Wind power daily power prediction method and system fusing physical mechanism and multi-model cooperation

Abstract

The application discloses a wind power daily power prediction method and system fusing physical mechanism and multi-model cooperation, and the method comprises the following steps: obtaining historical daily power data of a wind power station and prediction day meteorological data; obtaining time characteristics based on periodic coding of dates; mapping wind direction into wind direction efficiency characteristics; constructing a temperature effect table to obtain temperature effect characteristics; constructing wind speed physical characteristics based on a theoretical power curve model of a wind turbine; and constructing a physical guiding feature set.According to the wind speed segmentation, an adaptive prediction strategy is adopted: a special model is used in combination with historical similar day data for fusion prediction in a low wind speed section; a main model is used for prediction in combination with physical constraint correction in a high wind speed section; and the main model is directly used for prediction in a conventional wind speed section.The wind turbine physical mechanism is converted into model learnable input through the physical guiding feature set, a segmented prediction strategy is combined, the prediction accuracy, rationality and robustness in the whole wind speed section, especially in extreme working conditions, are effectively improved, and prediction results that violate physical common sense are reduced.

Description

Technical Field

[0001] This invention relates to the field of wind power forecasting technology, and in particular to a method and system for forecasting daily wind power output that integrates physical mechanisms and multi-model collaboration. Background Technology

[0002] Daily power generation forecasting for wind power plants is a crucial foundation for new energy power generation dispatch, electricity market trading, and operation and maintenance management. Accurate forecasting results are of great significance for improving the grid's capacity to accommodate wind power.

[0003] Currently, most existing wind power forecasting methods employ a single machine learning model and primarily rely on raw meteorological data (such as wind speed, wind direction, and temperature) or simple statistical features for prediction. However, existing technologies have the following significant shortcomings in practical applications: First, existing feature engineering is often disconnected from the actual physical mechanisms of wind turbines. This leads to biases in models at low wind speeds due to the combined effects of wind direction and temperature, while at high wind speeds, models are prone to overestimation, resulting in "high theoretical power generation but actual power curtailment."

[0004] Secondly, due to the significant differences in the operating characteristics of wind turbines across different wind speed ranges, a single model cannot adequately meet the forecasting needs of all operating conditions. For example, low wind speed ranges are characterized by "few samples and large fluctuations," while high wind speed ranges face "power curtailment uncertainty" and physical limit constraints. Existing forecasting schemes lack a refined segmentation mechanism based on meteorological conditions, fail to dynamically switch or integrate dedicated models (such as low wind speed rule models) across different wind speed ranges, and fail to impose physical boundary constraints based on historical experience on the output results. This leads to forecast results under extreme weather conditions often deviating from reality and exhibiting poor generalization ability.

[0005] Therefore, a new wind power forecasting scheme is needed. Summary of the Invention

[0006] The purpose of this invention is to provide a wind power daily power prediction method and system that integrates physical mechanism-guided feature construction with multi-model segmented collaborative prediction to improve the accuracy of daily power prediction across the entire wind speed range, especially under high and low wind speed boundary conditions.

[0007] To achieve the above objectives, this invention provides a method for predicting daily wind power output that integrates physical mechanisms and multi-model collaboration, comprising: Acquire historical daily power generation data of wind power plants and meteorological data for forecast days, including date, wind speed, wind direction, and temperature; Periodic encoding is performed based on the date to obtain time characteristics; Based on the preset wind direction sector division rules, the wind direction is mapped to a wind direction efficiency feature, which includes a wind direction sector flag and / or wind direction efficiency. A temperature effect table is constructed by performing joint segmented statistics on wind speed and temperature of historical samples. The temperature effect table includes several efficiency reduction factors. The temperature effect table is queried based on the wind speed and temperature to obtain temperature effect characteristics including the corresponding efficiency reduction factors. Based on the theoretical power curve model of wind turbine units and combined with the wind speed hard limiting mechanism, wind speed physical characteristics are constructed. The wind speed physical characteristics include at least the theoretical daily power calculated by the theoretical power curve model. Construct a physical guidance feature set that includes the time feature, the wind direction efficiency feature, the temperature effect feature, and the wind speed physical feature; When the wind speed falls into the preset low wind speed range, the physical guidance feature set is input into the low wind speed special prediction model, and combined with historical similar daily power generation data for fusion prediction to obtain the first predicted power value under low wind speed conditions. When the wind speed falls into the preset high wind speed range, the physical guidance feature set is input into the main prediction model for preliminary prediction, and the preliminary prediction result is corrected in combination with the physical operation constraints of the wind turbine to obtain the second predicted power value under high wind speed conditions. When the wind speed falls within the preset normal wind speed range, the physical guidance feature set is input into the main prediction model for prediction to obtain the third predicted power value under normal wind speed conditions.

[0008] Preferably, the physical guidance feature set further includes electricity-derived features, which are calculated from the historical daily electricity data and include at least one of the lag value of the daily electricity and a sliding statistic.

[0009] Preferably, during the training phase of the prediction model, the physical guidance feature set further includes power-restricted day features, the construction of which includes: For sample days in the historical samples that meet the preset wind speed conditions, the theoretical daily electricity consumption of that sample day is obtained based on the physical characteristics of the wind speed. Based on the wind direction efficiency characteristics and the efficiency reduction factor, the theoretical daily electricity consumption is corrected to obtain the corrected theoretical daily electricity consumption. Calculate the ratio of the actual daily electricity consumption of the sample day to the corrected theoretical daily electricity consumption. When the ratio is less than a preset threshold, mark the sample day as a suspected power rationing day and include the suspected power rationing day mark as a power rationing day feature in the physical guidance feature set.

[0010] Preferably, the meteorological data also includes air pressure, and the physical guidance feature set also includes air pressure features generated based on the air pressure that reflect changes in air pressure.

[0011] Preferably, the wind direction efficiency is obtained from historical sample statistics, and the statistical method includes: The wind speed of historical samples is divided into segments according to preset intervals to obtain several wind speed segments. Within each wind speed range, a unit power generation efficiency value is calculated for the corresponding wind direction sector. The unit power generation efficiency value is a statistical measure of the ratio of the actual daily electricity consumption of historical samples within the wind direction sector to the theoretical daily electricity consumption calculated by the theoretical electricity consumption curve model. The corresponding wind direction efficiency is determined based on the unit power generation efficiency value of each wind direction sector.

[0012] Preferably, the method for constructing the temperature effect table includes: The historical samples were divided into several wind speed ranges and the historical samples were divided into several temperature ranges. For each historical sample falling into a combination of wind speed and temperature, the ratio of the actual daily electricity consumption to the theoretical daily electricity consumption after wind direction efficiency correction is calculated, and the statistical value of the ratio is used as the efficiency reduction factor corresponding to that combination of wind speed and temperature. Several of the aforementioned efficiency reduction factors are written into a two-dimensional mapping table constructed using wind speed segment index and temperature segment index to obtain the temperature effect table.

[0013] Preferably, the wind speed hard limiting mechanism includes: when the wind speed exceeds a preset upper limit, modifying the wind speed to the upper limit wind speed; when the wind speed is lower than a preset lower limit, modifying the wind speed to the lower limit wind speed.

[0014] Preferably, the physical guidance feature set further includes physical interaction features, which include a joint data set of wind speed and temperature and a joint data set of wind speed and wind direction.

[0015] Preferably, the method for fusing predictions to obtain the first predicted energy value includes: By using multi-level similarity conditions, matching historical similar day power generation data is found from historical data, and rule-based prediction values ​​are calculated based on the historical similar day power generation data; Obtain the model prediction value output by the low wind speed dedicated prediction model; Calculate the coefficient of variation of the electricity values ​​corresponding to the historical similar daily power generation data, and the relative difference between the rule-predicted value and the model-predicted value; Based on the coefficient of variation and the relative difference, the first weight of the rule prediction value and the second weight of the model prediction value are dynamically allocated; Using the first weight and the second weight, the rule-predicted value and the model-predicted value are weighted and fused to obtain the first predicted power value.

[0016] Preferably, the method for calculating the rule prediction value includes: Based on the preset wind direction sector division rules, the wind direction is mapped to wind direction efficiency; A historical similar daily power generation sample set is obtained by searching based on at least two of the multi-level similarity conditions, including wind speed, wind direction efficiency, temperature, and month. The power generation of the historical similar daily power generation sample set is calculated using a weighted average or median, and the predicted value is obtained after fine-tuning with wind speed and temperature.

[0017] Preferably, in the process of predicting the second predicted power value, the method for correcting the preliminary prediction result in conjunction with the physical operating constraints of the wind turbine includes: The meteorological data also includes wind direction and temperature; based on the preset wind direction sector division rules, the wind direction is mapped to wind direction efficiency; Based on the wind direction efficiency, the temperature, and the wind speed, a set of similar condition operation statistics is retrieved from the historical power generation dataset. Select historical quantiles or maximum values ​​from the set of similar condition running statistics as dynamic upper bounds, and restrict the preliminary prediction results within the dynamic upper bounds; Based on the preset correction rules for the combined high wind speed and low temperature scenario, logical adjustments and protective lower bound constraints are applied to the preliminary prediction results.

[0018] Preferably, the method for correcting the preliminary prediction results in conjunction with the physical operating constraints of the wind turbine further includes: A temperature effect table is constructed, which includes several efficiency reduction factors corresponding to different wind speeds and temperatures. Based on the wind speed and the temperature, the temperature effect table is queried to obtain the corresponding efficiency reduction factor; Based on the obtained efficiency reduction factor and the preset correction rules for the combined high wind speed and low temperature scenario, logical adjustments and protective lower bound constraints are applied to the preliminary prediction results.

[0019] Preferably, the training methods for the main prediction model and the low-wind-speed-specific prediction model include: Obtain historical daily electricity consumption sequences and corresponding historical meteorological data to form a training sample set; The training sample set is weighted in segments based on the time difference between the sample date and the training deadline, so that the weight of recent samples is higher than that of distant samples. The physical guidance feature set is constructed based on the weighted full training sample set; The model is trained in the candidate model pool based on the physical guidance feature set, and the best model is selected based on the evaluation metrics.

[0020] The present invention also provides a wind power daily power generation prediction system, which predicts the daily power generation of wind power stations based on the wind power daily power generation prediction method described above.

[0021] The present invention also provides a wind power daily electricity generation prediction system, which includes: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the wind power daily electricity forecasting method as described above.

[0022] The present invention also provides a computer-readable storage medium comprising a computer program that can be executed by a processor to perform the wind power daily power prediction method as described above.

[0023] Compared with existing technologies, the wind power daily power generation prediction method provided by the above-mentioned technical solution constructs a physical guidance feature set that includes time period, wind direction efficiency, temperature effect, and theoretical power generation. This deeply transforms the actual physical operation mechanism of wind turbines into a highly efficient input that the model can learn, effectively reducing the probability of pure data-driven models giving predictions that violate physical common sense. Furthermore, in conjunction with an adaptive segmented prediction strategy based on wind speed conditions, a dedicated model is fused with historical similar patterns in the low wind speed range to overcome the problem of large data fluctuations. In the high wind speed range, physical operation constraints are introduced to post-process and correct the main model results to eliminate the overestimation risk caused by power curtailment and extreme weather. In addition, the deep integration of physical mechanism features and multi-model collaboration achieves complementary advantages between data-driven approaches and engineering experience, significantly improving the prediction accuracy, rationality, and robustness of the system across the entire wind speed range, especially under extreme high and low wind speed boundary conditions. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating the construction principle of the physical guidance feature set in an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the adaptive selection prediction model in an embodiment of the present invention. Detailed Implementation

[0026] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0027] This embodiment discloses a method for predicting daily wind power generation, applicable to scenarios of day-ahead planning, power generation scheduling, and operation and maintenance management of wind power stations, for predicting the daily power generation on the predicted day.

[0028] This prediction method includes a feature processing flow and a model adaptive selection flow, such as... Figure 1 The feature processing flow is as follows: S1. Data Acquisition and Preprocessing.

[0029] Obtain the historical daily power generation sequence of the wind power station, including the date and the actual power generation for that day; obtain the meteorological data for the forecast day, which includes the date and the corresponding wind speed, wind direction, temperature, etc. The meteorological data can come from wind measurement towers, weather forecasts, or multi-source fusion data.

[0030] To ensure consistency of samples from the same day, the electricity data and meteorological data are aligned at the daily granularity: when the meteorological data is hourly, a fixed time offset can be applied first (for example, the time is uniformly reduced by 1 hour to match the electricity statistics caliber), and then the data is aggregated by day to obtain the daily average wind speed, prevailing wind direction (or average wind direction), daily average temperature (or daily minimum temperature), and missing value imputation and outlier screening are completed.

[0031] S2. Date-based periodic encoding to obtain time features.

[0032] Dates are mapped to time features that characterize periodic patterns, enabling predictive models to capture seasonality and weekly cycles. Periodic encoding can be done using sine and cosine encoding forms, for example, constructing sin(2πm / 12) and cos(2πm / 12) for the month m (1–12), and sin(2πw / 7) and cos(2πw / 7) for the day of the week w (0–6).

[0033] In some optional implementations, the time features can further include year, quarter, day of the year, whether it is a weekend, day of the month, and their periodic encoding to adapt to the periodic differences of different wind fields. By using periodic encoding, the periodic boundaries (such as December and January, Sunday and Monday) remain continuous in the feature space, reducing the risk that the prediction model will misjudge periodic breakpoints as abrupt changes.

[0034] S3. Construct wind direction efficiency features based on wind direction sector division rules.

[0035] The wind direction is a cyclic variable from 0° to 360° and is affected by topography, wake, and unit layout. Different wind direction ranges may correspond to different effective inflow efficiencies.

[0036] This step uses a preset wind direction sector division rule to map wind direction to wind direction efficiency in order to obtain wind direction efficiency characteristics.

[0037] Specifically, the 360° space can be divided into several non-overlapping sector intervals, a flag bit (such as a One-hot flag) can be generated for each sector interval, and wind direction efficiency can be further configured for the sector interval to characterize the efficiency level of the sector interval relative to the reference sector.

[0038] The rules for dividing sectors by wind direction can be determined by wind field experience or by historical statistics to identify high-efficiency, sub-high-efficiency, and low-efficiency sectors. To ensure continuity, in some optional implementations, the wind direction efficiency feature can also include the sine and cosine components of wind direction obtained by cyclic encoding, such as sin(2πD / 360) and co(2πD / 360), to enhance the predictive model's ability to express changes near sector boundaries. By linking sector flags with wind direction efficiency, the nonlinear influence of wind direction on power generation efficiency is explicitly introduced into the feature space.

[0039] S4. Construct a temperature effect table through joint segmented statistics of wind speed and temperature, and generate temperature effect characteristics.

[0040] The effect of temperature on the power generation efficiency of wind turbines is usually coupled with wind speed. Specifically, under low temperature conditions, factors such as the risk of blade icing, changes in materials and lubrication conditions, and the triggering of control strategies may lead to efficiency degradation, and this degradation manifests differently in different wind speed ranges.

[0041] This step involves joint segmented statistical analysis of wind speed and temperature data from historical samples to construct a temperature effect table. The method for constructing the temperature effect table includes: The historical samples were divided into several wind speed ranges and the historical samples were divided into several temperature ranges. For historical samples falling into each wind speed and temperature combination, the ratio of actual daily electricity consumption to theoretical daily electricity consumption after wind direction efficiency correction is calculated, and the statistical value of the ratio is used as the efficiency reduction factor corresponding to that wind speed and temperature combination.

[0042] Several efficiency reduction factors are written into a two-dimensional mapping table constructed with wind speed segment index and temperature segment index to obtain a table of efficiency reduction factors with temperature effect.

[0043] When making predictions or inferences, the temperature effect table is queried based on the wind speed and temperature of the date to be predicted to obtain the corresponding reduction factor, and the reduction factor is output as part of the temperature effect characteristics.

[0044] S5. Construct wind speed physical characteristics based on theoretical power curve model and combined with wind speed hard limiting mechanism.

[0045] The power (electricity) of wind turbines has a clear physical boundary and a nonlinear relationship with wind speed, especially near the rated wind speed where saturation characteristics will appear.

[0046] This step first applies a hard limit to the wind speed: when the wind speed exceeds the upper limit, the wind speed is changed to the upper limit; when the wind speed is lower than the lower limit, the wind speed is changed to the lower limit.

[0047] The upper limit wind speed can be the unit's cut-out wind speed or the engineering setting value, and the lower limit wind speed can be 0 or the lower boundary near the cut-in wind speed.

[0048] The theoretical daily power consumption is then calculated based on the theoretical power consumption curve model. The theoretical power consumption curve model can use piecewise functions, interpolated curves, or lookup table curves. For example, a cubic relationship can be used to approximate the power consumption in the range below the rated wind speed, while a linear or saturated segment approximation can be used near the rated wind speed, and a power limit of 0 or upper limit protection can be set in the cut-off range.

[0049] By combining theoretical power and hard limiting mechanisms, wind speed physical characteristics consistent with physical constraints are formed, reducing the unconstrained extrapolation error of the prediction model in the high wind speed range.

[0050] S6. Construct a physical guidance feature set and form a prediction input.

[0051] The physical guidance feature set is formed by combining time features, wind direction efficiency features, temperature effect features, and wind speed physical features. Continuous features can be standardized or normalized. Discrete features can be directly encoded using one-hot encoding or categorical encoding. In some optional implementations, the list of feature column names and preprocessing parameters used in the training phase can be retained to ensure that the feature construction in the prediction phase is consistent with that in training.

[0052] like Figure 2 The adaptive model selection process is as follows: S7, Segmented Coordinated Prediction and Output of Wind Speed.

[0053] Set preset thresholds for low and high wind speeds, denoted as τlow and τhigh, respectively. The thresholds can be determined based on the unit's switching characteristics, the error-sensitive area near the rated value, the power curtailment characteristics of the site, and historical statistics. For example, τlow = 3 m / s and τhigh = 8.5 m / s can be used, but they are not limited to these. They can be adjusted to other ranges according to the characteristics of the site, as long as they can distinguish between the low wind speed sparse fluctuation area and the high wind speed power curtailment / saturation risk area.

[0054] Based on the predicted daily wind speed range, the following corresponding strategy will be implemented: 1) Low wind speed segment (wind speed v < τlow): Input the physical guidance feature set into the low wind speed dedicated prediction model to obtain the model prediction value; at the same time, combine it with historical similar daily power generation data for fusion prediction to obtain the first predicted power value under low wind speed conditions.

[0055] The fusion method can be weighted average, weighted superposition, or other computational forms that can achieve the synthesis of two types of information.

[0056] 2) Normal wind speed range (τlow≤v≤τhigh): The physical guidance feature set is input into the main prediction model to output a third predicted electricity value. The main prediction model can be a regression model or an ensemble model, including but not limited to ridge regression, random forest, gradient boosting tree, XGBoost, LightGBM, CatBoost, neural networks, etc., as long as it can output a daily electricity prediction value based on the features. The normal wind speed range mainly relies on the global fitting capability of the main model to obtain a stable baseline prediction.

[0057] 3) High wind speed segment (v>τhigh): Input the physical guidance feature set into the main prediction model to obtain the preliminary prediction results; then, combine the physical operation constraints of the wind turbine to correct the preliminary prediction results and obtain the second predicted power value.

[0058] The efficiency reduction factor and the physical operating constraints of the wind turbine are used to reflect the wind turbine's power curtailment capacity or historical operating range, so that the corrected predicted value does not exceed the physically feasible range.

[0059] Finally, the first, second, or third predicted energy value is output as the prediction result based on the wind speed segment.

[0060] In another embodiment, to further utilize the operational status and short-term trend information contained in the electricity consumption sequence, the physical guidance feature set can also include electricity consumption-derived features. These features are calculated from historical daily electricity consumption data and include at least one of lag values ​​and moving averages. Lag values ​​can be the daily electricity consumption of the previous 1 day, 7 days, or 30 days; moving averages can be the moving average, moving standard deviation, coefficient of variation, or trend quantities such as day-on-day and week-on-week comparisons for a 7-day or 30-day window.

[0061] During the forecasting phase, the calculation of lag and sliding statistics can be based on the observed historical electricity. When rolling multi-day forecasts are required, the sliding feature can be constructed by prioritizing the observed values ​​and recursively using the predicted values ​​for the missing parts, and reasonable protection can be set for the recursion error (such as truncation of abnormal jumps within the sliding window).

[0062] Since the power generation features provide information on recent operating baselines and fluctuation intensity, the model can still distinguish power generation differences caused by changes in equipment availability and adjustments to power curtailment strategies when meteorological inputs are similar, thereby improving short-term forecast stability and reducing drift risk.

[0063] In another embodiment, wind power stations are generally designated as curtailment days. Curtailment days refer to specific dates during the operation of wind power stations where, due to grid dispatch requirements, transmission line capacity limitations, or imbalances in power supply and demand, grid operators force wind power stations to reduce their power generation or stop generating electricity.

[0064] Setting power curtailment days is primarily to ensure the safe and stable operation of the power grid, preventing frequency fluctuations or equipment overload caused by excessive wind power output exceeding the grid's absorption capacity. It also allows for adjusting the power output structure during periods of low electricity demand to ensure overall power system balance. Although wind resources may be favorable at these times, wind farms must comply with grid dispatch instructions and cannot operate at full capacity.

[0065] To avoid discrepancies between the predicted values ​​output by the prediction model on days with power rationing and the actual values, features for days with power rationing are constructed during the training phase.

[0066] In one specific implementation, during the training phase, for sample days in the historical samples that meet preset wind speed conditions (e.g., wind speed not less than the high wind speed threshold of 8.5 m / s or within the rated range), the theoretical daily power generation for that sample day is obtained from the physical characteristics of wind speed. Then, the theoretical daily power generation is corrected using the wind direction efficiency characteristics and the efficiency reduction factor obtained from the temperature effect table, resulting in the corrected theoretical daily power generation. The correction can be performed using a multiplicative correction method: combining the theoretical daily power generation with the wind direction efficiency and the efficiency reduction factor to obtain a reasonable upper bound reference for power generation under the given wind direction sector and temperature conditions.

[0067] Then, the ratio of the actual daily electricity consumption to the corrected theoretical daily electricity consumption for the sample day is calculated. When the ratio is less than a preset threshold (e.g., 0.6), the sample day is marked as a suspected power rationing day, and the suspected power rationing day marker is included as a power rationing day feature in the physical guidance feature set.

[0068] By introducing features specific to days with power rationing, the "weather-power" relationship in the training samples more closely approximates the normal generation mechanism, resulting in a more stable mapping learned by the prediction model. This enables the model to identify and handle abnormal power rationing situations and avoid overestimation during high-wind-speed periods. In actual measured data, this feature reduced the prediction error for high-wind-speed periods (>8.5 m / s) by 18.5%, with a particularly significant improvement in prediction accuracy for days with severe power rationing.

[0069] In another embodiment, to further characterize the impact of air density changes on power generation efficiency, meteorological data may also include air pressure, and air pressure features reflecting air pressure changes may be introduced into the physical guidance feature set.

[0070] In one specific implementation, the pressure characteristics may include at least one of the following: the original pressure value, the pressure lag value, and the pressure sliding statistic. The pressure lag value may be the pressure of the previous day or the previous 7 days, and the pressure sliding statistic may be the 7-day window moving average and the moving standard deviation.

[0071] To enhance the physical meaning, in some optional implementations, air pressure features can be combined with temperature to generate an air density factor, which is then input into the prediction model as part of the air pressure-related features. Air pressure changes are usually slower than wind speed changes, but abrupt temperature changes or weather system transitions can affect air density. Introducing air pressure change features helps the prediction model distinguish the differences in power generation caused by different air densities under the same wind speed conditions, thereby improving the generalization consistency across weather patterns.

[0072] In another embodiment, the unit power generation efficiency value can be calculated based on historical samples for different wind speed segments, and the wind direction efficiency of each wind sector can be determined accordingly.

[0073] In one specific implementation, the wind speed of the historical samples is first segmented according to a preset interval. For example, it can be divided into continuous intervals such as [0,1), [1,2), [2,3) at 1 m / s intervals, or into intervals such as [0,3), [3,6), [6,8.5), (8.5,10.5] according to business thresholds.

[0074] Within each wind speed range, the unit power generation efficiency value for each wind direction sector is then statistically analyzed. The unit power generation efficiency value is defined as the statistical measure of the ratio of the actual daily electricity consumption of historical samples within that sector to the theoretical daily electricity consumption calculated from the theoretical electricity consumption curve model. The efficiency reduction factor statistic can be the mean, median, P50 quantile, P75 quantile, or a combination thereof.

[0075] If the unit power generation efficiency statistic for the primary high-efficiency sector within a certain wind speed range is 0.92, for the secondary high-efficiency sector it is 0.83, and for the inefficient sector it is 0.71, then the primary high-efficiency sector can be normalized to 1.00, and the corresponding secondary high-efficiency and inefficient wind direction efficiencies can be determined as 0.90 and 0.77, respectively; alternatively, the above statistics can be directly used as the wind direction efficiency.

[0076] The reason for using segmented wind speed statistics is that the difference in wind direction efficiency does not exist proportionally across all wind speeds. At low wind speeds, the impact of wake and terrain obstruction on power generation may be insignificant; however, at medium to high wind speeds, the difference in inflow quality caused by wind direction differences is more readily apparent. Therefore, the wind direction efficiency obtained through segmented statistics better reflects the actual wind field characteristics than a uniform coefficient across all wind speeds.

[0077] The wind efficiency constructed in this way can be directly used as a wind direction correction term in the theoretical power correction and temperature effect table construction, thereby improving the stability of the prediction model in multi-wind direction distribution scenarios.

[0078] In another embodiment, based on the above embodiments, to enhance the predictive model's ability to learn the coupling relationship across physical quantities, physical interaction features can be added to the physical guidance feature set. Physical interaction features can consist of joint data sets of wind speed and temperature, and joint data sets of wind speed and wind direction.

[0079] The joint data set can be input into the prediction model as explicit interaction terms or as a combined field for the prediction model to automatically learn nonlinear relationships. By introducing interactive expressions of wind speed and temperature, and wind speed and wind direction, the prediction model can more easily form an internal representation of "the difference in power consumption caused by changes in temperature or wind direction at the same wind speed", thereby improving the prediction consistency under complex operating conditions.

[0080] In another embodiment, in order to simultaneously consider the robustness of empirical estimation and the pattern recognition capability of the low-wind-speed-specific prediction model in the low-wind-speed segment, this embodiment introduces a dynamic weight allocation mechanism for the fusion prediction in the low-wind-speed segment.

[0081] Specifically: In low wind speed ranges, first, by using multi-level similarity conditions, search for matching historical data on similar days of power generation to identify efficiency reduction factors. Then, the rule-based predicted value is calculated based on historical similar daily power generation data; Then the low-wind-speed-specific prediction model outputs the model prediction value; Subsequently, the coefficient of variation (cv) of the power generation data corresponding to the historical similar days is calculated. cv is used to characterize the dispersion of the historical similar day sample set of efficiency reduction factor. The relative difference (δ) between the rule-predicted values ​​and the model-predicted values ​​is calculated. δ characterizes the degree of disagreement between the two types of predictions. δ can be calculated as follows:

[0082] in, For the rule-predicted value, The "1" in the denominator is used to prevent the ratio from becoming abnormal due to an excessively small denominator, which is the model's predicted value. Based on cv and δ, the first weight of the rule prediction value and the second weight of the model prediction value are determined by a preset dynamic weight allocation function, and the two are made to meet the normalization constraint (e.g., the weight sum is 1). The dynamic weight allocation function follows these principles: when cv increases or δ increases, the second weight is increased and the first weight is decreased; when cv is small and δ is small, the first weight is increased. After the dynamic weights are determined, the rule-predicted values ​​and model-predicted values ​​are weighted and fused to obtain the first predicted electricity value.

[0083] For example, CV and δ can be categorized into three levels—"low," "medium," and "high"—based on historical statistics. When either CV or δ is high, the second weight is higher than the first weight (e.g., 0.7 and 0.3); when both CV and δ are low, the first weight is higher than the second weight (e.g., 0.7 and 0.3); in other cases, a near-balanced weight is used (e.g., 0.5 and 0.5). The level division boundary can be determined by the statistical distribution of historical similar day sample sets.

[0084] This dynamic weighting mechanism reduces reliance on a single empirical estimate when historical samples fluctuate significantly or when there is a large discrepancy between the two types of predictions in the low-wind-speed segment. It also increases the contribution of empirical estimates when the samples are stable and the predictions are consistent, thereby reducing the unstable output of predictions in the low-wind-speed segment.

[0085] Furthermore, the methods for calculating the rule-predicted values ​​include: A historical similar daily power generation sample set is obtained by searching based on multi-level similarity conditions including wind speed, wind direction efficiency, temperature, and at least two of the following factors: month. In some optional implementations, wind speed and month can be used as the first-level screening conditions, and wind direction efficiency and temperature can be introduced as further subdivision conditions.

[0086] Multi-level retrieval can gradually relax conditions according to a "strict to lenient" strategy to obtain sufficient samples. For example, matching can be done first based on wind speed within ±0.2; if the sample is insufficient, the range can be expanded to ±0.5. Temperature conditions can be matched first at ±5, and if the sample is insufficient, the range can be expanded to ±10. Monthly conditions are used to suppress deviations caused by seasonal differences. Wind direction efficiency is used to reflect the differences in power generation efficiency under different wind directions due to wake, terrain, or layout. It can be obtained from historical operation statistics and discretized into groups according to preset intervals.

[0087] After obtaining a historical sample set of similar daily power generation, the daily power generation of the sample set is aggregated to obtain the rule-based prediction value. The aggregation method can be either weighted average or median: when there are a few outliers in the sample set, using the median can reduce the impact of outliers; when the samples have different confidence levels, a weighted average can be used, and samples that are closer to the conditions of the prediction day can be given higher weights.

[0088] Subsequently, the wind speed and temperature prediction values ​​can be fine-tuned: when the predicted daily wind speed or temperature is near the boundary of the sample set conditions, the prediction values ​​can be slightly corrected according to the direction of deviation to reduce the system bias caused by condition discretization.

[0089] Through multi-level similarity condition retrieval and robust aggregation, the rule-based predictions can provide interpretable empirical estimates in low-wind-speed, sparse-sample scenarios and provide a stable baseline for subsequent fusion.

[0090] In another embodiment, after obtaining the weighted fusion result for the low wind speed segment, the historical quantile interval of its power generation is calculated based on historical similar daily power generation data of the efficiency reduction factor, and the fusion result is restricted to this interval. In a specific example, the 25th quantile and the 75th quantile can be used to construct the interval: when the fusion result is below the 25th quantile, the output is truncated to the 25th quantile; when the fusion result is above the 75th quantile, the output is truncated to the 75th quantile; when the fusion result is within the interval, it remains unchanged. The value of the quantile interval can be adjusted according to the site's preference, as long as the statistical range under historical similar conditions is used as a reasonable boundary.

[0091] When the number of historical similar day samples is small, leading to unstable quantiles, one can choose to skip pruning or use a more robust boundary (e.g., using the minimum and maximum values ​​of the samples to form the boundary) to ensure the feasibility and numerical stability of the constraint operation.

[0092] This pruning constraint makes the output of the low wind speed segment less susceptible to amplification by a small number of outlier samples or fluctuations in the fusion weights, thereby reducing the extreme deviation probability of the low wind speed segment prediction and improving the smoothness of the prediction sequence.

[0093] In another embodiment, in order to make the prediction results under high wind speed conditions consistent with the historical reachable boundary and power curtailment characteristics of wind power stations, a dynamic upper bound constraint based on similarity condition statistics is introduced into the preliminary prediction results output by the main prediction model in the high wind speed segment, and a correction rule and a protection lower bound constraint for the composite scenario of high wind speed and low temperature are superimposed.

[0094] Specifically, based on wind direction efficiency, temperature, and wind speed, a set of statistics for similar operating conditions is retrieved from historical power generation datasets.

[0095] Similarity criteria retrieval can employ a progressively relaxed strategy to balance similarity and sample size: priority should be given to conditions such as similar wind direction efficiency (or falling into the same efficiency category), temperature within the same temperature range, and wind speed within a range similar to the predicted day. When samples are insufficient, the matching window for wind speed and temperature can be expanded (e.g., wind speed from a narrow window to a wider window, and temperature from a narrow window to a wider window), and seasonal conditions such as month can be retained to suppress the bias introduced by seasonal differences.

[0096] The retrieved set of similar conditional operation statistics includes at least the quantile, median, or maximum values ​​of the daily electricity volume corresponding to the similar sample set, which are used to characterize the historical achievable level under the meteorological combination.

[0097] After obtaining the set of statistics for similar conditions, historical quantiles or maximum values ​​are selected as dynamic upper bounds, and the preliminary prediction results of the master prediction model are restricted to these dynamic upper bounds.

[0098] The selection of the dynamic upper bound can be determined based on the wind power plant's risk appetite or the uncertainty of power curtailment. That is, when it is desirable to more conservatively suppress overestimation, a lower quantile can be selected as the upper bound; when it is desirable to reduce underestimation, the median or maximum value can be selected as the upper bound. This upper bound constraint is a boundary correction for the output layer, ensuring that the preliminary prediction results do not exceed the verifiable range under similar historical conditions, thereby reducing the systematic overestimation caused by "theoretical power generation being too high while actual power curtailment" in high wind speed sections.

[0099] After the upper bound is set, logical adjustments and protective lower bound constraints are applied to the prediction results based on the preset correction rules for the combined high wind speed and low temperature scenarios.

[0100] Composite scenario correction rules are used to handle the superimposed effects when high wind speed and low temperature occur simultaneously: for example, when the temperature falls into the low temperature range and the wind speed is in the high wind speed range, the results after the upper limit are applied with a reduction correction or further contraction, and a boundary is set for the correction magnitude to avoid over-compression.

[0101] The lower bound constraint is used to prevent the forecast result from falling into an obviously unreasonable low range after multiple corrections. It can be formed by selecting a lower quantile or other statistics from the set of statistics for similar operating conditions. When the corrected forecast result is lower than the lower bound, it is raised to the lower bound. Finally, a non-negativity constraint can be applied to avoid negative daily electricity forecast values.

[0102] In this embodiment, wind direction differences are transformed into efficiency differences and incorporated into similarity condition retrieval, making the historical statistical boundary closer to the actual achievable capacity of the unit under different incoming flow directions; by using dynamic upper bound restrictions based on similarity condition statistics, the output of the main prediction model in the high wind speed range is constrained to the historically verifiable range; by superimposing high wind speed and low temperature composite rules and protective lower bound constraints, the prediction results can suppress the risk of overestimation while avoiding unreasonable underestimation, thereby improving the physical consistency and operational availability of the prediction output under high wind speed and special meteorological conditions.

[0103] Based on the above embodiments, in order to perform calculable efficiency reduction correction for high wind speed predictions in scenarios including the influence of low temperatures, the method for correcting the preliminary prediction results of efficiency reduction factors in conjunction with the physical operating constraints of the wind turbine further includes: Based on the predicted wind speed and temperature, the temperature effect table is consulted to obtain the corresponding efficiency reduction factor. The efficiency reduction factor is combined with the predicted value after dynamic upper bound correction, for example, by multiplying and scaling the predicted value to obtain a corrected value that takes into account the effect of temperature. Subsequently, combining the obtained efficiency reduction factor with the preset correction rules for the combined high wind speed and low temperature scenarios, logical adjustments and protection lower bound constraints are applied to the correction value. When both high wind speed and low temperature conditions are met, the predicted value can be further reduced according to the efficiency reduction factor based on the upper bound constraint, and the predicted value is ensured to be no lower than the protection lower bound formed by the set of statistics for similar operating conditions of the efficiency reduction factor.

[0104] In another embodiment, in order to obtain a master prediction model and a low-wind-speed-specific prediction model that can be used for the prediction phase, this embodiment provides a training method, which includes data construction, time decay weighting, candidate model training, and evaluation and selection.

[0105] First, obtain historical daily electricity consumption sequences and corresponding historical meteorological data to form a training sample set.

[0106] Subsequently, the training sample set is weighted in segments based on the time difference between the sample date and the training deadline, so that the weight of recent samples is higher than that of distant samples.

[0107] In a specific example, weighting can be segmented into segments: high weight for the last 2 months, standard weight for 2–6 months, reduced weight for 6–12 months, and low weight for over one year. The boundaries of these segments and the weight values ​​can be adjusted according to the rate of change in the station's operating mode, as long as a monotonic decay relationship of "higher in the near term and lower in the long term" is satisfied. The weights of the efficiency degradation factor are used as sample weights in the model training loss calculation, thus making the model more focused on recent operating modes.

[0108] Then, a physical guidance feature set is constructed based on the weighted full training sample set.

[0109] Finally, the model is trained in the candidate model pool based on the physical guidance feature set, and the best model is selected based on the evaluation metrics.

[0110] In some alternative implementations, a subset of samples with wind speeds falling into the low wind speed range can be selected from the training sample set, features can be constructed or selected separately for low wind speed scenarios, and a low wind speed-specific prediction model can be obtained by training and selecting the best model in the candidate model pool.

[0111] By using time decay weighting and multi-model pool training optimization, the prediction model can maintain its adaptability to recent data even as the operating mode changes over time, while providing reusable model components for the three-layer architecture of the prediction stage.

[0112] In another preferred embodiment of the present invention, a wind power daily power generation prediction system is also disclosed, which predicts the daily power generation of a wind power station based on the wind power daily power generation prediction method in the above embodiments.

[0113] Therefore, the prediction process of this prediction system is as follows: (1) Data level: Collect historical daily electricity consumption and multi-source meteorological data, perform time offset alignment, daily aggregation, missing and anomaly processing, and configure processing strategies for special samples such as maintenance days and suspected power restriction days; (2) Feature level: Construct time periodic coding features, wind direction sector and wind direction efficiency features, temperature effect features of wind speed-temperature joint segmentation, theoretical power curve model and wind speed physical features of hard limit; add power derived features, power restriction day features, air pressure features and physical interaction features as needed, and solidify feature column names and preprocessing parameters for training and prediction consistency. (3) Model level: Train and select the main prediction model in the model pool based on time decay weighting; train a low wind speed special model for a subset of low wind speed samples; save the model, normalizer and feature list; (4) Prediction execution level: For the prediction day, construct a physical guidance feature set according to the same feature engineering logic and standardize it; trigger the segmentation strategy based on the wind speed threshold: for the low wind speed segment, perform the fusion of "similar day rule estimation and low wind speed special model" and perform quantile pruning; for the regular wind speed segment, the main model outputs directly; for the high wind speed segment, after the main model gives the initial value, construct a dynamic upper bound based on the wind direction efficiency, similarity condition statistics of wind speed segment and temperature segment, and superimpose the efficiency reduction factor, composite rule and protection lower bound constraint for post-processing; finally output non-negative daily electricity prediction results, and optionally output branches and constraint intermediate quantities for verification.

[0114] Because the system integrates physical mechanism characteristics (theoretical power boundary, wind direction efficiency difference, temperature coupling degradation) with multi-model segmented collaboration (low wind speed fusion, conventional master model, high wind speed constraint post-processing) in a unified manner, the prediction results are closer to historical achievable levels and have interpretable correction paths under the combined conditions of sparse fluctuations in low wind speed data, uncertainty of power rationing at high wind speeds, and low temperature, thereby improving the stability and engineering usability of predictions across the entire wind speed range.

[0115] This invention also discloses another wind power daily energy forecasting system, which includes one or more processors, a memory, and one or more programs. The one or more programs are stored in a degradation factor memory and configured to be executed by the degradation factor processors. The degradation factor program includes instructions for executing the wind power daily energy forecasting method described above. The processor can be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, used to execute the relevant programs to implement the functions required by the modules in the wind power daily energy forecasting system of this application embodiment, or to execute the wind power daily energy forecasting method of this application embodiment.

[0116] This invention also discloses a computer-readable storage medium comprising a computer program. The computer program, which is executable by a processor, performs the wind power daily electricity prediction method described above. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be read-only memory (ROM), random access memory (RAM), or magnetic media, such as floppy disks, hard disks, magnetic tapes, magnetic disks, or optical media, such as digital versatile discs (DVDs), or semiconductor media, such as solid-state disks (SSDs).

[0117] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned wind power daily electricity prediction method.

[0118] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for predicting daily wind power output that integrates physical mechanisms and multi-model collaboration, characterized in that, include: Acquire historical daily power generation data of wind power plants and meteorological data for forecast days, including date, wind speed, wind direction, and temperature; Periodic encoding is performed based on the date to obtain time characteristics; Based on the preset wind direction sector division rules, the wind direction is mapped to a wind direction efficiency feature, which includes a wind direction sector flag and / or wind direction efficiency. A temperature effect table is constructed by performing joint segmented statistics on wind speed and temperature of historical samples. The temperature effect table includes several efficiency reduction factors. The temperature effect table is queried based on the wind speed and temperature to obtain temperature effect characteristics including the corresponding efficiency reduction factors. Based on the theoretical power curve model of wind turbine generators and combined with the wind speed hard limiting mechanism, wind speed physical characteristics are constructed. The wind speed physical characteristics include at least the theoretical daily power calculated by the theoretical power curve model. Construct a physical guidance feature set that includes the time feature, the wind direction efficiency feature, the temperature effect feature, and the wind speed physical feature; When the wind speed falls into the preset low wind speed range, the physical guidance feature set is input into the low wind speed special prediction model, and combined with historical similar daily power generation data for fusion prediction to obtain the first predicted power value under low wind speed conditions. When the wind speed falls into the preset high wind speed range, the physical guidance feature set is input into the main prediction model for preliminary prediction, and the preliminary prediction result is corrected in combination with the physical operation constraints of the wind turbine to obtain the second predicted power value under high wind speed conditions. When the wind speed falls within the preset normal wind speed range, the physical guidance feature set is input into the main prediction model for prediction to obtain the third predicted power value under normal wind speed conditions.

2. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The physical guidance feature set also includes electricity-derived features, which are calculated from the historical daily electricity data and include at least one of the lag value of the daily electricity and a sliding statistic.

3. The wind power daily electricity generation prediction method according to claim 1, characterized in that, During the training phase of the prediction model, the physical guidance feature set also includes features related to power rationing days, the construction of which includes: For sample days in the historical samples that meet the preset wind speed conditions, the theoretical daily electricity consumption of that sample day is obtained based on the physical characteristics of the wind speed. Based on the wind direction efficiency characteristics and the efficiency reduction factor, the theoretical daily electricity consumption is corrected to obtain the corrected theoretical daily electricity consumption. Calculate the ratio of the actual daily electricity consumption of the sample day to the corrected theoretical daily electricity consumption. When the ratio is less than a preset threshold, mark the sample day as a suspected power rationing day and include the suspected power rationing day mark as a power rationing day feature in the physical guidance feature set.

4. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The meteorological data also includes air pressure, and the physical guidance feature set also includes air pressure features generated based on the air pressure that reflect changes in air pressure.

5. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The wind direction efficiency is obtained from historical sample statistics, and the statistical method includes: The wind speed of historical samples is divided into segments according to preset intervals to obtain several wind speed segments. Within each wind speed range, a unit power generation efficiency value is calculated for the corresponding wind direction sector. The unit power generation efficiency value is a statistical measure of the ratio of the actual daily electricity consumption of historical samples within the wind direction sector to the theoretical daily electricity consumption calculated by the theoretical electricity consumption curve model. The corresponding wind direction efficiency is determined based on the unit power generation efficiency value of each wind direction sector.

6. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The method for constructing the temperature effect table includes: The historical samples were divided into several wind speed ranges and the historical samples were divided into several temperature ranges. For each historical sample falling into a combination of wind speed and temperature, the ratio of the actual daily electricity consumption to the theoretical daily electricity consumption after wind direction efficiency correction is calculated, and the statistical value of the ratio is used as the efficiency reduction factor corresponding to that combination of wind speed and temperature. Several of the aforementioned efficiency reduction factors are written into a two-dimensional mapping table constructed using wind speed segment index and temperature segment index to obtain the temperature effect table.

7. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The wind speed hard limiting mechanism includes: when the wind speed exceeds a preset upper limit, the wind speed is modified to the upper limit wind speed; when the wind speed is lower than a preset lower limit, the wind speed is modified to the lower limit wind speed.

8. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The physical guidance feature set also includes physical interaction features, which include a joint data set of wind speed and temperature and a joint data set of wind speed and wind direction.

9. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The method for fusing predictions to obtain the first predicted energy value includes: By using multi-level similarity conditions, matching historical similar day power generation data is found in historical data, and rule-based prediction values ​​are calculated based on the historical similar day power generation data; Obtain the model prediction value output by the low wind speed dedicated prediction model; Calculate the coefficient of variation of the electricity values ​​corresponding to the historical similar daily power generation data, and the relative difference between the rule-predicted value and the model-predicted value; Based on the coefficient of variation and the relative difference, the first weight of the rule prediction value and the second weight of the model prediction value are dynamically allocated; Using the first weight and the second weight, the rule-predicted value and the model-predicted value are weighted and fused to obtain the first predicted power value.

10. The wind power daily electricity generation prediction method according to claim 9, characterized in that, The method for calculating the rule-predicted value includes: Based on the preset wind direction sector division rules, the wind direction is mapped to wind direction efficiency; A historical similar daily power generation sample set is obtained by searching based on at least two of the multi-level similarity conditions, including wind speed, wind direction efficiency, temperature, and month. The power generation of the historical similar daily power generation sample set is calculated using a weighted average or median, and the predicted value is obtained after fine-tuning with wind speed and temperature.

11. The wind power daily electricity generation prediction method according to claim 1, characterized in that, In the process of predicting the second predicted power value, the method for correcting the preliminary prediction result in combination with the physical operating constraints of the wind turbine includes: The meteorological data also includes wind direction and temperature; based on the preset wind direction sector division rules, the wind direction is mapped to wind direction efficiency; Based on the wind direction efficiency, the temperature, and the wind speed, a set of similar condition operation statistics is retrieved from the historical power generation dataset. Select historical quantiles or maximum values ​​from the set of similar condition running statistics as dynamic upper bounds, and restrict the preliminary prediction results within the dynamic upper bounds; Based on the preset correction rules for the combined high wind speed and low temperature scenario, logical adjustments and protective lower bound constraints are applied to the preliminary prediction results.

12. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The method for correcting the preliminary prediction results based on the physical operating constraints of the wind turbine also includes: A temperature effect table is constructed, which includes several efficiency reduction factors corresponding to different wind speeds and temperatures. Based on the wind speed and the temperature, the temperature effect table is queried to obtain the corresponding efficiency reduction factor; Based on the obtained efficiency reduction factor and the preset correction rules for the combined high wind speed and low temperature scenario, logical adjustments and protective lower bound constraints are applied to the preliminary prediction results.

13. The wind power daily electricity generation prediction method according to claim 1, characterized in that, The training methods for the main prediction model and the low-wind-speed-specific prediction model include: Obtain historical daily electricity consumption sequences and corresponding historical meteorological data to form a training sample set; The training sample set is weighted in segments based on the time difference between the sample date and the training deadline, so that the weight of recent samples is higher than that of distant samples. The physical guidance feature set is constructed based on the weighted full training sample set; The model is trained in the candidate model pool based on the physical guidance feature set, and the best model is selected based on the evaluation metrics.

14. A wind power daily electricity generation prediction system, characterized in that, The prediction system predicts the daily electricity generation of wind power plants based on the wind power daily electricity generation prediction method according to any one of claims 1 to 13.

15. A wind power daily electricity generation prediction system, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the wind power daily power forecasting method as claimed in any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that, Includes a computer program that can be executed by a processor to perform the wind power daily power prediction method as described in any one of claims 1 to 13.

Images