New energy output prediction method and system based on sub-seasonal weather forecast

Abstract

The application discloses a new energy output prediction method and system based on sub-seasonal weather forecast, comprising: forming aligned weather input set and aligned output label set by aligning sub-seasonal weather forecast records and station operation observation records, reducing system deviation introduced by sample mismatch; obtaining a circulation modal feature vector by modal projection, combining a prediction result dispersion and a historical re-prediction statistical benchmark data set to generate a prediction reliability weight vector, screening weak predictable signals, and enhancing the recognition stability of persistent strong / weak wind and continuous cloudy / sunny hot processes; by performing probability scale reduction mapping, outputting a station local weather factor probability distribution set, making uncertainty adaptive propagation with reliability; by completing output conversion, ensuring that the output distribution is consistent with the physical constraints of the unit; outputting the weekly average, fluctuation interval and extreme low output risk index within the 2-6 week evaluation window, providing a reference for backup arrangement and risk control.

Description

Technical Field

[0001] This invention relates to the technical field of power system renewable energy output forecasting and meteorological information processing, and particularly to a renewable energy output forecasting method and system based on sub-seasonal weather forecasts. Background Technology

[0002] With the continuous growth of wind and solar power installed capacity, the uncertainty of new energy output places higher demands on grid dispatching plans, reserve capacity configuration, and inter-regional mutual assistance. Existing technologies, during the 2-6 week seasonal cycle, shift the forecasting focus from short-term local weather to slowly varying predictable signals of large-scale circulation. Ensemble forecast products generally exhibit characteristics such as weak predictable signals, high dispersion, and cumulative bias with increasing step length, making it difficult to stably identify persistent strong / weak winds, and prolonged cloudy / hot weather events at the site scale. Current common practices include ordering meteorological elements or power sequences... While methods such as establishing regional average regression models or integrating meteorological forecasts with historical power output models are feasible, shortcomings are still easily exposed: First, there are inconsistencies in the alignment of multi-source data in terms of start time, forecast step size, and station identification dimension, and mismatched samples will introduce systematic bias; Second, there is a lack of a reliability quantification mechanism that combines ensemble forecast dispersion with historical error statistics; Third, if the mapping of local meteorological factors to power output at stations is not included in operational boundary conditions such as rated power upper limit, cut-in / cut-out threshold, and nighttime irradiance threshold, the prediction results are prone to inconsistencies with physical operational constraints.

[0003] CN119067263A discloses a regional subseasonal wind energy resource assessment method based on dynamic climate models. This method focuses on the assessment and correction of regional wind energy resources. The prediction objects are mostly regional average or grid point elements, which are difficult to directly cover the output prediction under the metering boundary of new energy power stations. At the same time, its reliability is usually reflected by the overall error or correction term, and it lacks a weight expression that links the dispersion of the forecast results with the historical error statistics and establishes a corresponding relationship with the circulation mode.

[0004] CN120566429A discloses a method for predicting regional medium- and long-term renewable energy power, but this method is difficult to cover the hierarchical transmission relationship of circulation modes, local factors, unit boundaries, power output distribution, and risk indicators. If there is a lack of an alignment mechanism and a reliability weight transmission mechanism with joint constraints of reporting time, forecast step index, and station identification, problems such as unstable identification of continuous processes and indicator jumps during weekly rolling updates are likely to occur.

[0005] Given the shortcomings of existing sub-seasonal renewable energy forecasting technologies in terms of data alignment consistency, reliability quantification and transmission, site probability downscaling and unit operation constraint consistency, and weekly statistics and extreme risk output, this invention proposes a renewable energy output forecasting method based on sub-seasonal meteorological forecasts. This method involves jointly aligning sub-seasonal meteorological forecast records with renewable energy station operation observation records, performing modal projection within a preset circulation feature extraction window to obtain circulation mode feature vectors, and generating forecast reliability weight vectors corresponding to the circulation modes by combining the dispersion of forecast results with historical reforecast statistical benchmark datasets. Then, conditional probability local downscaling mapping is performed using the circulation mode feature vectors and forecast reliability weight vectors as conditions to obtain a set of local meteorological factor probability distributions for renewable energy stations. Under the constraints of the upper limit of rated power of renewable energy power generation units and operating boundary conditions, output conversion is completed to form a set of renewable energy station output probability distributions. Finally, within a 2-6 week evaluation window, weekly average output forecasts, intra-week fluctuation range forecasts, and extreme low output risk indicators are output, providing a reference for planning, reserve arrangements, and risk control. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A first aspect of the present invention provides a method for predicting new energy output based on next-season weather forecasts, comprising: Collect the next season's meteorological forecast records and the operation observation records of the new energy power station, and align them according to the unified start time, forecast step index and new energy power station identifier to obtain the aligned meteorological input set and the aligned output label set; Modal projection operation is performed on the aligned meteorological input set within a preset circulation feature extraction region window to obtain circulation modal feature vectors. Based on the dispersion of forecast results and historical reforecast statistical benchmark dataset of the sub-seasonal meteorological forecast records, a forecast reliability weight vector corresponding to the circulation modal feature vectors is generated. Using the circulation mode feature vector and the forecast reliability weight vector as conditional inputs, a conditional probability local downscaling mapping operation is performed to output a set of probability distributions of local meteorological factors at new energy power stations. Based on the upper limit constraint of the rated power of the new energy power generation unit and the operating boundary conditions, the output probability distribution set of the local meteorological factors of the new energy power station is converted to obtain the output probability distribution set of the new energy power station. Based on the statistical results of the power output probability distribution set of the new energy power station within the 2-6 week evaluation window, the weekly average power output prediction value, the weekly fluctuation range prediction value, and the extreme low power output risk index are generated and output.

[0009] A second aspect of the present invention provides a new energy output prediction system based on sub-seasonal weather forecasts, comprising: The data acquisition module is used to collect next season's weather forecast records and new energy power station operation observation records; The alignment processing module is used to align the sub-seasonal meteorological forecast records and the new energy power station operation observation records according to a unified start time, forecast step index and new energy power station identifier, so as to obtain an aligned meteorological input set and an aligned output label set. The circulation mode extraction and weight generation module is used to perform modal projection operation on the aligned meteorological input set within a preset circulation feature extraction region window to obtain circulation mode feature vectors, and to generate a forecast reliability weight vector corresponding to the circulation mode feature vectors based on the forecast result dispersion and historical reforecast statistical benchmark dataset of the sub-seasonal meteorological forecast records. The local downscaling probability mapping module is used to take the circulation mode feature vector and the forecast reliability weight vector as conditional inputs, perform conditional probability local downscaling mapping operation, and output the set of probability distributions of local meteorological factors of new energy power stations. The output probability conversion module is used to convert the output probability distribution set of the local meteorological factors of the new energy power station according to the upper limit constraint of the rated power of the new energy power generation unit and the operating boundary conditions, so as to obtain the output probability distribution set of the new energy power station. The evaluation window statistical output module is used to generate and output the weekly average output prediction value, the weekly fluctuation range prediction value, and the extreme low output risk index based on the statistical results of the output probability distribution set of the new energy power station within a 2-6 week evaluation window.

[0010] A third aspect of the present invention provides a computer device comprising: one or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the new energy output prediction method based on sub-seasonal weather forecasts described above.

[0011] A fourth aspect of the present invention provides a computer-readable medium for storing software, the software including instructions executable by one or more computers, the instructions causing the one or more computers to perform operations, the operations including the flow of the aforementioned method for predicting new energy output based on sub-seasonal weather forecasts.

[0012] The beneficial effects of this invention are as follows: This invention aligns the sub-seasonal meteorological forecast records and station operation observation records according to the start time, forecast step index, and station identifier to form an aligned meteorological input set and an aligned output label set, reducing the systematic bias introduced by sample mismatch; it obtains the circulation mode feature vector by performing modal projection on the aligned meteorological input set within the circulation feature extraction area window, and generates a forecast reliability weight vector by combining the dispersion of forecast results and the historical reforecast statistical benchmark dataset, thereby filtering weakly predictable signals and enhancing the stability of identifying persistent strong / weak winds and continuous cloudy / sunny and hot processes; it performs probability downscaling mapping based on the circulation mode feature vector and the forecast reliability weight vector to output the probability distribution set of local meteorological factors at the station, so that uncertainty adapts to the propagation of reliability; it completes the output conversion by introducing the rated power upper limit and operating boundary conditions to ensure that the output distribution is consistent with the physical constraints of the unit; finally, it outputs the weekly average, fluctuation range, and extreme low output risk indicators within a 2-6 week evaluation window, providing a reference for planning, reserve arrangement, and risk control. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the new energy output prediction method based on sub-seasonal weather forecasts as shown in this invention. Detailed Implementation

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0015] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a method for predicting new energy output based on next-season weather forecasts, which specifically includes the following steps: S1. Collect the next season's meteorological forecast records and the operational observation records of the new energy power stations, and align them according to the unified start time, forecast step index, and new energy power station identifiers to obtain the aligned meteorological input set and the aligned output label set. Note that the following points should be noted in this step: S1.1 When each forecast start time arrives, the data acquisition module reads the start time field, forecast step index field, and meteorological element field from the next season's meteorological forecast record, and reads the observation time field, new energy station identification field, and output observation value field from the new energy power station operation observation record.

[0018] In a preferred embodiment, the weather forecast records for the next season are stored in a structured file format using ensemble forecast products, and the operation observation records of new energy power stations are recorded in the historical database of the dispatching side or the power station side SCADA database. To facilitate unified alignment, the starting time field is selected and stored in Beijing time, and the time zone is kept consistent within the same data domain.

[0019] As an example, the forecast start time is 00:00 on January 13, 2026; the start time field is 00:00 on January 13, 2026; the forecast step index field is "Step 1, Step 2...Step 42", where each step corresponds to a 6-hour forecast interval; the meteorological element field includes at least two of the following: 10-meter wind speed, 2-meter air temperature, surface air pressure, shortwave irradiance, and relative humidity, and each element is accompanied by latitude and longitude grid markers and numerical units; the observation time field is 06:00 on January 13, 2026; the new energy station identifier field is WF-037 (wind farm) or PV-112 (photovoltaic farm); the output observation value field is 48.6MW (corresponding to the average grid-connected active power), and preferably it is formed by a 5-minute moving average from 1-minute sampling to reduce the impact of short-term sampling jumps on alignment.

[0020] In a preferred embodiment, the data acquisition module is completed using programmable control and edge acquisition equipment commonly used in industrial fields, such as a Siemens SIMATIC S7-1500 series PLC (CPU 1511-1 PN) with an industrial Ethernet communication interface to read data from the SCADA / EMS side, and the edge industrial control computer parses and stores the weather forecast files; the industrial control computer is an Advantech UNO-2484G or equivalent industrial computer, used to perform timed fetching, field verification, data storage and index generation.

[0021] S1.2. Generate the forecast target time based on the forecast start time field and the forecast step size index field, and use the forecast target time and the new energy power station identifier field as the joint key to filter the output observation value field corresponding to the forecast target time from the new energy power station operation observation records to obtain the aligned output label set.

[0022] In a preferred embodiment, the forecast target time is determined by the start time field and the forecast step index field. For example, when the forecast step index is step 1, the forecast target time is 6 hours after the start time; when the forecast step index is step 2, the forecast target time is 12 hours after the start time, and so on.

[0023] In a preferred embodiment, when selecting the force observation value field from the new energy power station operation observation records after forming a joint key with the predicted target time and the new energy power station identification field, the following filtering rules can be adopted: In the observation records of new energy power stations, the observation sequence of the corresponding station is located by the new energy power station identification field; in the observation sequence, the observation time field that is consistent with the forecast target time is retrieved by timestamp; if the observation time field is not completely consistent with the forecast target time (for example, the observation is a 5-minute grid while the forecast target time is an hour), the nearest neighbor matching method with a time difference of no more than half an observation interval is adopted: that is, the observation record with the smallest time difference is selected within one observation interval before and after the forecast target time; when no observation record that meets the time difference constraint is matched, the joint key entry is marked as missing and removed from the aligned output label set to maintain the stability of the one-to-one correspondence of the training samples.

[0024] S1.3 Using the forecast target time, forecast step size index field and new energy power station identification field as indexes, write the meteorological element fields in the next season's meteorological forecast records into alignment entries and generate an aligned meteorological input set, wherein the alignment entries and the aligned output label set correspond one-to-one by the joint key.

[0025] In a preferred embodiment, the alignment entries are organized using the forecast target time, forecast step length index field, and new energy power station identification field as index fields, and are stored in row format; each row corresponds to a meteorological input and output label pairing sample for a certain forecast step length and a certain power station under a single forecast.

[0026] Specifically, using the reporting start time field as the outer loop, all forecast step size indices under the same reporting start time are traversed step by step; a corresponding forecast target time is generated for each forecast step size index; the identifier field of each new energy power station is traversed, and the power output observation value field corresponding to the power station at the forecast target time is read and written into the corresponding entry of the aligned power output label set; the grid meteorological element field corresponding to the forecast target time in the next season's meteorological forecast record is read simultaneously, and each element is written into the meteorological element field area of ​​the aligned entry in a unified order; only when there is a label record with the same composite key as the aligned entry in the aligned power output label set is the aligned meteorological input set confirmed to be written into the aligned meteorological input set, so as to ensure that the two correspond one-to-one according to the composite key.

[0027] It should be noted that the reason for aligning the sub-seasonal meteorological forecast records and the new energy power station operation observation records with a unified start time, forecast step size index, and new energy power station identifier in S1 is that the error sources of sub-seasonal scale (2-6 weeks) prediction include not only local meteorological uncertainties but also statistical distribution drift caused by differences in start batches and forecast step sizes. Through alignment, subsequent steps S2-S5 can establish sample associations for the same start time, the same step size, and the same power station under the same joint key constraint, thereby stabilizing the correspondence between circulation modal characteristics and power station output labels and avoiding mixing statistical characteristics from different start batches or different step sizes into the same training / evaluation set. Compared with the existing technology that directly regresses output based solely on daily or hourly forecasts, this embodiment retains the start time and step size index in S1, which facilitates the subsequent consistency processing of reliability weights and weekly aggregate statistics, thereby improving the accuracy of weekly scale prediction results.

[0028] S2. Perform modal projection operation on the aligned meteorological input set within the preset circulation feature extraction region window to obtain the circulation modal feature vector. Then, based on the dispersion of forecast results from the next season's meteorological forecast records and the historical reforecast statistical benchmark dataset, generate a forecast reliability weight vector corresponding to the circulation modal feature vector. Note that the following should be noted in this step: S2.1 Based on the latitude and longitude grid point identifiers in the aligned meteorological input set, the corresponding grid point meteorological element fields are extracted according to the preset latitude and longitude boundary range to obtain the window meteorological element matrix within the preset circulation feature extraction area window.

[0029] In a preferred embodiment, the latitude and longitude grid marker consists of a latitude serial number, a longitude serial number, and a grid resolution marker, wherein the grid resolution marker is used to distinguish products with different grid spacings, such as 1° and 0.5°; the preset latitude and longitude boundary range is a rectangular window that can cover the main circulation background area affecting the target wind farm, such as 15° to 60° north latitude and 70° to 140° east longitude, to cover the East Asian monsoon region and its upstream transmission channels; for coastal wind farms, the upper limit of east longitude can be extended to 160° to cover the position changes of the Northwest Pacific subtropical high.

[0030] Within this window, the meteorological element matrix is ​​organized in three dimensions according to element type, latitude grid, and longitude grid, and forms a set of matrix slices under the same forecast step index; among them, the element type includes at least one of 10-meter wind speed, 500 hPa geopotential height, or sea level pressure to characterize the circulation pattern.

[0031] S2.2 Normalize the window meteorological element matrix according to element type and grid identifier, and rearrange the window meteorological element matrix according to the forecast step index to obtain the window feature matrix that corresponds one-to-one with the forecast step index.

[0032] In a preferred embodiment, when normalizing the window meteorological element matrix, normalization benchmarks are set according to element types, and the normalization benchmarks are derived from the statistics of windows in the same region and season in the historical reforecast statistical benchmark dataset, so that different elements can be compared in terms of numerical magnitude. After normalization, the two-dimensional grid points are expanded into a one-dimensional sequence according to the fixed scanning order of the grid point identifiers (e.g., latitude first and longitude second), and different element types are spliced ​​in sequence to obtain the window feature matrix corresponding to each forecast step index. The row dimension of the window feature matrix corresponds to the feature position after the element is expanded, and the column dimension corresponds to different element types.

[0033] Furthermore, when rearranging the window meteorological element matrix according to the forecast step length index, multiple step lengths at the same reporting time are arranged into a sequence according to time, so that subsequent projection calculations can independently obtain the circulation mode feature vector at each step length.

[0034] S2.3. Perform projection operation on the window feature matrix and the preset circulation mode basis vector matrix to obtain the circulation mode feature vector corresponding to the prediction step size index.

[0035] In a preferred embodiment, the preset circulation mode basis vector matrix is ​​obtained by performing mode decomposition on the historical re-prediction statistical benchmark dataset within a preset circulation feature extraction region window. The mode decomposition uses an empirical orthogonal function to obtain several principal modes, and the top few modes (e.g., 10 to 30) that ensure coverage of the main variance contributions are fixedly retained. The structure of the preset circulation mode basis vector matrix is ​​as follows: each column corresponds to the spatial morphological basis vector of a mode, and each row corresponds to a feature position after the window feature matrix is ​​expanded, so that the window feature matrix is ​​projected onto the basis to obtain feature vector entries arranged by mode index.

[0036] When performing projection operations, the window feature matrix corresponding to each prediction step index is projected and the circulation mode feature vector corresponding to the prediction step index is output, so that it can be directly used as the condition input for the subsequent S3.

[0037] S2.4 Perform projection operations consistent with the preset circulation mode basis vector matrix on different forecast results of the next season's meteorological forecast records at the same start time and under the same forecast step index to obtain a set of projection coefficients arranged by mode index, and calculate the dispersion of the forecast results from the set of projection coefficients.

[0038] Specifically, under the same forecast start time and the same forecast step size index, for each forecast result in different forecast results, a projection coefficient entry arranged by mode index is obtained based on the projection operation consistent with the preset circulation mode basis vector matrix. The projection coefficient entries of each forecast result under the same mode index are then aggregated to obtain the projection coefficient subset corresponding to that mode index. The projection coefficient subset corresponding to each mode index is sorted, and the lower quartile value and upper quartile value of the projection coefficient subset are determined respectively. The difference between the upper quartile value and the lower quartile value is used as the mode dispersion value corresponding to that mode index. The mode dispersion values ​​corresponding to each mode index are arranged by mode index to form a forecast result dispersion sequence, and the forecast result dispersion sequence is used as the forecast result dispersion.

[0039] S2.5. Based on the start time and forecast step size index, read the error statistics corresponding to the modality index from the historical re-forecast statistical benchmark dataset to obtain a set of historical error statistics arranged by modality index.

[0040] In a preferred embodiment, the historical reforecasting statistical baseline dataset includes at least: the ensemble reforecasting projection coefficient sequence under multiple historical reporting batches, the corresponding projection coefficient sequence, and the error statistics grouped by month.

[0041] Error statistics should include at least: systematic bias statistics under the same modality index (e.g., long-term mean offset), absolute error scaling statistics (e.g., root mean square error statistics), and discrete representations of error growth curves as the forecast step size changes.

[0042] During reading, the system uses a triple index—starting season category, forecast step size index, and modality index—to locate the corresponding set of error statistics.

[0043] S2.6. The dispersion of the forecast results and the set of historical error statistics are fused and calculated according to the modal index to obtain the reliability weight value corresponding to each mode. After applying upper and lower limit constraints to the reliability weight value, a forecast reliability weight vector is formed.

[0044] In a preferred embodiment, the fusion calculation follows the consistency rule that the greater the dispersion, the lower the reliability, and the greater the historical error, the lower the reliability. Specifically, the dispersion sequence and the set of historical error statistics are first mapped to the reliability scoring interval of the same dimension, and then synthesized according to the fusion ratio to obtain the reliability weight value corresponding to each mode. In order to avoid excessive divergence in the subsequent downscaling distribution due to excessively small weights or excessive convergence due to excessively large weights in extreme cases, upper and lower limits are imposed on the reliability weight values. For example, the weights are restricted to a closed interval of 0.1 to 0.9, and the out-of-bounds parts are truncated back to the boundary value to finally form the forecast reliability weight vector.

[0045] It should be noted that in this preferred embodiment, S2 obtains a circulation mode feature vector that can represent the large-scale circulation state by performing modal projection within a preset circulation feature extraction region window. This is because the main predictable signals of sub-seasonal scale new energy output often originate from the persistence and phase transition of large-scale circulation patterns. At the same time, by fusing the ensemble dispersion under the same reporting step size with the historical re-prediction error statistics to form a forecast reliability weight vector, the subsequent S3 can distinguish the current strong / weak predictability modal contributions during local downscaling, avoiding treating low-confidence modes equally. Compared with the existing technology that only directly inputs and outputs a single definite value for the original grid point elements, this embodiment introduces predictability constraints at the circulation level, so that the subsequent output expresses the source and scale of uncertainty in the form of a probability distribution, which facilitates maintaining statistical consistency during weekly-scale aggregation.

[0046] S3. Using the circulation mode feature vector and the forecast reliability weight vector as input conditions, perform conditional probability local downscaling mapping operation to output the set of local meteorological factor probability distributions for new energy power plants. Note that the following points should be noted in this step: S3.1. Based on the identification of new energy power stations, read the latitude, longitude and altitude information of the power stations, and combine the circulation mode feature vector and the forecast reliability weight vector to generate the power station condition feature entries.

[0047] In a preferred embodiment, when reading the latitude, longitude and altitude information of a new energy power station based on its identifier, it is preferable to read the three fields of longitude of the station center point, latitude of the station center point, and altitude from the station static parameter database, along with the unit type or component type identifier.

[0048] As an example, the center point of wind farm WF-037 can be located at latitude 38.25°N, longitude 118.60°E, and altitude 35m; the solar farm PV-112 can be located at latitude 36.10°N, longitude 116.90°E, and altitude 120m.

[0049] When generating site condition feature entries by combining circulation mode feature vectors and forecast reliability weight vectors, it is preferable to concatenate the three types of information into a single condition record in a fixed field order: the first part is the site static field, the middle part is the sequence of circulation mode feature vector entries, and the last part is the sequence of forecast reliability weight vector entries arranged in the same dimension. The forecast step size index field is also written in so that the network can distinguish the location within different weeks.

[0050] S3.2 Input the station condition feature items into the conditional probability parameter generation network, and output the set of local meteorological factor distribution parameters corresponding to the new energy station identifier. The set of local meteorological factor distribution parameters includes mean parameters and dispersion parameters.

[0051] In a preferred embodiment, the conditional probability parameter generation network is constructed based on the conditional density parameterization of a deep neural network. Its input is the station conditional feature entries, and its output is the set of distribution parameters corresponding to the local meteorological factors of the new energy station.

[0052] Specifically, local meteorological factors include at least one of the following: equivalent wind speed at hub height and wind direction stability indication for wind farms, and at least one of the following of the following: horizontal irradiance at the site and module temperature indication for photovoltaic farms.

[0053] Specifically, the structure of the conditional probability parameter generation network includes: Feature embedding layer: Discrete / continuous feature embedding and scale unification are performed on the station static fields and the forecast step size index field; Conditional fusion layer: The embedded static features are concatenated with the circulation modal feature vector and the prediction reliability weight vector, and the conditional expression is extracted through several fully connected layers or gated fusion layers; Parameter output header: Outputs the mean parameter and dispersion parameter of each local meteorological factor respectively, and applies a positive value constraint to the dispersion parameter during output to ensure the validity of the distribution parameter.

[0054] In a preferred embodiment, the training samples of the conditional probability parameter generation network are derived from S1-aligned meteorological inputs and historical local meteorological observations, and cross-validation is performed by grouping by station to avoid overestimation caused by spatial correlation between different stations.

[0055] S3.3. Perform weighted modulation on the dispersion parameter according to the forecast reliability weight vector to obtain the weighted modulation dispersion parameter, wherein the dispersion parameter corresponding to the mode with smaller weight takes a larger value, and the dispersion parameter corresponding to the mode with larger weight takes a smaller value.

[0056] In this embodiment, the mode refers to each circulation main mode entry obtained by S2 projection. For example, the first mode corresponds to the main change in large-scale potential height, the second mode corresponds to the meridional circulation oscillation, and the third mode corresponds to the subtropical high position shift.

[0057] For example, if a wind farm has a reliability weight of 0.8 for the first mode and a reliability weight of 0.2 for the third mode under a certain forecast step index, then the modulation method for the dispersion parameter is as follows: the dispersion parameter corresponding to the first mode is compressed to a smaller scale, and the dispersion parameter corresponding to the third mode is amplified to a larger scale, so that the probability distribution of the local meteorological factors formed in the end is more dispersed when the low reliability mode dominates and more concentrated when the high reliability mode dominates.

[0058] S3.4 Combine the mean parameter with the weighted modulation dispersion parameter to form a set of probability distributions of local meteorological factors at new energy power stations.

[0059] Specifically, the parameters are organized according to a set of distribution parameters corresponding to each local meteorological factor, and each set of parameters is accompanied by a new energy station identification field and a forecast step index field. If there are multiple local meteorological factors, a set of parameters with multi-factor joint distribution is formed.

[0060] In this preferred embodiment, S3 uses the circulation mode feature vector and the forecast reliability weight vector as conditional inputs to perform conditional probability local downscaling mapping because the grid information of the sub-seasonal forecast has significant spatial scale differences at the station scale. Directly taking the nearest grid point or simple interpolation often cannot characterize the local offset caused by differences in topography, underlying surface, and unit height. The network outputs a set of probability distributions of local meteorological factors through conditional probability parameters, and modulates the dispersion parameter with reliability weights so that the dispersion of the local distribution adapts to the predictability, thereby providing an input basis consistent with uncertainty for the probability output conversion in S4. Compared with downscaling schemes that only output a single definite value, this embodiment outputs a probability distribution and introduces reliability modulation, so that the weekly scale risk index can be directly obtained from the distribution statistics.

[0061] S4. Based on the rated power upper limit constraint and operating boundary conditions of the new energy power generation unit, perform output conversion on the probability distribution set of local meteorological factors of the new energy power station to obtain the output probability distribution set of the new energy power station. Note that the following should be noted in this step: S4.1. Based on the new energy power station identification, read the parameter records of the new energy power generation unit to obtain the rated power upper limit parameter and the operating boundary parameter. The rated power upper limit parameter is the grid-connected rated power value of the new energy power generation unit. The operating boundary parameter includes at least the cut-in wind speed threshold (e.g., 3 m / s) and cut-out wind speed threshold (e.g., 25 m / s) for wind power, and the nighttime irradiance threshold (e.g., 20 W / m²) for photovoltaic power. 2 ).

[0062] It should be noted that the wind speed threshold for wind power integration is taken from the wind speed point on the unit's power curve at which it begins to stably generate electricity on the grid, and is determined after eliminating frequently starting and stopping intervals in low wind speed areas based on historical operating statistics; the wind speed threshold for wind power disconnection is taken from the safe shutdown wind speed setting of the unit, and is set to not exceed the upper limit of the unit's allowed continuous operation based on local extreme wind event protection strategies; the photovoltaic nighttime irradiance threshold is taken from the irradiance level corresponding to the minimum DC input power allowed for grid connection of the inverter, and is set to the upper limit of the irradiance interval that keeps the DC power below the grid connection threshold for a long period of time, based on the module temperature and shading conditions in the early morning / evening, in order to avoid generating false non-zero output estimates in low irradiance intervals.

[0063] S4.2. Perform boundary gating processing on the probability distribution set of local meteorological factors of new energy power stations according to the operation boundary parameters to obtain the boundary gating meteorological distribution set. The boundary gating processing includes at least zero output for wind power when the wind speed is lower than the cut-in wind speed threshold or higher than the cut-out wind speed threshold, and zero output for photovoltaic power when the irradiance is lower than the nighttime irradiance threshold.

[0064] S4.3. Perform segmented power output conversion on the boundary gated meteorological distribution set according to the type of new energy power generation unit to obtain the intermediate power output distribution set. The segmented power output conversion is as follows: wind power generates segmented power distribution based on wind speed segmented intervals, and photovoltaic generates segmented DC power distribution based on irradiance and temperature and converts it into AC power distribution through inverter efficiency parameters.

[0065] In a preferred embodiment, the new energy power generation unit types include at least wind power generation units and photovoltaic power generation units; segmented output conversion is performed for different types, and the probability distribution of local meteorological factors is mapped to an output probability distribution. Wherein: (1) Wind power segmentation (power curve segmentation) The wind speed is divided into four segments: below the cut-in speed, from the cut-in speed to the rated wind speed, from the rated wind speed to the cut-out speed, and above the cut-out speed. Power is mapped to each segment. in, For wind power units at wind speeds of The converted value of active power at that time. For the local wind speed factor, To cut off the wind speed threshold, The rated wind speed threshold, To cut off the wind speed threshold, This is the reference value for converting the rated power of wind power units; (2) Photovoltaic segmented DC to AC power conversion (irradiance and temperature) First, the DC power is obtained based on irradiance and temperature, then the inverter efficiency is considered to convert it to AC power: in, This is the converted value of the DC power of the photovoltaic system. This is the converted value of the photovoltaic AC side power. For the local irradiance factor; For the local temperature factor; The nominal power of a component or array under standard test conditions; This is the baseline irradiation value under standard test conditions; This is the temperature reference value under standard test conditions; Temperature coefficient; These are inverter efficiency parameters.

[0066] For example, photovoltaic fields take , , , per degree Celsius, Local radiation factor 600W / m 2 When the local temperature factor T is 35°C, the corresponding AC power conversion value falls in the middle range below the rated power.

[0067] S4.4 Apply the upper limit truncation of the rated power upper limit parameter to the intermediate power output distribution set to obtain the power output probability distribution set of new energy power plants.

[0068] Specifically, the power portion exceeding the grid-connected rated power value is uniformly mapped to the rated power value, and the corresponding probability mass is accumulated into the rated power state bucket, thereby obtaining a set of output probability distributions of new energy power plants that meet the upper limit constraint of rated power.

[0069] In this preferred embodiment, S4 converts the set of local meteorological factor probability distributions into a set of output probability distributions under the constraints of rated power upper limit and operating boundary conditions. This is because the power output of the power station is not only driven by meteorology, but also affected by engineering constraints such as unit safety boundary, grid-connected rated power and inverter efficiency. If only the meteorological distribution is statistically analyzed without gating and truncation, it will lead to results that do not conform to engineering facts, such as exceeding the rated power or non-zero power output at night. Compared with the existing technology that directly regresses the weekly average power output using empirical coefficients, this embodiment introduces equipment boundary and power curve constraints at the probability level, so that the output distribution reflects both predictability and engineering feasibility, and can directly support the probability calculation of extreme low output risk.

[0070] S5. Based on the statistical results of the power output probability distribution set of new energy power plants within a 2-6 week evaluation window, generate and output the weekly average power output forecast, the intra-week fluctuation range forecast, and the extreme low power output risk indicator. Note that the following should be noted in this step: S5.1 Based on the 2-6 week evaluation window, the power output probability distribution set of new energy power plants is divided into weekly power output probability distribution subsets according to natural weeks. The power output probability distribution corresponding to each forecast step index in each week is weighted by time and aggregated weekly to obtain the weekly aggregated power output probability distribution set.

[0071] In a preferred embodiment, the 2-6 week evaluation window starts from the natural week corresponding to the reporting start time and covers the forecast target time set for 2 to 6 consecutive weeks; when dividing the power output probability distribution set of new energy power plants according to natural weeks, the boundary of a natural week is from Monday 00:00 to Sunday 24:00, and the power output probability distribution corresponding to each forecast step index falling within the range of that week is assigned to the power output probability distribution subset of the same week.

[0072] Furthermore, when performing weekly aggregation calculations on the output probability distributions corresponding to each forecast step length index within each week, the time coverage length represented by each forecast step length is used as the weight. For example, a 6-hour step length corresponds to a weight of 6 hours. Weighted mixing is then performed on all step length distributions within the week to obtain the weekly aggregated output probability distribution.

[0073] For example, the mathematical formula for performing weekly aggregation operations is as follows: in, The probability density function of the weekly aggregate output force for a given natural cycle; The value is taken as the output force; This is the set of forecast step size indices that fall within this natural cycle; The forecast step size index is The output probability density function corresponding to the time; Step size Corresponding time weights; Step size The corresponding time coverage length.

[0074] S5.2 Calculate the weekly average output prediction value, the intra-week fluctuation range prediction value, and the extreme low output risk index for the weekly aggregate output probability distribution set. The weekly average output prediction value is the expected value of the weekly aggregate output probability distribution, the intra-week fluctuation range prediction value is the interval corresponding to the upper and lower quantiles of the weekly aggregate output probability distribution, and the extreme low output risk index is the probability that the weekly aggregate output probability distribution is lower than the preset low output threshold.

[0075] As an example, the mathematical formula for calculating the weekly average output forecast is: in, This is the predicted weekly average power output for that natural cycle. The value is taken as the output force; Let be the probability density function of the weekly aggregate output force for this natural cycle.

[0076] As an example, the mathematical formula for calculating the predicted value of the weekly fluctuation range is: in, This is the predicted value for the intra-week fluctuation range of this natural week; To achieve the cumulative probability The corresponding output quantile; To achieve the cumulative probability The corresponding output quantile; This represents the probability value of the lower quantile (e.g., 0.10). This represents the probability value of the upper quantile (e.g., 0.90).

[0077] As an example, the mathematical formula for calculating the risk index of extremely low output is: in, This serves as an indicator of the extreme low output risk during that natural cycle. This is a preset low output threshold.

[0078] For example, a preset low output threshold is used. The power can be set to 15% of the grid-connected rated power value; for example, when the grid-connected rated power value of a wind farm is 100MW, the preset low output threshold is set to 15MW; when the grid-connected rated power value of a photovoltaic farm is 80MW, the preset low output threshold is set to 12MW.

[0079] In this preferred embodiment, S5 outputs the weekly average output prediction value, the intra-week fluctuation range prediction value, and the extreme low output risk index based on the weekly aggregated output probability distribution within a 2-6 week evaluation window. This is because the core focus of the next season's business evaluation is not a single point in time, but rather the power level, fluctuation range, and supply guarantee risk managed on a natural week basis. By aggregating the step distribution weekly with time weights, the step contribution of different time coverage lengths becomes comparable. Compared with existing technologies that only provide a weekly average value, this embodiment simultaneously provides the fluctuation range and the probability of low output, enabling the dispatching side to simultaneously evaluate the average level, fluctuation amplitude, and extreme risk under the same output system, while maintaining consistency with the aforementioned reliability weights and boundary constraints, thus reducing inconsistencies caused by subsequent manual secondary corrections.

[0080] In applying the above embodiments, other aspects of the present invention also propose a new energy output prediction system based on sub-seasonal weather forecasts, including: The data acquisition module is used to collect next season's weather forecast records and new energy power station operation observation records; The alignment processing module is used to align the next season's meteorological forecast records and the new energy power station operation observation records according to a unified start time, forecast step index and new energy power station identifier, so as to obtain an aligned meteorological input set and an aligned output label set. The circulation mode extraction and weight generation module is used to perform modal projection operation on the aligned meteorological input set within a preset circulation feature extraction region window to obtain circulation mode feature vectors, and generate forecast reliability weight vectors corresponding to the circulation mode feature vectors based on the forecast result dispersion of the next season's meteorological forecast records and the historical reforecast statistical benchmark dataset. The local downscaling probability mapping module is used to take the circulation mode feature vector and the forecast reliability weight vector as conditional inputs, perform conditional probability local downscaling mapping operation, and output the set of probability distributions of local meteorological factors of new energy power stations. The output probability conversion module is used to convert the output probability distribution set of local meteorological factors of new energy power generation units according to the upper limit constraint of rated power and the operating boundary conditions, so as to obtain the output probability distribution set of new energy power generation units. The evaluation window statistical output module is used to generate and output the weekly average output forecast, the weekly fluctuation range forecast, and the extreme low output risk index based on the statistical results of the power output probability distribution set of new energy power plants within a 2-6 week evaluation window.

[0081] Other aspects disclosed in the embodiments of the present invention also provide a computer device including one or more processors and a memory.

[0082] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the renewable energy output prediction method based on sub-seasonal weather forecasts described in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0083] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the renewable energy output prediction method based on sub-seasonal weather forecasts of the foregoing embodiments, particularly... Figure 1 The flowchart of the method is shown.

[0084] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0085] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0086] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if required, the program can be implemented in assembly or machine language.

[0087] In any case, the language can be either compiled or interpreted.

[0088] Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit.

[0089] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0090] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0091] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0092] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0093] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting new energy output based on next-season weather forecasts, characterized in that, include: Collect the next season's meteorological forecast records and the operation observation records of the new energy power station, and align them according to the unified start time, forecast step index and new energy power station identifier to obtain the aligned meteorological input set and the aligned output label set; Modal projection operation is performed on the aligned meteorological input set within a preset circulation feature extraction region window to obtain circulation modal feature vectors. Based on the dispersion of forecast results and historical reforecast statistical benchmark dataset of the sub-seasonal meteorological forecast records, a forecast reliability weight vector corresponding to the circulation modal feature vectors is generated. The process of obtaining the circulation mode feature vector includes: based on the latitude and longitude grid point identifiers in the aligned meteorological input set, extracting the corresponding grid point meteorological element fields according to a preset latitude and longitude boundary range to obtain a window meteorological element matrix within the preset circulation feature extraction area window; performing normalization processing on the window meteorological element matrix according to the element type and grid point identifier, and rearranging the window meteorological element matrix according to the forecast step index to obtain a window feature matrix that corresponds one-to-one with the forecast step index; and performing a projection operation on the window feature matrix and the preset circulation mode basis vector matrix to obtain the circulation mode feature vector corresponding to the forecast step index. Generating the forecast reliability weight vector includes: performing a projection operation consistent with the preset circulation mode basis vector matrix on different forecast results of the sub-seasonal weather forecast records under the same start time and the same forecast step index, obtaining a set of projection coefficients arranged by mode index, and calculating the dispersion of the forecast results from the set of projection coefficients; based on the start time and the forecast step index, reading the error statistics corresponding to the mode index from the historical reforecast statistical benchmark dataset, obtaining a set of historical error statistics arranged by the mode index; fusing the dispersion of the forecast results and the set of historical error statistics according to the mode index to obtain the reliability weight value corresponding to each mode, and applying upper and lower limit constraints to the reliability weight value to form the forecast reliability weight vector; Using the circulation mode feature vector and the forecast reliability weight vector as conditional inputs, a conditional probability local downscaling mapping operation is performed to output a set of probability distributions of local meteorological factors at new energy power stations. The output set of probability distributions of local meteorological factors for new energy power stations includes: reading the latitude, longitude, and altitude information of the power station based on the new energy power station identifier, and generating station condition feature entries by combining the circulation mode feature vector and the forecast reliability weight vector; inputting the station condition feature entries into a conditional probability parameter generation network to output a set of local meteorological factor distribution parameters corresponding to the new energy power station identifier, wherein the set of local meteorological factor distribution parameters includes a mean parameter and a dispersion parameter; performing weighted modulation on the dispersion parameter according to the forecast reliability weight vector to obtain a weighted modulation dispersion parameter, wherein the dispersion parameter corresponding to the mode with a smaller weight takes a larger value, and the dispersion parameter corresponding to the mode with a larger weight takes a smaller value; combining the mean parameter and the weighted modulation dispersion parameter to form a set of probability distributions of local meteorological factors for new energy power stations. Based on the upper limit constraint of the rated power of the new energy power generation unit and the operating boundary conditions, the output probability distribution set of the local meteorological factors of the new energy power station is converted to obtain the output probability distribution set of the new energy power station. Based on the statistical results of the power output probability distribution set of the new energy power station within the 2-6 week evaluation window, the weekly average power output prediction value, the weekly fluctuation range prediction value, and the extreme low power output risk index are generated and output.

2. The method for predicting new energy output based on sub-seasonal weather forecasts according to claim 1, characterized in that, The obtained aligned meteorological input set and aligned output label set include: When each forecast start time arrives, the data acquisition module reads the start time field, forecast step index field, and meteorological element field from the next season's meteorological forecast record, and reads the observation time field, new energy station identification field, and output observation value field from the new energy station operation observation record. Based on the starting time field and the forecast step index field, a forecast target time is generated. Using the forecast target time and the new energy power station identifier field as a joint key, the output observation value field corresponding to the forecast target time is filtered from the operation observation records of the new energy power station to obtain the aligned output label set. Using the forecast target time, the forecast step size index field, and the new energy power station identification field as indexes, the meteorological element fields in the sub-season meteorological forecast record are written into alignment entries and the alignment meteorological input set is generated, wherein the alignment entries and the alignment output label set correspond one-to-one according to the joint key.

3. The method for predicting new energy output based on sub-seasonal weather forecasts according to claim 1, characterized in that, The obtained set of power output probability distributions for new energy power plants includes: Based on the new energy power station identification, the parameter records of the new energy power generation unit are read to obtain the rated power upper limit parameter and the operating boundary parameter. The rated power upper limit parameter is the grid-connected rated power value of the new energy power generation unit. The operating boundary parameter includes at least the cut-in wind speed threshold and cut-out wind speed threshold of wind power, and the nighttime irradiance threshold of photovoltaic power. Based on the operational boundary parameters, the probability distribution set of local meteorological factors of the new energy power station is subjected to boundary gating processing to obtain a boundary gating meteorological distribution set. The boundary gating processing includes at least zero output of wind power when the wind speed is lower than the cut-in wind speed threshold or higher than the cut-out wind speed threshold, and zero output of photovoltaic power when the irradiance is lower than the nighttime irradiance threshold. The boundary-gated meteorological distribution set is subjected to segmented power output conversion according to the type of new energy power generation unit to obtain the intermediate power output distribution set. The segmented power output conversion is that wind power generates segmented power distribution according to wind speed segmented intervals, and photovoltaic generates segmented DC power distribution according to irradiance and temperature and converts it into AC power distribution through inverter efficiency parameters. By applying the upper limit truncation of the rated power upper limit parameter to the intermediate power output distribution set, the power output probability distribution set of new energy power plants is obtained.

4. The method for predicting new energy output based on sub-seasonal weather forecasts according to claim 3, characterized in that, The generation and output of the weekly average power output forecast, the intra-week fluctuation range forecast, and the extreme low power output risk indicator include: Based on the 2-6 week evaluation window, the set of output probability distributions of the new energy power stations is divided into weekly output probability distribution subsets according to the natural week. The output probability distributions corresponding to each forecast step index in each week are then aggregated weekly by time weighting to obtain the weekly aggregated output probability distribution set. For the set of weekly aggregated output probability distributions, calculate the predicted value of the weekly average output, the predicted value of the intra-week fluctuation range, and the risk index of extreme low output. The predicted value of the weekly average output is the expected value of the weekly aggregated output probability distribution, the predicted value of the intra-week fluctuation range is the interval corresponding to the upper and lower quantiles of the weekly aggregated output probability distribution, and the risk index of extreme low output is the probability that the weekly aggregated output probability distribution is lower than a preset low output threshold.

5. A new energy output prediction system based on sub-seasonal weather forecasting, based on the new energy output prediction method based on sub-seasonal weather forecasting as described in any one of claims 1 to 4, characterized in that, include: The data acquisition module is used to collect next season's weather forecast records and new energy power station operation observation records; The alignment processing module is used to align the sub-seasonal meteorological forecast records and the new energy power station operation observation records according to a unified start time, forecast step index and new energy power station identifier, so as to obtain an aligned meteorological input set and an aligned output label set. The circulation mode extraction and weight generation module is used to perform modal projection operation on the aligned meteorological input set within a preset circulation feature extraction region window to obtain circulation mode feature vectors, and to generate a forecast reliability weight vector corresponding to the circulation mode feature vectors based on the forecast result dispersion and historical reforecast statistical benchmark dataset of the sub-seasonal meteorological forecast records. The local downscaling probability mapping module is used to take the circulation mode feature vector and the forecast reliability weight vector as conditional inputs, perform conditional probability local downscaling mapping operation, and output the set of probability distributions of local meteorological factors of new energy power stations. The output probability conversion module is used to convert the output probability distribution set of the local meteorological factors of the new energy power station according to the upper limit constraint of the rated power of the new energy power generation unit and the operating boundary conditions, so as to obtain the output probability distribution set of the new energy power station. The evaluation window statistical output module is used to generate and output the weekly average output prediction value, the weekly fluctuation range prediction value, and the extreme low output risk index based on the statistical results of the output probability distribution set of the new energy power station within a 2-6 week evaluation window.

6. A computer device, characterized in that, include: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the new energy output prediction method based on sub-seasonal weather forecasts as described in any one of claims 1 to 4.

7. A computer-readable medium for storing software, characterized in that: The software includes instructions executable by one or more computers, which cause the one or more computers to perform operations, including the flow of the new energy output prediction method based on sub-seasonal weather forecasts as described in any one of claims 1 to 4.

Images