Segmented prediction method and system for generated power of wind power plant

Abstract

According to the method, real-time wind speed and wind direction data are collected based on a wind power plant multi-mode environment sensing model, historical power generation data are called through a power generation power multi-dimensional sensing analysis model, and future power generation time is divided into different time periods. By analyzing the time sequence and spatial distribution characteristics of real-time and historical data in each time period, the potential incidence relation of the data is mined, the generation power is predicted segment by segment by adopting a calculation model with a dynamic weight adjustment mechanism, and structured integration and formatted output are performed on the predicted data. The system comprises six units including a multi-mode environment data acquisition unit and a historical data storage unit, all the units work cooperatively, the whole process function from data acquisition, analysis processing to prediction result output is achieved, accurate wind power plant generation power prediction data is provided for power grid dispatching, and the accuracy and reliability of power dispatching are improved.

Description

Technical Field

[0001] This invention relates to the field of wind farm power generation prediction, and more particularly to a method and system for segmented prediction of wind farm power generation. Background Technology

[0002] With the ever-increasing global demand for clean energy, the accurate prediction of wind farm power generation, as an important renewable energy source, is crucial for the stable operation and efficient dispatch of the power system. Traditional power supply models rely on conventional energy sources such as thermal power, and the intermittent and fluctuating nature of wind farm power generation presents numerous challenges when integrating it into the grid. Accurate prediction of wind farm power generation can enhance the grid's capacity to absorb wind power, promote the efficient utilization of clean energy, and has profound significance for driving energy structure transformation and achieving "dual-carbon" goals.

[0003] However, existing wind farm power generation prediction technologies have significant shortcomings. On the one hand, most methods only consider a single environmental factor or a small amount of historical data, lacking comprehensive perception and analysis of multimodal environmental information and power generation data. Wind farm power generation is affected by various environmental factors such as wind speed, wind direction, temperature, and air pressure. These factors are interconnected and dynamically changing, making it difficult to capture the changing patterns of power generation under complex environments through analysis of a single factor. On the other hand, existing prediction models often employ fixed calculation patterns, failing to adapt to the dynamic changes in the environment and equipment status during wind farm operation. In actual operation, the performance of wind farm equipment changes over time, and external environmental conditions are constantly changing. Fixed-pattern prediction models struggle to flexibly adjust prediction strategies, resulting in significant deviations between predicted results and actual power generation, thus failing to meet the needs of refined grid dispatching. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a method and system for segmented prediction of wind farm power generation.

[0005] The technical solution utilized in this invention is a segmented prediction method for wind farm power generation, comprising the following steps:

[0006] Step S1: Based on the wind farm multimodal environment perception model, obtain real-time wind speed and wind direction data within the wind farm area. The multimodal environment perception model collects data through multiple wind speed and wind direction sensor arrays distributed within the wind farm. Each sensor array includes no less than three sensors at different heights and positions.

[0007] Step S2: Using the multi-dimensional perception analysis model of power generation, retrieve the historical power generation data of the wind farm. The historical power generation data includes hourly power generation records for at least one full calendar year.

[0008] Step S3: Divide the future power generation time of the wind farm into multiple time periods according to a preset time interval. The time interval ranges from 15 minutes to 1 hour, and the different time periods are seamlessly connected.

[0009] Step S4: For each time period, combine the real-time wind speed and wind direction change data and historical power generation data corresponding to that time period, and use analysis strategies based on time series and spatial distribution characteristics to analyze the potential correlation between the data;

[0010] Step S5: Based on the correlation obtained from the analysis, the power generation of different time periods is predicted segment by segment using a calculation model with a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism adjusts the weights according to the degree of matching between real-time data and historical data.

[0011] Step S6: The predicted power generation data for different time periods are structured, integrated, and formatted to generate a data format output suitable for the power grid dispatching system. The output data includes the start time of different time periods, the predicted power generation value, and the data confidence index. The data confidence index is determined based on the number of iterations of the calculation model and the degree of data convergence.

[0012] Furthermore, in step S4, when analyzing the potential correlations between data, a multimodal environment perception fusion model is introduced. This model calculates the correlation degree of environmental data using the following formula:

[0013]

[0014] Among them, R ij V represents the correlation between the i-th set of real-time environmental data and the j-th set of historical environmental data; i V j These are real-time and historical wind speed values, respectively; D i D j These are real-time and historical wind direction values, respectively; P i P j These represent the real-time and historical power generation at the corresponding time points, respectively; α1, α2, and α3 are weighting coefficients determined based on the characteristics of wind farm equipment and statistical analysis of historical data, and satisfy α1+α2+α3=1.

[0015] Furthermore, in step S5, the computational model with a dynamic weight adjustment mechanism utilizes the power prediction optimization formula:

[0016]

[0017] Among them, P p β represents the power generation during the predicted period; n represents the number of historical data samples used in the calculation; β kP represents the dynamic weight of the k-th historical data sample, and its value is dynamically adjusted based on the similarity between the real-time data and the k-th historical data sample. h，k P represents the power generation of the k-th historical data sample. r，k γ1 represents the corresponding power generation calculated based on real-time data simulation; γ2 and γ1 are correction coefficients determined according to the operating status of wind farm equipment and environmental conditions, and γ1+γ2=1.

[0018] Furthermore, in step S1, the wind farm multimodal environmental perception model utilizes a spatial interpolation compensation algorithm during data acquisition. Specifically, for any two adjacent sensors, the wind speed V... c Wind direction D c Calculated using the following formula:

[0019]

[0020] Where V1 and V2 are the wind speed measurements of two adjacent sensors, D1 and D2 are the wind direction measurements of two adjacent sensors, and d1 and d2 are the distances from the point to be calculated to the two adjacent sensors.

[0021] Furthermore, in step S2, when the multi-dimensional power generation perception and analysis model extracts features from historical power generation data, it uses the time-power feature mapping formula:

[0022]

[0023] Among them, F t，p Let ΔP be the characteristic value of power change within the time interval Δt; t ω represents the change in power generation within a time interval Δt; t The time weighting coefficient is set according to the characteristics of electricity demand in different time periods.

[0024] Furthermore, in step S3, when dividing future power generation periods, a dynamic time period division formula is used:

[0025]

[0026] Among them, T s T represents the adjusted time period length. t The initial time period length is denoted by m; m is the number of sampling points for real-time wind speed data; V i Let be the real-time wind speed at the i-th sampling point; This represents the average real-time wind speed.

[0027] Furthermore, step S3 includes the following sub-steps:

[0028] Step S31: Determine the initial time period division scheme. Based on the regular needs of power grid dispatch and the fluctuation characteristics of historical data, set the initial time interval and number of time periods.

[0029] Step S32: Collect real-time wind speed data and obtain the real-time wind speed sequence for the current moment and a certain future time period through the wind farm multimodal environment perception model;

[0030] Step S33: Calculate the wind speed fluctuation coefficient. Calculate the standard deviation and coefficient of variation of wind speed based on the real-time wind speed sequence to assess the degree of wind speed fluctuation.

[0031] Step S34: Adjust the time period division. Dynamically adjust the initial time period division scheme according to the wind speed fluctuation coefficient. When the wind speed fluctuation exceeds the preset value, shorten the time period length. When the wind speed fluctuation is lower than the preset value, extend the time period length.

[0032] Furthermore, step S4 includes the following sub-steps:

[0033] Step S41: Construct a data feature matrix by arranging real-time wind speed and wind direction data and historical power generation data according to time order and parameter category to form a multi-dimensional data feature matrix;

[0034] Step S42: Perform data normalization processing, normalize the different parameters in the data feature matrix to make the data fall within the same order of magnitude range;

[0035] Step S43: Using the sliding window algorithm, a fixed-size window is slid across the data feature matrix to extract combinations of data features within different time windows;

[0036] Step S44: Calculate feature similarity. Use the cosine similarity algorithm to calculate the similarity between data feature combinations and other combinations within different time windows, and identify similar data patterns.

[0037] Furthermore, step S5 includes the following sub-steps:

[0038] Step S51: Determine the basic prediction model. Based on the equipment type and historical data characteristics of the wind farm, select a basic calculation model that meets the preset benchmark as the prediction framework.

[0039] Step S52: Initialize the weight parameters by assigning initial weight values ​​to different input parameters in the basic prediction model. These weight values ​​are set based on the statistical analysis results of historical data.

[0040] Step S53: Perform model training by inputting real-time and historical data into the basic prediction model and adjusting the weight parameters through iterative calculation to minimize the error between the model output and the actual power generation.

[0041] Step S54: Output prediction results. The trained model predicts power generation at different time periods and outputs the prediction results and related confidence information.

[0042] A segmented prediction system for wind farm power generation, comprising:

[0043] The multimodal environmental data acquisition unit is deployed in the wind farm area to collect real-time wind speed and direction data through multiple wind speed and direction sensor arrays and transmit the data to the data processing unit.

[0044] The historical data storage unit is used to store hourly power generation records of the wind farm for at least one full calendar year in the past, and can interact with the data processing unit.

[0045] The time period segmentation unit, connected to the multimodal environment data acquisition unit and the historical data storage unit, is used to divide the future power generation time of the wind farm into multiple time periods according to preset rules;

[0046] The data correlation analysis unit is connected to the multimodal environment data acquisition unit, the historical data storage unit, and the time period segmentation unit, respectively, and is used to analyze the potential correlation between real-time data and historical data in different time periods;

[0047] The power prediction unit, connected to the data correlation analysis unit, is used to predict the power generation of different time periods segment by segment based on the correlation obtained by analysis and using a calculation model with a dynamic weight adjustment mechanism.

[0048] The data output unit, connected to the power prediction unit, is used to output the predicted power generation data for different time periods to the power grid dispatch system after structured integration and formatting.

[0049] Beneficial Effects: This invention proposes a segmented prediction method and system for wind farm power generation. This method and system construct a multimodal environmental perception model, comprehensively collecting real-time wind speed, wind direction, and other multi-dimensional environmental data through multiple sensor arrays distributed within the wind farm. This overcomes the limitations of traditional methods that rely solely on a single environmental factor, achieving three-dimensional perception of complex environmental information. Simultaneously, the system utilizes a multi-dimensional power generation perception analysis model to deeply mine historical power generation data, revealing potential correlations between data in both time and space dimensions, providing richer and more accurate data for prediction. In terms of prediction strategy, the system adopts a segmented prediction mechanism, dynamically adjusting the time period division based on real-time wind speed fluctuations, avoiding the problem of fixed time period divisions failing to adapt to environmental changes. Through a dynamic weight adjustment mechanism, weights are flexibly allocated based on the matching degree between real-time and historical data, enabling the model to adapt to the dynamic changes in wind farm equipment status and environmental conditions, significantly improving the adaptability and accuracy of prediction. Furthermore, the system constructs a multi-dimensional prediction model through data correlation analysis and feature mining, deeply integrating environmental data with power generation, effectively capturing the changing patterns of power generation under complex environments. In terms of data processing, the system performs structured integration and formatting of the prediction results, outputting standardized data including confidence level indicators, providing a more reliable and valuable decision-making basis for power grid dispatch. This invention, through comprehensive perception, in-depth analysis, dynamic adjustment, and standardized output, successfully overcomes the problems of incomplete consideration of environmental factors and poor model adaptability in traditional prediction technologies. It provides an accurate, flexible, and practical solution for wind farm power generation prediction, powerfully promoting the efficient coordinated operation of wind power and the power grid. Attached Figure Description

[0050] Figure 1 This is a flowchart of the method steps of the present invention;

[0051] Figure 2 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] like Figure 1 As shown, the method for segmented prediction of wind farm power generation includes the following steps:

[0054] Step S1: Based on the wind farm multimodal environment perception model, obtain real-time wind speed and wind direction data within the wind farm area. The multimodal environment perception model collects data through multiple wind speed and wind direction sensor arrays distributed within the wind farm. Each sensor array includes no less than three sensors at different heights and positions.

[0055] Specifically, step S1 mainly revolves around the multimodal environmental perception model of the wind farm, aiming to acquire accurate real-time wind speed and direction data within the wind farm area. This step achieves data acquisition through multiple wind speed and direction sensor arrays deployed within the wind farm. Each sensor array includes at least three sensors at different heights and locations. This multi-sensor layout design is intended to capture environmental information from multiple dimensions. Sensors at different heights can collect wind speed and direction data from different atmospheric layers, while sensors at different locations can cover different areas of the wind farm, thus avoiding data bias caused by the location limitations of a single sensor. The data acquired through this multimodal environmental perception model can more comprehensively and realistically reflect the current actual environmental conditions of the wind farm, providing an accurate environmental data foundation for subsequent power generation prediction.

[0056] In practical implementation, these sensor arrays are installed within the wind farm according to a pre-planned layout. During installation, factors such as the wind farm's topography and turbine distribution are fully considered to ensure the sensors acquire representative environmental data. After installation, the real-time wind speed and direction data are transmitted to the data processing center via wired or wireless communication. During data transmission, appropriate encryption and verification technologies are employed to ensure the accuracy and integrity of the data, preventing loss or tampering, thus providing reliable support for subsequent analysis and prediction based on this data.

[0057] Step S2: Using the multi-dimensional perception analysis model of power generation, retrieve the historical power generation data of the wind farm. The historical power generation data includes hourly power generation records for at least one full calendar year.

[0058] Specifically, step S2 utilizes a multi-dimensional power generation perception analysis model to retrieve historical power generation data from the wind farm. This historical data covers hourly power generation records for at least one full calendar year. Collecting such a long-term span and fine-grained historical data is crucial because wind farm power generation is affected by various cyclical factors such as seasons and weather. Data from a full calendar year comprehensively reflects these cyclical patterns. Analyzing historical power generation data allows us to understand the wind farm's power generation capacity under different seasons and weather conditions, uncovering hidden trends and patterns in power generation. These patterns are vital for predicting future power generation, as future power generation often follows certain historical patterns, and historical data provides important reference for prediction.

[0059] In terms of implementation, wind farm power generation data is recorded in real time and stored in a dedicated database. The multi-dimensional power generation sensing and analysis model connects to the database through a specific data interface and retrieves the required historical power generation data according to preset rules. During data retrieval, the data undergoes preliminary screening and processing to remove abnormal and invalid data, ensuring the accuracy and reliability of the obtained data. Simultaneously, to improve data retrieval efficiency, the database employs a rational indexing and storage structure design, enabling the multi-dimensional power generation sensing and analysis model to quickly and accurately obtain hourly power generation records for at least one complete calendar year, saving time for subsequent data processing and analysis and improving the efficiency of the entire prediction process.

[0060] Step S3: Divide the future power generation time of the wind farm into multiple time periods according to a preset time interval. The time interval ranges from 15 minutes to 1 hour, and the different time periods are seamlessly connected.

[0061] Specifically, step S3 divides the future power generation time of the wind farm into multiple time periods according to a preset time interval, with the time interval ranging from 15 minutes to 1 hour, and the time periods are seamlessly connected. This division of time periods is to finely segment the future power generation time in order to more accurately predict the power generation in each time period. Different time intervals are suitable for different prediction needs and environmental conditions. Shorter time intervals (such as 15 minutes) can capture rapid changes in environmental factors such as wind speed and wind direction more promptly, and are suitable for situations with large wind speed fluctuations; longer time intervals (such as 1 hour) are suitable for relatively stable environmental conditions, which can reduce the amount of calculation required for prediction. The seamless connection of each time period ensures comprehensive coverage of the future power generation time, with no prediction blind spots, thereby providing continuous and complete power generation prediction information for grid dispatch.

[0062] In implementing this step, an initial time interval is first determined based on grid dispatch requirements and the fluctuation characteristics of historical data. Then, the system monitors the wind farm's environmental conditions in real time, particularly wind speed changes. When significant wind speed fluctuations are detected, the time interval is automatically shortened, increasing the prediction frequency to more promptly reflect changes in power generation. Conversely, when wind speeds are relatively stable, the time interval is appropriately extended to reduce computational resource consumption. This dynamic adjustment of the time interval allows the time period division to better adapt to the actual operating conditions of the wind farm, providing a more suitable time dimension for subsequent power generation prediction and improving the accuracy and practicality of the predictions.

[0063] Step S4: For each time period, combine the real-time wind speed and wind direction change data and historical power generation data corresponding to that time period, and use analysis strategies based on time series and spatial distribution characteristics to analyze the potential correlation between the data;

[0064] Specifically, step S4, for each time period, combines real-time wind speed and direction change data with historical power generation data, employing an analysis strategy based on time series and spatial distribution characteristics to uncover potential correlations between the data. There are complex nonlinear relationships between wind farm power generation and real-time environmental and historical power generation data; these relationships are hidden within the time series variations and spatial distribution differences of the data. Analyzing the time series characteristics reveals the trends in power generation and environmental data over time, such as the patterns of change at different times of the day. Analyzing the spatial distribution characteristics allows us to consider the impact of environmental differences at different locations within the wind farm on power generation. This comprehensive analysis strategy can uncover subtle but significant correlations that influence power generation, providing deeper data support for accurate power generation prediction.

[0065] In the implementation process, the first step is to integrate real-time wind speed and direction data for each time period with historical power generation data to construct a dataset containing multiple parameters. Then, specialized data analysis algorithms are used to process this dataset, extracting data features from both temporal and spatial dimensions. In the temporal dimension, the changing trends and periodic patterns of the data at different points in time are analyzed; in the spatial dimension, the differences in data collected by sensors at different locations and their impact on power generation are considered. By analyzing and comparing these features, potential correlation patterns between the data are identified, such as the correspondence between specific wind speed and direction combinations and power generation. These uncovered correlations will serve as an important basis for subsequent power generation prediction, helping to improve the accuracy and reliability of the prediction model.

[0066] Step S5: Based on the correlation obtained from the analysis, the power generation of different time periods is predicted segment by segment using a calculation model with a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism adjusts the weights according to the degree of matching between real-time data and historical data.

[0067] Specifically, step S5, based on the correlations mined in step S4, uses a computational model with a dynamic weight adjustment mechanism to predict the power generation for each time period segment by segment. Since the operating environment of a wind farm is dynamically changing, the correlation between real-time data and historical data varies at different times. Therefore, a computational model that can flexibly adjust according to data changes is needed. A computational model with a dynamic weight adjustment mechanism can automatically adjust the weights of various data factors in the prediction process based on the degree of matching between real-time and historical data. When real-time data has a high similarity to a set of historical data, a higher weight is assigned to that set of historical data, allowing it to play a greater role in the prediction; conversely, when the similarity is low, its weight is reduced to avoid inaccurate historical data having an excessive impact on the prediction results. This dynamic weight adjustment method enables the prediction model to better adapt to different environmental conditions, improving the accuracy and reliability of the prediction.

[0068] In implementing this step, a suitable basic calculation model is first selected as the prediction framework based on factors such as the type of equipment in the wind farm and the characteristics of historical data. Then, the correlations and related data mined in step S4 are input into this model. During model operation, the matching degree between real-time data and historical data is monitored in real time, and the weight parameters in the model are dynamically adjusted according to a preset algorithm. Through continuous iterative calculations, the predicted results output by the model are made as close as possible to the actual power generation. After each prediction is completed, the model's prediction results are evaluated and verified. Based on the evaluation results, the model parameters and weight adjustment strategies are further optimized to ensure that the model can continuously provide accurate power generation prediction results in subsequent prediction processes.

[0069] Step S6: The predicted power generation data for different time periods are structured, integrated, and formatted to generate a data format output suitable for the power grid dispatching system. The output data includes the start time of different time periods, the predicted power generation value, and the data confidence index. The data confidence index is determined based on the number of iterations of the calculation model and the degree of data convergence.

[0070] Specifically, step S6 involves structuring and formatting the predicted power generation data for each time period to generate a data format output suitable for the power grid dispatching system. The power grid dispatching system has specific formats and requirements for input data; only by structuring and formatting the predicted data can it be accurately received and used by the system. During structuring, key information such as the start time and predicted power generation values ​​for each time period are arranged and organized in an orderly manner, forming a clear and standardized data structure. In terms of formatting, the data is encoded and converted according to the data format standards specified by the power grid dispatching system to meet the system's interface requirements. Simultaneously, the output data also includes a data confidence indicator, determined based on the number of iterations of the calculation model and the degree of data convergence. This indicator reflects the reliability of the prediction results, providing a reference for power grid dispatchers and enabling them to comprehensively consider the accuracy of the prediction results when formulating dispatching plans.

[0071] In practical implementation, a dedicated data processing module is developed to handle structured integration and formatting. This module interacts with the power prediction unit to obtain predicted power generation data for each time period. Then, according to preset data processing rules, the data is structured and integrated, organizing relevant information into specific data tables or data objects. Next, based on the data format requirements of the power grid dispatching system, the structured data is formatted, for example, by converting it to standard data formats such as XML and JSON, and performing necessary encoding operations. When determining the data confidence level, the corresponding indicator value is calculated using a specific algorithm based on the number of iterations of the calculation model during the prediction process and the data convergence status. Finally, the processed data is output to the power grid dispatching system using the prescribed communication protocol and interface, achieving effective integration with the power grid dispatching system and providing strong support for the stable operation of the power grid and the rational allocation of power resources.

[0072] Preferably, in step S4, when analyzing the potential correlations between data, a multimodal environment perception fusion model is introduced. This model calculates the correlation degree of environmental data using the following formula:

[0073]

[0074] Among them, R ij V represents the correlation between the i-th set of real-time environmental data and the j-th set of historical environmental data; i V j These are real-time and historical wind speed values, respectively; D i D j These are real-time and historical wind direction values, respectively; P i P jThese represent the real-time and historical power generation at the corresponding time points, respectively; α1, α2, and α3 are weighting coefficients determined based on the characteristics of wind farm equipment and statistical analysis of historical data, and satisfy α1+α2+α3=1.

[0075] Specifically, step S4 introduces a multimodal environmental perception fusion model, which mines the potential relationship between real-time and historical data by calculating the correlation degree of environmental data. In wind farm power generation prediction, the correlation degree of environmental data is a key indicator for evaluating the degree of matching between current environmental conditions and similar historical scenarios. The model calculates the correlation degree through three dimensions of parameters: wind speed ratio, wind direction difference, and power ratio. The wind speed ratio reflects the relative magnitude of real-time wind speed and historical wind speed, the wind direction difference reflects the angular difference of wind direction change, and the power ratio is directly related to the change in power generation. The weight coefficients of these three parameters are determined based on the characteristics of wind farm equipment and statistical analysis of historical data to ensure accurate capture of the correlation between data under different wind farm environments. The correlation degree value calculated by this model can help the system quickly identify historical scenarios similar to the current environmental conditions, thereby providing a more reliable reference for prediction. In the implementation process, the system first preprocesses the real-time collected wind speed and wind direction data, and then compares them group by group with the environmental data in the historical database to calculate the correlation degree of each group of data. Based on the correlation value, the system filters out historical data with high correlation for subsequent power generation prediction and analysis, effectively improving the accuracy and relevance of the prediction.

[0076] Preferably, in step S5, the computational model with a dynamic weight adjustment mechanism utilizes the power prediction optimization formula:

[0077]

[0078] Among them, P p β represents the power generation during the predicted period; n represents the number of historical data samples used in the calculation; β k P represents the dynamic weight of the k-th historical data sample, and its value is dynamically adjusted based on the similarity between the real-time data and the k-th historical data sample. h，k P represents the power generation of the k-th historical data sample. r，k γ1 represents the corresponding power generation calculated based on real-time data simulation; γ2 and γ1 are correction coefficients determined according to the operating status of wind farm equipment and environmental conditions, and γ1+γ2=1.

[0079] Specifically, the power prediction optimization formula proposed in step S5 employs a dynamic weight adjustment mechanism, which adaptively adjusts prediction parameters based on the similarity between real-time and historical data, thereby significantly improving prediction accuracy. In actual wind farm operation, environmental conditions and equipment status are constantly changing, making it difficult for traditional prediction models to adapt to such dynamic changes. This formula introduces dynamic weight coefficients, assigning weight values ​​corresponding to the similarity of each historical data sample. Historical samples with higher similarity have a larger weight in the prediction calculation, and vice versa. Simultaneously, the formula also considers the power generation calculated based on real-time data simulation, using correction coefficients to comprehensively balance historical and real-time simulation data. This dual adjustment mechanism enables the prediction model to better adapt to the dynamic characteristics of wind farms. During implementation, the system first determines the initial weight values ​​based on the statistical characteristics of historical data, and then dynamically adjusts them continuously based on the comparison results between real-time and historical data during the prediction process. Through iterative calculation, the error between the predicted power output and the actual power generation gradually decreases, ultimately yielding a high-precision prediction result. The application of this formula effectively solves the problem of poor adaptability of traditional prediction models when facing complex and ever-changing wind farm environments.

[0080] Preferably, in step S1, the wind farm multimodal environment perception model utilizes a spatial interpolation compensation algorithm during data acquisition. Specifically, for any two adjacent sensors, the wind speed V... c Wind direction D c Calculated using the following formula:

[0081]

[0082] Where V1 and V2 are the wind speed measurements of two adjacent sensors, D1 and D2 are the wind direction measurements of two adjacent sensors, and d1 and d2 are the distances from the point to be calculated to the two adjacent sensors.

[0083] Specifically, the spatial interpolation compensation algorithm proposed in step S1 is a key technology for improving the accuracy of wind farm environmental data acquisition. In actual wind farms, due to the complexity of terrain and the limitations of sensor deployment, data acquisition blind spots may exist between sensors. This algorithm can accurately estimate the wind speed and direction data in these blind spots by weighted interpolation of adjacent sensor data. The core idea of ​​the algorithm is based on the inverse distance weighting principle, that is, the wind speed and direction values ​​of the point to be calculated are inversely proportional to the measurement values ​​of adjacent sensors, and the closer the distance, the greater the influence. This method fully considers the influence of spatial location on environmental data and can more accurately reflect the distribution of environmental parameters within the wind farm. In the implementation process, the system first constructs the spatial coordinate system of the wind farm and determines the position coordinates of each sensor. Then, for any point where data needs to be estimated, the system automatically identifies its adjacent sensors and obtains the measurement values ​​and position information of these sensors. The distance from the point to be calculated to each adjacent sensor is calculated according to the distance formula, and then the wind speed and direction values ​​of that point are calculated using a weighted average method. In this way, the system can obtain more comprehensive and accurate environmental data within the wind farm, providing a solid data foundation for subsequent power generation prediction and effectively making up for the data loss problem caused by insufficient sensor deployment.

[0084] Preferably, in step S2, when the multi-dimensional power generation perception and analysis model extracts features from historical power generation data, it uses the time-power feature mapping formula:

[0085]

[0086] Among them, F t，p Let ΔP be the characteristic value of power change within the time interval Δt; t ω represents the change in power generation within a time interval Δt; t The time weighting coefficient is set according to the characteristics of electricity demand in different time periods.

[0087] Specifically, the time-power feature mapping formula proposed in step S2 is an important tool for in-depth mining of historical power generation data. During wind farm power generation, the change in power output over time exhibits complex nonlinear characteristics. These characteristics contain rich information and are crucial for predicting future power output. This formula, by calculating the power change characteristic value per unit time and combining it with a time weighting coefficient, can accurately capture the pattern of power output change over time. The time weighting coefficient is set considering the characteristics of electricity demand in different time periods. For example, during peak electricity consumption periods, the change in power output has a greater impact on power dispatch, thus it is given a higher weight. In this way, the system can focus more on the power change characteristics of time periods that have a significant impact on power dispatch. In the implementation process, the system first preprocesses the historical power generation data, arranging it chronologically and dividing it into appropriate time intervals. Then, for each time interval, the power change is calculated and combined with the time weighting coefficient to obtain the feature mapping value. These feature mapping values ​​constitute a feature vector reflecting the time-varying pattern of power output. By analyzing this feature vector, the system can identify important characteristics such as periodic and trend changes in power output, providing more valuable input information for subsequent prediction models and improving the accuracy and reliability of predictions.

[0088] Preferably, in step S3, when dividing future power generation periods, a dynamic time period division formula is used:

[0089]

[0090] Among them, T s T represents the adjusted time period length. t The initial time period length is denoted by m; m is the number of sampling points for real-time wind speed data; V i Let be the real-time wind speed at the i-th sampling point; This represents the average real-time wind speed.

[0091] Specifically, the dynamic time-segmentation formula proposed in step S3 addresses the limitations of traditional methods that use fixed time intervals, which are ill-suited to the dynamic characteristics of wind speed fluctuations in wind farms. This new formula dynamically adjusts the time-segment length by calculating the real-time wind speed fluctuation coefficient, resulting in a more rational division of time periods. When wind speed fluctuations are significant, the time-segment length is shortened, increasing the prediction frequency to more promptly capture the impact of wind speed changes on power generation. Conversely, when wind speeds are relatively stable, the time-segment length is extended, reducing computational load and improving prediction efficiency. This dynamic adjustment mechanism allows the prediction system to better adapt to the actual operating conditions of wind farms. In implementation, the system first sets an initial time-segment length, then collects wind speed data in real time and calculates its fluctuation coefficient. The fluctuation coefficient is calculated based on the standard deviation and coefficient of variation of wind speed, accurately reflecting the degree of wind speed fluctuation. Based on the calculated fluctuation coefficient, the system automatically adjusts the time-segment length according to the formula and redivides future power generation time. Through this dynamic division method, the system can optimize the use of computational resources while ensuring prediction accuracy, improving the efficiency and adaptability of the entire prediction process and providing a more reasonable and accurate prediction time-segmentation scheme for grid dispatch.

[0092] Preferably, step S3 includes the following sub-steps:

[0093] Step S31: Determine the initial time period division scheme. Based on the regular needs of power grid dispatch and the fluctuation characteristics of historical data, set the initial time interval and number of time periods.

[0094] Step S32: Collect real-time wind speed data and obtain the real-time wind speed sequence for the current moment and a certain future time period through the wind farm multimodal environment perception model;

[0095] Step S33: Calculate the wind speed fluctuation coefficient. Calculate the standard deviation and coefficient of variation of wind speed based on the real-time wind speed sequence to assess the degree of wind speed fluctuation.

[0096] Step S34: Adjust the time period division. Dynamically adjust the initial time period division scheme according to the wind speed fluctuation coefficient. When the wind speed fluctuation exceeds the preset value, shorten the time period length. When the wind speed fluctuation is lower than the preset value, extend the time period length.

[0097] Specifically, step S3 was refined, proposing a four-step time-segmentation method. This method forms a complete time-segmentation process from initial scheme determination to final dynamic adjustment. When determining the initial time-segmentation scheme, the system comprehensively considers the routine needs of power grid dispatching and the fluctuation characteristics of historical data, setting a basic time interval and number of time periods to provide a benchmark for subsequent dynamic adjustments. Collecting real-time wind speed data is the foundation for subsequent analysis; the wind speed sequence obtained through the wind farm's multimodal environmental perception model reflects the actual operating status of the wind farm. Calculating the wind speed fluctuation coefficient is a key step in this method. Through the calculation of standard deviation and coefficient of variation, the system can accurately assess the degree of wind speed fluctuation, providing a quantitative basis for time-segmentation adjustments. Finally, the initial time-segmentation scheme is dynamically adjusted based on the wind speed fluctuation coefficient. When fluctuations are large, the time period length is shortened; when fluctuations are small, the time period length is extended, making the time-segmentation more consistent with actual wind speed changes. This step-by-step implementation makes the time-segmentation process more scientific and reasonable, dynamically optimizing predicted time periods based on real-time environmental changes, improving the accuracy and adaptability of predictions, and providing a more precise time reference for power grid dispatching.

[0098] Preferably, step S4 includes the following sub-steps:

[0099] Step S41: Construct a data feature matrix by arranging real-time wind speed and wind direction data and historical power generation data according to time order and parameter category to form a multi-dimensional data feature matrix;

[0100] Step S42: Perform data normalization processing, normalize the different parameters in the data feature matrix to make the data fall within the same order of magnitude range;

[0101] Step S43: Using the sliding window algorithm, a fixed-size window is slid across the data feature matrix to extract combinations of data features within different time windows;

[0102] Step S44: Calculate feature similarity. Use the cosine similarity algorithm to calculate the similarity between data feature combinations and other combinations within different time windows, and identify similar data patterns.

[0103] Specifically, step S4 was refined, proposing a data correlation analysis method comprising four sub-steps. This method forms a complete data mining process, from data matrix construction to similar pattern recognition. Constructing the data feature matrix is ​​a crucial step in integrating real-time and historical data. By arranging data according to time sequence and parameter categories, a multi-dimensional feature matrix is ​​formed, providing a structured data foundation for subsequent analysis. Data normalization eliminates the impact of differences in the order of magnitude of different parameters on the analysis results, ensuring that each parameter has equal importance in the analysis process and improving the accuracy of the analysis. The application of the sliding window algorithm can extract data feature combinations within different time windows, capturing local changes in the data and helping to discover short-term correlation patterns. Calculating feature similarity is the core of this method. Through the cosine similarity algorithm, the system can quickly and accurately identify similar data patterns and uncover potential correlations hidden in the data. This sub-step analysis method comprehensively considers the time series and spatial distribution characteristics of the data, enabling in-depth mining of complex correlations between data and providing stronger support for power generation prediction.

[0104] Preferably, step S5 includes the following sub-steps:

[0105] Step S51: Determine the basic prediction model. Based on the equipment type and historical data characteristics of the wind farm, select a basic calculation model that meets the preset benchmark as the prediction framework.

[0106] Step S52: Initialize the weight parameters by assigning initial weight values ​​to different input parameters in the basic prediction model. These weight values ​​are set based on the statistical analysis results of historical data.

[0107] Step S53: Perform model training by inputting real-time and historical data into the basic prediction model and adjusting the weight parameters through iterative calculation to minimize the error between the model output and the actual power generation.

[0108] Step S54: Output prediction results. The trained model predicts power generation at different time periods and outputs the prediction results and related confidence information.

[0109] Specifically, step S5 was refined, proposing a power prediction method comprising four sub-steps. This method forms a complete prediction process from basic model selection to final result output. Determining the basic prediction model is the starting point of the entire prediction process. Based on the equipment type and historical data characteristics of the wind farm, the system selects the most suitable calculation model as the prediction framework, ensuring that the model accurately reflects the power generation characteristics of the wind farm. Initializing weight parameters provides the initial calculation basis for the model. These weight values ​​are set based on the statistical analysis results of historical data, enabling the model to have good prediction performance in the initial stage. Model training is the core of this method. By inputting real-time and historical data into the model and performing iterative calculations, the weight parameters are continuously adjusted to minimize the error between the model output and the actual power generation. During training, the system determines whether to stop iteration based on preset convergence conditions to ensure the model achieves optimal prediction results. Finally, the fully trained model predicts power generation for each time period and outputs the prediction results and related confidence information, providing accurate and reliable decision-making basis for grid dispatch. This step-by-step prediction method fully considers the dynamic changes in the wind farm environment and the adaptability of the model, and can provide high-precision power generation prediction results.

[0110] like Figure 2 As shown, a segmented prediction system for wind farm power generation includes:

[0111] The multimodal environmental data acquisition unit is deployed in the wind farm area to collect real-time wind speed and direction data through multiple wind speed and direction sensor arrays and transmit the data to the data processing unit.

[0112] The historical data storage unit is used to store hourly power generation records of the wind farm for at least one full calendar year in the past, and can interact with the data processing unit.

[0113] The time period segmentation unit, connected to the multimodal environment data acquisition unit and the historical data storage unit, is used to divide the future power generation time of the wind farm into multiple time periods according to preset rules;

[0114] The data correlation analysis unit is connected to the multimodal environment data acquisition unit, the historical data storage unit, and the time period segmentation unit, respectively, and is used to analyze the potential correlation between real-time data and historical data in different time periods;

[0115] The power prediction unit, connected to the data correlation analysis unit, is used to predict the power generation of different time periods segment by segment based on the correlation obtained by analysis and using a calculation model with a dynamic weight adjustment mechanism.

[0116] The data output unit, connected to the power prediction unit, is used to output the predicted power generation data for different time periods to the power grid dispatch system after structured integration and formatting.

[0117] The proposed method and system for segmented prediction of wind farm power generation utilizes a multimodal environmental perception model. It deploys multiple sensor arrays at different heights and locations to comprehensively collect real-time wind speed and direction data, achieving a three-dimensional perception of the complex wind farm environment. Simultaneously, the multi-dimensional power generation perception and analysis model retrieves at least one year's worth of hourly historical power generation data, deeply mining the data from multiple temporal and spatial dimensions. This overcomes the limitations of traditional methods that rely on only a limited amount of data, providing a richer and more comprehensive data foundation for prediction.

[0118] To address the challenge of poor adaptability in traditional forecasting models, this method and system employ a dynamic and flexible forecasting strategy. On one hand, the time period division is dynamically adjusted based on real-time wind speed fluctuations; the time period is shortened when wind speed fluctuations are large and extended when they are small, ensuring relatively stable environmental conditions within each forecast period and improving the accuracy of the forecasts. On the other hand, a computational model with a dynamic weight adjustment mechanism is used to flexibly allocate weights based on the degree of matching between real-time and historical data. This allows the model to adaptively adjust as wind farm equipment performance changes and environmental conditions evolve, effectively avoiding the problem of large forecasting deviations under fixed models.

[0119] Furthermore, the system enhances predictive accuracy through close collaboration across multiple stages. In the data processing stage, the collected multi-source data is structured, integrated, and formatted to output standardized data including start times for each period, predicted power generation values, and confidence level indicators, providing clear and reliable decision-making support for grid dispatch. During data correlation analysis, analysis strategies based on time series and spatial distribution characteristics are employed to deeply mine the potential correlations between real-time and historical data, accurately capturing the changing patterns of power generation under complex environments. This end-to-end optimization from data acquisition and analysis to predictive output enables the method and system to effectively overcome the shortcomings of traditional technologies, providing an efficient and accurate solution for wind farm power generation prediction and facilitating more refined and reliable power dispatch for the grid.

[0120] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0121] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for segmented prediction of wind farm power generation, characterized in that, The method includes: Step S1: Based on the wind farm multimodal environment perception model, obtain real-time wind speed and wind direction data within the wind farm area. The multimodal environment perception model collects data through multiple wind speed and wind direction sensor arrays distributed within the wind farm. Each sensor array includes no less than three sensors at different heights and positions. Step S2: Using the multi-dimensional perception analysis model of power generation, retrieve the historical power generation data of the wind farm. The historical power generation data includes hourly power generation records for at least one full calendar year. Step S3: Divide the future power generation time of the wind farm into multiple time periods according to a preset time interval. The time interval ranges from 15 minutes to 1 hour, and the different time periods are seamlessly connected. Step S4: For each time period, combine the real-time wind speed and wind direction change data and historical power generation data corresponding to that time period, and use analysis strategies based on time series and spatial distribution characteristics to analyze the potential correlation between the data; Step S5: Based on the correlation obtained from the analysis, the power generation of different time periods is predicted segment by segment using a calculation model with a dynamic weight adjustment mechanism. The dynamic weight adjustment mechanism adjusts the weights according to the degree of matching between real-time data and historical data. Step S6: The predicted power generation data for different time periods are structured, integrated, and formatted to generate a data format output suitable for the power grid dispatching system. The output data includes the start time of different time periods, the predicted power generation value, and the data confidence index. The data confidence index is determined based on the number of iterations of the calculation model and the degree of data convergence.

2. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, In step S4, when analyzing the potential correlations between the data, a multimodal environment perception fusion model is introduced. This model calculates the correlation degree of the environmental data using the following formula: Among them, R ij V represents the correlation between the i-th set of real-time environmental data and the j-th set of historical environmental data; i V j These are real-time and historical wind speed values, respectively; D i D j These are real-time and historical wind direction values, respectively; P i P j These represent the real-time and historical power generation at the corresponding time points, respectively; α1, α2, and α3 are weighting coefficients determined based on the characteristics of wind farm equipment and statistical analysis of historical data, and satisfy α1+α2+α3=1.

3. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, In step S5, the computational model with dynamic weight adjustment mechanism utilizes the power prediction optimization formula: Among them, P p β represents the power generation during the predicted period; n represents the number of historical data samples used in the calculation; β k P represents the dynamic weight of the k-th historical data sample, and its value is dynamically adjusted based on the similarity between the real-time data and the k-th historical data sample. h，k P represents the power generation of the k-th historical data sample. r，k γ1 represents the corresponding power generation calculated based on real-time data simulation; γ2 and γ1 are correction coefficients determined according to the operating status of wind farm equipment and environmental conditions, and γ1+γ2=1.

4. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, In step S1, the wind farm multimodal environmental perception model utilizes a spatial interpolation compensation algorithm during data acquisition. Specifically, for any two adjacent sensors, the wind speed V... c Wind direction D c Calculated using the following formula: Where V1 and V2 are the wind speed measurements of two adjacent sensors, D1 and D2 are the wind direction measurements of two adjacent sensors, and d1 and d2 are the distances from the point to be calculated to the two adjacent sensors.

5. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, In step S2, when the multi-dimensional power generation perception and analysis model extracts features from historical power generation data, it uses the time-power feature mapping formula: Among them, F t，p Let ΔP be the characteristic value of power change within the time interval Δt; t ω represents the change in power generation within a time interval Δt; t The time weighting coefficient is set according to the characteristics of electricity demand in different time periods.

6. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, In step S3, when dividing future power generation periods, the dynamic time period division formula is used: Among them, T s T represents the adjusted time period length. t The initial time period length is denoted by m; m is the number of sampling points for real-time wind speed data; V i Let be the real-time wind speed at the i-th sampling point; This represents the average real-time wind speed.

7. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, Step S3 includes the following sub-steps: Step S31: Determine the initial time period division scheme. Based on the regular needs of power grid dispatch and the fluctuation characteristics of historical data, set the initial time interval and number of time periods. Step S32: Collect real-time wind speed data and obtain the real-time wind speed sequence for the current moment and a certain future time period through the wind farm multimodal environment perception model; Step S33: Calculate the wind speed fluctuation coefficient. Calculate the standard deviation and coefficient of variation of wind speed based on the real-time wind speed sequence to assess the degree of wind speed fluctuation. Step S34: Adjust the time period division. Dynamically adjust the initial time period division scheme according to the wind speed fluctuation coefficient. When the wind speed fluctuation exceeds the preset value, shorten the time period length. When the wind speed fluctuation is lower than the preset value, extend the time period length.

8. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, Step S4 includes the following sub-steps: Step S41: Construct a data feature matrix by arranging real-time wind speed and wind direction data and historical power generation data according to time order and parameter category to form a multi-dimensional data feature matrix; Step S42: Perform data normalization processing, normalize the different parameters in the data feature matrix to make the data fall within the same order of magnitude range; Step S43: Using the sliding window algorithm, a fixed-size window is slid across the data feature matrix to extract combinations of data features within different time windows; Step S44: Calculate feature similarity. Use the cosine similarity algorithm to calculate the similarity between data feature combinations and other combinations within different time windows, and identify similar data patterns.

9. The method for segmented prediction of wind farm power generation according to claim 1, characterized in that, Step S5 includes the following sub-steps: Step S51: Determine the basic prediction model. Based on the equipment type and historical data characteristics of the wind farm, select a basic calculation model that meets the preset benchmark as the prediction framework. Step S52: Initialize the weight parameters by assigning initial weight values ​​to different input parameters in the basic prediction model. These weight values ​​are set based on the statistical analysis results of historical data. Step S53: Perform model training by inputting real-time and historical data into the basic prediction model and adjusting the weight parameters through iterative calculation to minimize the error between the model output and the actual power generation. Step S54: Output prediction results. The trained model predicts power generation at different time periods and outputs the prediction results and related confidence information.

10. A segmented prediction system for wind farm power generation, characterized in that, The system includes: The multimodal environmental data acquisition unit is deployed in the wind farm area to collect real-time wind speed and direction data through multiple wind speed and direction sensor arrays and transmit the data to the data processing unit. The historical data storage unit is used to store hourly power generation records of the wind farm for at least one full calendar year in the past, and can interact with the data processing unit. The time period segmentation unit, connected to the multimodal environment data acquisition unit and the historical data storage unit, is used to divide the future power generation time of the wind farm into multiple time periods according to preset rules; The data correlation analysis unit is connected to the multimodal environment data acquisition unit, the historical data storage unit, and the time period segmentation unit, respectively, and is used to analyze the potential correlation between real-time data and historical data in different time periods; The power prediction unit, connected to the data correlation analysis unit, is used to predict the power generation of different time periods segment by segment based on the correlation obtained by analysis and using a calculation model with a dynamic weight adjustment mechanism. The data output unit, connected to the power prediction unit, is used to output the predicted power generation data for different time periods to the power grid dispatch system after structured integration and formatting.

Images