd, th, and th historical wind speed local temporal correlation implicit feature encoding vectors in the set of historical wind speed local temporal correlation implicit feature encoding vectors, and are respectively the historical wind speed weight matrix and the historical wind speed bias vector, is the A set of historical wind speed local temporal correlation depth implicit encoding vectors. It should be understood that the set of the historical wind speed local temporal correlation implicit feature encoding vectors is also processed independently. Different from the single representation of the real-time wind speed temporal correlation implicit feature encoding vector, the set of the historical wind speed local temporal correlation implicit feature encoding vectors contains multiple features that need to be encoded one by one. Specifically, since the historical wind speed contains local correlation patterns in different time periods (such as minute-level fluctuations and hourly trends), and these patterns may show spatial heterogeneity due to geographical environment or meteorological conditions, a single global encoding is difficult to capture the semantic differences of all local features at the same time. In the technical solution of this application, a fully connected network is also used to extract the depth implicit representation of each historical wind speed local temporal correlation implicit feature encoding vector. By processing each historical wind speed local temporal correlation implicit feature encoding vector independently, the system can perform customized mining on the local laws of different time windows or physical positions, avoiding feature confusion caused by global encoding. The dense parameter matrix of the fully connected network provides such a flexible non-linear mapping ability that it can perform high-dimensional projection on each historical wind speed local temporal correlation implicit feature encoding feature and explicitly reconstruct its implicit temporal correlation pattern (such as wind speed gradient change or turbulence intensity fluctuation). In this way, through depth implicit encoding, non-linear compression and semantic enhancement of historical local features are realized, which not only retains short-term perturbation details (such as sudden gust events), but also extracts long-term statistical laws (such as seasonal wind speed distribution), constructs multi-granularity feature anchors for subsequent dynamic matching of global-local features of real-time wind speed, and thus reduces overfitting and improves generalization ability through multi-historical scenario feature learning.

[0040] Specifically, wind speed temporal semantic response decision anchoring is respectively performed on the real-time wind speed temporal correlation depth implicit encoding vector and each historical wind speed local temporal correlation depth implicit encoding vector in the set of the historical wind speed local temporal correlation depth implicit encoding vectors to obtain the set of real-time wind speed-historical wind speed local temporal feature response anchoring encoding matrices, which is expressed by the formula as:

[0041]

[0042] Wherein, is the transposed vector of, is the length of, is the A local temporal feature response anchored encoding matrix of real-time wind speed - historical wind speed. It should be understood that although the global encoding of real-time wind speed has captured the monthly-level wind speed evolution law through the LSTM model, its potential non-linear interaction relationship still needs to be explicitly decoupled; while the local encoding set of historical wind speed contains multi-scale features such as minute-level gusts and hourly-level fluctuations, and a semantic association with real-time features needs to be established through a dynamic anchoring mechanism. Through decision anchoring technology, the global hidden features of the real-time wind speed temporal correlation depth implicit encoding vector are used as the dynamic benchmark, the semantic contribution weights of each local temporal correlation depth implicit encoding vector of historical wind speed are quantified, and a set of real-time wind speed - historical wind speed local temporal feature response anchored encoding matrices with global reference characteristics is constructed. This anchoring mechanism not only explicitly represents the coupling relationship between real-time and historical features, but also provides combinable building block components for subsequent feature fusion, which is suitable for being combined and spliced in a specific way to form a deep implicit representation of a complete decision anchor. Among them, each real-time wind speed - historical wind speed local temporal feature response anchored encoding matrix corresponds to the feature response pattern of a specific time window or meteorological condition, and multi-granularity semantic alignment can be achieved through weight normalization. Here, the set of real-time wind speed - historical wind speed local temporal feature response anchored encoding matrices not only represents the interaction information between the real-time wind speed temporal correlation depth implicit encoding features and the historical wind speed local temporal correlation depth implicit encoding features, but also can serve as a bridge for capturing the global-local semantic alignment relationship.

[0043] Figure 6 The block diagram of the splicing weight factor calculation secondary subunit in the wind power control system based on wind speed prediction according to the embodiment of the present application. As Figure 6 shown, in the embodiment of the present application, the splicing weight factor calculation secondary subunit 1313-2 includes: a splicing factor calculation tertiary subunit 1313-21, configured to determine the decision anchor adaptive splicing factors of each real-time wind speed - historical wind speed local temporal feature response anchored encoding matrix based on the feature distribution of each real-time wind speed - historical wind speed local temporal feature response anchored encoding matrix in the set of real-time wind speed - historical wind speed local temporal feature response anchored encoding matrices to obtain a set of real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factors; a weighting processing tertiary subunit 1313-22, configured to perform weighting processing based on the Softmax function on the set of real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factors to obtain a set of real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing weight factors.

[0044] In an embodiment of the present application, the splicing factor calculation three - level subunit 1313 - 21 includes: obtaining the real - time wind speed - historical wind speed local time - series decision - anchor adaptive splicing factor corresponding to the real - time wind speed - historical wind speed local time - series feature response anchor encoding matrix based on the maximum value, mean value, variance, and difference amplification coefficient of the real - time wind speed - historical wind speed local time - series feature response anchor encoding matrix and the real - time wind speed - historical wind speed drift coefficient; wherein, the difference amplification coefficient is the number of eigenvalues of the real - time wind speed - historical wind speed local time - series feature response anchor encoding matrix; the real - time wind speed - historical wind speed drift coefficient is obtained by performing global smoothing state transition processing on the real - time wind speed - historical wind speed local time - series feature response anchor encoding matrix.

[0045] Specifically, the splicing factor calculation three - level subunit 1313 - 21 is used to determine the decision - anchor adaptive splicing factor of each real - time wind speed - historical wind speed local time - series feature response anchor encoding matrix based on the feature distribution of each real - time wind speed - historical wind speed local time - series feature response anchor encoding matrix in the set of real - time wind speed - historical wind speed local time - series feature response anchor encoding matrices, so as to obtain a set of real - time wind speed - historical wind speed local time - series decision - anchor adaptive splicing factors, which is expressed by the formula:

[0046]

[0047]

[0048]

[0049]

[0050] Wherein, represents the variance of represents the mean value of is for calculating the number of eigenvalues of represents the difference amplification coefficient of represents the maximum value of ， represents the real - time wind speed - historical wind speed drift coefficient, is the intermediate state transition, is the th eigenvalue of is the A local temporal decision anchor adaptive splicing factor for real-time wind speed - historical wind speed. It should be understood that the global encoding of real-time wind speed and the set of local encodings of historical wind speed respectively carry wind speed laws with different time granularities (such as long-term trends and short-term disturbances), but the non-linear coupling relationship between their feature spaces needs to be dynamically aligned through an interpretable weight mechanism. Traditional feature fusion methods are prone to problems such as "long-tail feature drowning" or "local noise amplification" due to the lack of refined modeling of feature contributions. In the technical solution of this application, after generating the real-time wind speed - historical wind speed local temporal feature response anchor encoding matrix, the module needs to further extract and model the feature contributions of each real-time wind speed - historical wind speed local temporal feature response anchor encoding matrix, which is the role of the real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factor. By analyzing the feature distributions of the respective real-time wind speed - historical wind speed local temporal feature response anchor encoding matrices, the real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factors of the respective real-time wind speed - historical wind speed local temporal feature response anchor encoding matrices are determined. These real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factors measure the dominance of local features in a specific task or context and serve as the weight basis for subsequent feature fusion.

[0051] Preferably, for the drift coefficient in the real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factor , for the state transition of the input feature set distribution of the real-time wind speed - historical wind speed local temporal feature response anchor encoding matrix from weak overall interpretability of the mean to strong local interpretability of the maximum value, the global dominance basis of local features is enhanced through the weak-to-strong interpretable generalization of the drift coefficient .

[0052] In particular, is used as the intermediate state transition representation from weak interpretability to strong interpretability. For each eigenvalue of the real-time wind speed - historical wind speed local temporal feature response anchor encoding matrix, it is used as the importance score of the input real-time wind speed - historical wind speed local temporal feature response anchor encoding matrix for the global smooth state transition to perform importance score weight global control on the intermediate state transition relative to the global state transition, so as to achieve the interpretable generalization inference of the weight basis of the real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factor.

[0053] Specifically, the weight processing three-level subunit 1313 - 22 is used to perform weight processing on the set of real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing factors based on the Softmax function to obtain the set of real-time wind speed - historical wind speed local temporal decision anchor adaptive splicing weight factors, which is expressed by the formula:

[0054]

[0055] Among them, is a normalization function, is the th real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing weight factor in the set of real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing weight factors. It should be understood that since the local feature coding sets of real-time wind speed and historical wind speed carry non-linear correlation information of different time scales (such as long-term trends and short-term perturbations), and the dynamic weights of these features need to be adjusted in real time according to the actual working conditions, traditional linear weighting methods are difficult to balance global stability and local sensitivity. In the technical solution of this application, weight processing based on the Softmax function is performed on the set of real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing factors, that is, the Softmax function is used to normalize the sequence of real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing factors into a weight sequence with the property of probability distribution, and the set of real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing factors is transformed into a probability distribution form. In this way, both the normalization constraint that the sum of weight values is 1 is ensured, and the relative importance of key features is highlighted through the non-linear amplification mechanism. This transformation can effectively solve the problem that the contribution of local features at a specific moment (such as during a wind speed mutation period) may be diluted by global statistical features, and at the same time avoid prediction biases caused by uneven weight distribution. Through the probabilistic weight representation, a discriminative decision basis can be provided for subsequent feature fusion, that is, features corresponding to high-probability weight factors (such as historical local patterns highly matching the current working conditions) will be preferentially adopted, while low-probability factors will be suppressed, so as to achieve the adaptive screening and dynamic alignment of global-local features.

[0056] Specifically, the real-time wind speed semantic query feature fusion secondary subunit 1313-3 is used to fuse the set of real-time wind speed-historical wind speed local time-series feature response anchor encoding matrices based on the set of real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing weight factors to obtain the real-time wind speed semantic query dynamic response encoding vector. This process is expressed by the formula:

[0057]

[0058] Among them, is the th real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing weight factor in the set of real-time wind speed-historical wind speed local time-series decision anchor adaptive splicing weight factors, is the real-time wind speed semantic query dynamic response encoding matrix, is a shape reshaping operation, is the dynamic response encoding vector for real-time wind speed semantic query. It should be understood that the real-time wind speed time series provides global dynamic information at the current moment, reflecting the overall trend of the current operating state of the wind farm; while the historical wind speed data contains rich local patterns, such as short-term fluctuations or periodic patterns. Traditional methods are difficult to establish an effective association between global and local features, resulting in increased prediction errors and control lags. In the technical solution of this application, the concepts of decision anchor adaptive splicing weight factor and feature response anchor encoding matrix are introduced to achieve refined processing and dynamic fusion of these features. Specifically, the sets of real-time wind speed-historical wind speed local time series decision anchor adaptive splicing weight factors and the sets of real-time wind speed-historical wind speed local time series feature response anchor encoding matrices are fused through weighted summation.

[0059] In an embodiment of this application, the short-term wind speed prediction unit 132 is used to: input the dynamic response encoding vector for real-time wind speed semantic query into a decoder-based short-term wind speed predictor to obtain the short-term wind speed prediction value. It should be understood that the dynamic response encoding vector for real-time wind speed semantic query integrates complex interaction information of global long-term dependencies (such as monthly wind speed trends) and local perturbation features (such as hourly turbulence events), but its essence still needs to reverse-map the abstract semantic features into quantifiable wind speed prediction values through reverse mapping. The decoder, through its time series modeling ability, transforms the non-linear features extracted by the encoder (such as implicit global-local associations) into accurate predictions of future wind speeds. Its structure (such as a gated recurrent unit or a Transformer) allows for adaptive capture of time series dependencies, avoiding the limitations of traditional linear regression models, and finally outputs a high-precision short-term wind speed prediction value. It is worth mentioning that after wind speed prediction, in order to control power more accurately. The short-term wind speed prediction value is converted into a preliminary power prediction value through a power curve (such as an S-shaped curve). At this time, the predicted power may contain high-frequency jitters or non-linear distortions (such as power jumps caused by sudden changes in wind speed gradients). Therefore, it is necessary to dynamically correct the predicted power obtained based on wind speed prediction. That is to say, smoothing helps to reduce noise interference and enhance the stability and continuity of the signal. Common smoothing techniques include the moving average method, the exponentially weighted moving average method, and the Kalman filter. Through these methods, the system can effectively remove high-frequency noise and retain low-frequency trend components, thereby improving the accuracy of subsequent predictions. Preferably, in a specific embodiment of this application, first, the wind speed prediction value is mapped to a power value by using a wind power curve. Then, by introducing a "weighted smoothing model", the predicted power data is smoothed to prevent sudden fluctuations from affecting the power grid. Among them, the weighted smoothing model formula is expressed as:

[0060]

[0061] Among them, represents the output power at time , represents the predicted power at time , represents the output power at the previous moment , is the smoothing factor, where is dynamically adjusted according to the actual fluctuations. Through smoothing, the data quality and model performance can be significantly improved. That is, smoothing is applied to the preliminary power prediction value to suppress the impact of short-term power mutations on the grid voltage stability.

[0062] In the above-mentioned wind power control system 100 based on wind speed prediction, the limited power strategy determination module 140 is used to calculate the difference between the short-term predicted value of the wind speed and the cut-out wind speed value, and based on the comparison between the difference and a preset threshold, determine whether to adopt the limited power strategy. It should be understood that the instantaneous fluctuations of the wind speed may cause a deviation between the predicted value and the actual operating conditions. If there is no dynamic response mechanism, it may cause severe jitter of the power output or the grid voltage to exceed the limit. The fixed threshold strategy is prone to "over-limiting power" or "insufficient power limiting" because it cannot adapt to the non-linear change of the wind speed. Therefore, this application takes the difference comparison as the core basis for dynamic threshold adjustment. By calculating the real-time deviation between the predicted value and the cut-out wind speed, combined with a preset threshold (such as a dynamic confidence interval based on historical data statistics), it is judged whether the current working condition needs to trigger a hierarchical power limiting strategy. Specifically, if the deviation exceeds the threshold, power limiting protection is started to ensure the stability of the grid voltage; if the deviation is within the threshold, the wind farm is allowed to output at the maximum power, thus avoiding the loss of power generation efficiency caused by conservative power limiting. Then, the smoothed power value is used as the input of the control strategy to avoid mis-triggering power limiting protection due to prediction noise.

[0063] In a specific embodiment, a historical wind speed sequence containing 432 data points (sampling interval of 10 minutes, covering 72 hours) and a real-time wind speed time series (14 data points, interval of 10 minutes) are used as inputs. The LSTM model is used to perform global temporal encoding on the real-time wind speed sequence to extract the implicit feature vector containing long-term dependencies; after the historical wind speed sequence is segmented, the LSTM model is used to generate a set of implicit feature encoding vectors for local temporal correlation of the historical wind speed. Further, through the semantic adaptive query encoding of the decision anchor points of the wind speed temporal features, the decision anchor adaptive splicing weight factor between the real-time features and the historical scenarios is calculated, and the real-time wind speed semantic query dynamic response encoding vector is fused accordingly, and the decoder is used to decode the real-time wind speed semantic query dynamic response encoding vector to obtain the short-term wind speed prediction value. The results show that the mean absolute error (MAE) between the predicted wind speed sequence and the true value of this solution is 0.32 m / s, a 60.9% reduction compared to 0.82 m / s of the traditional LSTM method; the root mean square error (RMSE) is 0.45 m / s, a decrease of 57.1%; the coefficient of determination (R²) reaches 0.982, indicating that the model has strong interpretability. Through the feature response anchoring mechanism and dynamic weight calculation, the system decision-making efficiency is increased by 63%, and the total processing delay is reduced to 85 ms, meeting the real-time requirements of wind power control. The hierarchical power limit strategy reduces the power volatility from 21.4% of the traditional solution to 12.7%, and the number of times the cut-out wind speed trigger level is reduced by 100%; the grid voltage fluctuation amplitude is controlled within ±1.02%, and the total harmonic distortion rate (THD) is 3.1%, significantly better than ±2.35% and 7.8% of the traditional solution. The above data verify the effectiveness of this solution in improving prediction accuracy, shortening control delay, optimizing power generation efficiency, and ensuring grid stability, providing reliable technical support for large-scale wind power grid connection.

[0064] In summary, the wind power control system 100 based on wind speed prediction according to the embodiments of the present application is clarified. It obtains the time series of real-time wind speed values within a predetermined time period collected by the wind speed sensor, and obtains the historical data of the wind speed values. Using deep learning-based data analysis and encoding techniques, it encodes the real-time wind speed time series and the historical subsequence respectively, extracts the global long-term dependencies and local perturbation features, and then dynamically fuses the real-time features and the semantic associations of the historical similar scenarios through the decision anchor points, and then generates a high-precision wind speed prediction value based on the fusion, and based on the real-time deviation analysis between the prediction value and the cut-out wind speed, combines the dynamic threshold adjustment mechanism to trigger the hierarchical power limit strategy. This solution automatically captures the non-linear temporal law of the wind speed through deep learning, dynamically responds to the changes of the actual working conditions, effectively improves the prediction accuracy, shortens the control delay, and at the same time optimizes the power generation efficiency and ensures the grid voltage stability.

[0065] As described above, the wind power control system 100 based on wind speed prediction according to the embodiments of the present application can be implemented in various terminal devices. In one example, the wind power control system 100 based on wind speed prediction can be integrated into the terminal device as a software module and / or a hardware module. For example, the wind power control system 100 based on wind speed prediction can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the wind power control system 100 based on wind speed prediction can also be one of the many hardware modules of the terminal device.

[0066] Alternatively, in another example, the wind power control system 100 based on wind speed prediction and the terminal device can also be separate devices, and the wind power control system 100 based on wind speed prediction can be connected to the terminal device through a wired and / or wireless network, and transmit and interact information according to a predefined data format.

[0067] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation.

[0068] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0069] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware, or in the form of a hardware plus software functional module.

[0070] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0071] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to cover all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0072] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. A plurality of units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The terms such as "second" are used to denote names and do not denote any particular order.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit of the technical solutions of the present invention.

Claims

1. A wind power control system based on wind speed prediction, characterized in that Including: A real-time wind speed value acquisition module for obtaining a time series of real-time wind speed values within a predetermined time period collected by a wind speed sensor; A historical wind speed data acquisition module for obtaining historical data of wind speed values; A query encoding module for performing query encoding on the time series of the real-time wind speed values and the historical data of the wind speed values to obtain a short-term wind speed prediction value. Among them, the query encoding module includes: a real-time and historical wind speed time series encoding unit for performing wind speed semantic query dynamic encoding based on time series correlation features on the time series of the real-time wind speed values and the historical data of the wind speed values to obtain real-time wind speed semantic query dynamic response encoding features; a short-term wind speed prediction unit for obtaining a short-term wind speed prediction value based on the real-time wind speed semantic query dynamic response encoding features; A restricted power strategy determination module for calculating the difference between the short-term wind speed prediction value and the cut-out wind speed value, and determining whether to adopt a restricted power strategy based on the comparison between the difference and a preset threshold; Among them, the real-time and historical wind speed time series encoding unit includes: A wind speed value historical data sequence splitting subunit for splitting the historical data of the wind speed values based on the predetermined time period to obtain a set of historical wind speed value time series subsequences; A wind speed time series encoding subunit for performing time series encoding on the time series of the real-time wind speed values and each historical wind speed value time series subsequence in the set of historical wind speed value time series subsequences to obtain a set of real-time wind speed time series correlation implicit feature encoding vectors and historical wind speed local time series correlation implicit feature encoding vectors; A real-time wind speed semantic query subunit for performing wind speed time series feature decision anchor semantic adaptive query encoding on the set of real-time wind speed time series correlation implicit feature encoding vectors and the set of historical wind speed local time series correlation implicit feature encoding vectors to obtain a real-time wind speed semantic query dynamic response encoding vector as the real-time wind speed semantic query dynamic response encoding feature.

2. The wind power control system based on wind speed prediction according to claim 1, wherein, The wind speed time series encoding subunit is used for: performing time series encoding based on the LSTM model on the time series of the real-time wind speed values and each historical wind speed value time series subsequence in the set of historical wind speed value time series subsequences to obtain the set of real-time wind speed time series correlation implicit feature encoding vectors and the set of historical wind speed local time series correlation implicit feature encoding vectors.

3. The wind power control system based on wind speed prediction according to claim 2, wherein The real-time wind speed semantic query subunit includes: An anchor encoding matrix calculation secondary subunit for calculating an anchor encoding matrix of each historical wind speed local time series correlation implicit feature encoding vector in the set of real-time wind speed time series correlation implicit feature encoding vectors and the set of historical wind speed local time series correlation implicit feature encoding vectors to obtain a set of real-time wind speed - historical wind speed local time series feature response anchor encoding matrices; A splicing weight factor calculation secondary subunit for determining a decision anchor adaptive splicing weight factor of each real-time wind speed - historical wind speed local time series feature response anchor encoding matrix in the set of real-time wind speed - historical wind speed local time series feature response anchor encoding matrices to obtain a set of real-time wind speed - historical wind speed local time series decision anchor adaptive splicing weight factors; The real-time wind speed semantic query feature fusion secondary subunit is used to fuse the set of real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing weight factors, and fuse the set of real-time wind speed-historical wind speed local temporal feature response anchor encoding matrices to obtain the real-time wind speed semantic query dynamic response encoding vector.

4. The wind power control system based on wind speed prediction according to claim 3, characterized in that, The anchor encoding matrix calculation secondary subunit is used for: Performing deep hidden feature fully connected encoding on the real-time wind speed temporal correlation hidden feature encoding vector to obtain a real-time wind speed temporal correlation deep hidden encoding vector; Performing deep hidden feature fully connected encoding on each historical wind speed local temporal correlation hidden feature encoding vector in the set of historical wind speed local temporal correlation hidden feature encoding vectors to obtain a set of historical wind speed local temporal correlation deep hidden encoding vectors; Performing wind speed temporal semantic response decision anchoring on the real-time wind speed temporal correlation deep hidden encoding vector and each historical wind speed local temporal correlation deep hidden encoding vector in the set of historical wind speed local temporal correlation deep hidden encoding vectors respectively to obtain the set of real-time wind speed-historical wind speed local temporal feature response anchor encoding matrices.

5. The wind power control system based on wind speed prediction according to claim 4, characterized in that, The splicing weight factor calculation secondary subunit includes: The splicing factor calculation tertiary subunit is used to determine the decision anchor adaptive splicing factors of the respective real-time wind speed-historical wind speed local temporal feature response anchor encoding matrices based on the feature distributions of the respective real-time wind speed-historical wind speed local temporal feature response anchor encoding matrices in the set of real-time wind speed-historical wind speed local temporal feature response anchor encoding matrices, so as to obtain the set of real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing factors; The weight normalization processing tertiary subunit is used to perform weight normalization processing based on the Softmax function on the set of real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing factors to obtain the set of real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing weight factors.

6. The wind power control system based on wind speed prediction according to claim 5, wherein, The splicing factor calculation tertiary subunit includes: Based on the maximum value, mean value, variance, and difference amplification coefficient of the real-time wind speed-historical wind speed local temporal feature response anchor encoding matrix, as well as the real-time wind speed-historical wind speed drift coefficient, obtaining the real-time wind speed-historical wind speed local temporal decision anchor adaptive splicing factor corresponding to the real-time wind speed-historical wind speed local temporal feature response anchor encoding matrix; Wherein, the difference amplification coefficient is the number of eigenvalues of the real-time wind speed-historical wind speed local temporal feature response anchor encoding matrix; The real-time wind speed-historical wind speed drift coefficient is obtained by performing global smoothing state transition processing on the real-time wind speed-historical wind speed local temporal feature response anchor encoding matrix.

7. The wind power control system based on wind speed prediction according to claim 6, wherein The wind speed short-term prediction unit is used to: input the real-time wind speed semantic query dynamic response encoding vector into a wind speed short-term predictor based on a decoder to obtain the wind speed short-term prediction value.

Images