Software data processing method, system and equipment for medium and long term power generation capacity prediction and medium

Abstract

The invention discloses a software data processing method, system and device for medium and long term power generation capacity prediction and a medium, and the method comprises the steps: obtaining multi-dimensional data of new energy power generation, preprocessing the multi-dimensional data, obtaining a time sequence feature vector, dividing new energy resources, obtaining a plurality of resource subsets, distributing prediction parameters for each resource subset, establishing a multi-scene hierarchical framework of medium and long term prediction, and selecting a prediction model to perform power generation capability prediction on each resource subset to obtain a preliminary prediction value; introducing constraint conditions of the power system to construct a capability distribution matrix, and adjusting the preliminary prediction value to obtain a power generation capability prediction result; monitoring a deviation between a power generation capacity prediction result and an actual result, and triggering a corresponding correction and optimization mechanism; and when the external disturbance event is detected, safety boundary calibration is performed and the prediction curve is updated, so that the technical level and the practical application value of long-term power generation capacity prediction in new energy are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power generation capacity prediction technology, and in particular to a software data processing method, system, equipment and medium for medium- and long-term power generation capacity prediction. Background Technology

[0002] New energy sources generally refer to new clean energy sources relative to traditional fossil fuels, including solar, wind, hydro, biomass, geothermal, and ocean energy. New energy sources are characterized by their renewability, cleanliness, and low carbon emissions, making them an important direction for promoting energy transformation and achieving the "dual-carbon" goal. In the adjustment of the energy structure, the installed capacity and power generation share of new energy sources have been increasing year by year, becoming a core growth point for the future power system. Power generation capacity forecasting refers to estimating the power supply capacity that the power system can provide in a future period by analyzing power generation capacity, energy resource conditions, technological level, policy factors, and load demand. For new energy sources, power generation capacity forecasting needs to consider uncertainties such as the volatility of natural resources, equipment utilization rate, energy storage, and dispatching methods. Therefore, forecasting not only requires statistical modeling but also optimization analysis combined with the operating characteristics of the power system to provide a scientific basis for power planning and grid dispatching. In new energy power generation capacity forecasting, software data processing methods are the core supporting tool. It involves the collection, cleaning, transformation, modeling, and analysis of raw data. Typical methods include time series analysis, machine learning models, simulation, and scenario analysis. Automating data processing and forecasting through software can improve the accuracy and efficiency of forecasts, and support multi-option comparison and visualized decision-making.

[0003] In existing technologies, the prediction of medium- and long-term power generation capacity of new energy sources mainly relies on statistical methods based on single historical power generation data or time series models based on simplified meteorological factors. These methods suffer from the following problems in application: First, the data sources are singular and of low quality. Meteorological data, power operation data, and policy planning data from different sources are often not processed uniformly, leading to incomplete or even distorted inputs to the prediction models. Second, the models fail to adequately characterize the volatility and uncertainty of new energy power generation. Existing methods often employ linear regression or simplified time series algorithms, making it difficult to capture the complex nonlinear characteristics of new energy in the medium- and long-term dimensions, influenced by multiple factors such as seasonal climate patterns, policy changes, and the expansion of installed capacity. Third, there is a lack of dynamic adjustment mechanisms. Traditional methods typically generate fixed prediction results, unable to be corrected based on actual data feedback during operation. When encountering extreme weather events, energy policy adjustments, or abnormal fluctuations in electricity demand, prediction errors can easily accumulate and amplify. Finally, systemic constraints, such as grid transmission capacity, energy storage regulation capacity, and the proportion of new energy consumption, are not fully considered, leading to deviations between the prediction results and the actual available capacity of the power system, affecting the scientific nature of medium- and long-term power planning and dispatch. In summary, existing technologies have significant shortcomings in terms of comprehensive data processing, accurate prediction models, dynamic adaptability, and integration of system constraints. Existing technologies suffer from incomplete data processing, insufficient model accuracy, lack of dynamic correction, and neglect of system constraints in predicting the medium- and long-term power generation capacity of new energy sources, making it difficult for the prediction results to meet the needs of long-term safe operation and planning of the power system. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] This invention provides a software data processing method, system, equipment, and medium for medium- and long-term power generation capacity prediction. It addresses the shortcomings of existing technologies in predicting medium- and long-term power generation capacity of new energy sources, such as incomplete data processing, insufficient model accuracy, lack of dynamic correction, and neglect of system constraints. These shortcomings lead to prediction results that fail to meet the needs of long-term safe operation and planning of the power system.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a software data processing method for predicting medium- and long-term power generation capacity, comprising:

[0008] Acquire multidimensional data on new energy power generation, preprocess the multidimensional data to obtain a time-series feature vector;

[0009] Based on the time-series feature vector, new energy resources are divided into multiple resource subsets. Prediction parameters are assigned to each resource subset to establish a multi-scenario hierarchical framework for medium- and long-term prediction.

[0010] Based on the multi-scenario hierarchical framework, a prediction model is selected to predict the power generation capacity of each resource subset, and a preliminary prediction value is obtained.

[0011] By introducing constraints of the power system to construct a capacity allocation matrix, the preliminary prediction values ​​are adjusted to obtain the power generation capacity prediction results.

[0012] Monitor the deviation between the predicted and actual power generation capacity, and trigger corresponding correction and optimization mechanisms based on the deviation characteristics;

[0013] When an external disturbance event is detected, safety boundary calibration is performed and the prediction curve is updated to achieve prediction of the medium- and long-term power generation capacity of new energy sources.

[0014] As a preferred embodiment of the software data processing method for medium- and long-term power generation capacity prediction described in this invention, the preprocessing of the multidimensional data includes:

[0015] The multidimensional data is processed to improve its quality, resulting in the first multidimensional data.

[0016] Align the first multidimensional data and construct a time-series feature mapping table using a multi-granularity aggregation method;

[0017] Feature variables are selected from the time-series feature mapping table, and higher-order features are constructed through feature combination and derivation methods to obtain the time-series feature vector.

[0018] As a preferred embodiment of the software data processing method for medium- and long-term power generation capacity prediction described in this invention, the establishment of a multi-scenario hierarchical framework for medium- and long-term prediction includes:

[0019] The first clustering algorithm is used to divide new energy resources into multiple resource subsets, and the differences between each subset are statistically verified.

[0020] External driving factors are identified and quantified into prediction parameters, and prediction parameters are assigned to each resource subset.

[0021] By coupling resource subsets with prediction parameters, a multi-scenario hierarchical framework for medium- and long-term prediction is constructed.

[0022] The beneficial effects of this preferred technical solution are that it can accurately characterize resource differences, flexibly adapt to changing external factors, and significantly improve the scientificity and adaptability of medium- and long-term forecasts.

[0023] As a preferred embodiment of the software data processing method for medium- and long-term power generation capacity prediction according to the present invention, the method includes: selecting a prediction model to predict the power generation capacity of each resource subset, which comprises:

[0024] Select at least one prediction model to predict the power generation capacity of each resource subset, and use the model fusion method to fuse the output results of at least one prediction model to obtain the preliminary prediction value.

[0025] External constraints are introduced to limit the preliminary predicted values.

[0026] The advantages of this preferred technical solution are that it can improve prediction accuracy, enhance model adaptability, and ensure the reliability and practicality of prediction results.

[0027] As a preferred embodiment of the software data processing method for medium- and long-term power generation capacity prediction described in this invention, the construction of the capacity allocation matrix includes:

[0028] The objective function is to maximize the total absorption benefit or minimize the total abandoned electricity.

[0029] The constraints include the upper limit of subset output, matching of total power system demand, dynamic balance of energy storage, and non-negativity and normalization;

[0030] The objective function is solved using an optimization algorithm to obtain the optimal solution for each element in the capability allocation matrix, and the capability allocation matrix is ​​constructed using the optimal solution for each element.

[0031] The capacity allocation matrix is ​​applied to the initial prediction values ​​to obtain the adjusted power generation capacity trajectory of each subset.

[0032] The adjusted power generation capacity trajectories of each subset are used as new observation data to update the time series feature mapping table and the corresponding feature vectors.

[0033] Based on the update results, trigger parameter retraining of the prediction model or iteration of prediction parameters for a subset of resources.

[0034] The beneficial effects of this preferred technical solution are that it can achieve precise power allocation, dynamic optimization and prediction, and improve the overall operating efficiency and reliability of the system.

[0035] As a preferred embodiment of the software data processing method for medium- and long-term power generation capacity prediction described in this invention, the corresponding correction and optimization mechanisms include:

[0036] Calculate at least one deviation index between the predicted and actual power generation capacity;

[0037] The deviation index is decomposed into multiple dimensions, and the corresponding correction process is triggered for different dimensions.

[0038] The deviation index is compared with a multi-level threshold, which includes at least an early warning threshold, an intervention threshold, and an emergency threshold.

[0039] When the deviation indicators trigger the intervention threshold and emergency threshold, the corresponding re-optimization strategy is activated;

[0040] The optimization strategies include retraining the prediction model, reconstructing the scheduling matrix, and calibrating subset parameters.

[0041] As a preferred embodiment of the software data processing method for medium- and long-term power generation capacity prediction described in this invention, the safety boundary calibration includes:

[0042] Obtain disturbance indices, input the disturbance indices into the security assessment model, and output trend adjustment coefficients and boundary condition correction coefficients;

[0043] The long-term power generation capacity prediction trend of the resource subset is corrected by using the trend adjustment coefficient, and the system constraints in the capacity allocation matrix are updated by using the boundary condition correction coefficient, so as to obtain the corrected prediction curve and boundary conditions.

[0044] Based on the revised prediction curve and boundary conditions, the medium- and long-term power supply security indicators are recalculated and the security margin threshold is updated to achieve the prediction of the medium- and long-term power generation capacity of new energy sources.

[0045] Secondly, the present invention provides a software data processing system for medium- and long-term power generation capacity prediction, comprising:

[0046] The preprocessing module is used to acquire multidimensional data of new energy power generation, preprocess the multidimensional data, and obtain time-series feature vectors.

[0047] The partitioning module is used to partition new energy resources based on the time-series feature vector to obtain multiple resource subsets, assign prediction parameters to each resource subset, and establish a multi-scenario hierarchical framework for medium- and long-term prediction.

[0048] The prediction module is used to select a prediction model based on the multi-scenario hierarchical framework to predict the power generation capacity of each resource subset and obtain preliminary prediction values.

[0049] The adjustment module is used to introduce the constraints of the power system to construct a capacity allocation matrix, adjust the preliminary prediction value, and obtain the power generation capacity prediction result.

[0050] The monitoring module is used to monitor the deviation between the predicted power generation capacity and the actual results, and trigger corresponding correction and optimization mechanisms based on the deviation characteristics.

[0051] The calibration module is used to perform safety boundary calibration and update the prediction curve when an external disturbance event is detected, so as to realize the prediction of the medium and long-term power generation capacity of new energy.

[0052] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the software data processing method for predicting medium- and long-term power generation capacity.

[0053] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the software data processing method for predicting medium- and long-term power generation capacity.

[0054] Compared with existing technologies, the beneficial effects of this invention are as follows: The software data processing method for predicting the medium- and long-term power generation capacity of new energy proposed in this invention effectively overcomes the shortcomings of existing technologies through a complete process. By acquiring multi-source heterogeneous data and time-series mapping, it achieves unified collection, cleaning, and modeling of multi-dimensional data such as meteorological, operational, and policy data, ensuring the integrity and accuracy of input information. Through feature pattern recognition and scenario layering, different regions and power source types are divided into resource subsets, and differentiated prediction parameters are introduced, making the prediction more consistent with medium- and long-term resource endowments and development scenarios. Through prediction modeling and capacity allocation matrix construction, it integrates time series, machine learning, and physical simulation methods, and introduces constraints such as grid transmission and energy storage regulation, achieving dynamic matching between prediction results and system operating conditions. Through dynamic deviation correction and safety boundary calibration, it can continuously monitor and correct during prediction execution, especially when extreme weather events or policy adjustments occur, quickly generating correction factors and adjusting prediction curves, thereby improving the long-term adaptability and robustness of prediction results. Ultimately, this invention not only improves prediction accuracy and reliability but also enhances the response capability to uncertainties and external disturbances, providing strong data support for the medium- and long-term planning, scheduling, and policy formulation of power systems. By constructing a complete closed-loop process of multi-source data processing, pattern recognition, capacity allocation, and dynamic correction, not only are the deficiencies in data comprehensiveness and model accuracy compensated for, but also the prediction results are made more scientific, robust, and adaptable through dynamic adjustment and system constraint fusion, thereby significantly improving the technical level and practical application value of new energy medium- and long-term power generation capacity prediction. Attached Figure Description

[0055] 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.

[0056] Figure 1This is a schematic diagram of the overall flow logic of a software data processing method for predicting medium- and long-term power generation capacity, provided as an embodiment of the present invention. Detailed Implementation

[0057] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0058] Example 1, referring to Figure 1 As an embodiment of the present invention, a software data processing method for predicting medium- and long-term power generation capacity is provided, comprising:

[0059] S100: Acquire multi-dimensional data of new energy power generation, preprocess the multi-dimensional data, and obtain time-series feature vectors;

[0060] S200: Based on time-series feature vectors, new energy resources are divided into multiple resource subsets, prediction parameters are assigned to each resource subset, and a multi-scenario hierarchical framework for medium- and long-term prediction is established.

[0061] S300: Based on a multi-scenario hierarchical framework, a prediction model is selected to predict the power generation capacity of each resource subset and obtain preliminary prediction values.

[0062] S400: Introduce the constraints of the power system to construct a capacity allocation matrix, adjust the preliminary prediction values, and obtain the power generation capacity prediction results;

[0063] S500: Monitors the deviation between the predicted and actual power generation capacity and triggers corresponding correction and optimization mechanisms based on the characteristics of the deviation.

[0064] S600: When an external disturbance event is detected, safety boundary calibration is performed and the prediction curve is updated to achieve prediction of the medium- and long-term power generation capacity of new energy sources.

[0065] It should be noted that through end-to-end optimization, from multi-dimensional data preprocessing to precise resource subset partitioning, and then to the integration of multi-model predictions and system constraint adjustments, dynamic deviation correction and safety boundary calibration are ultimately achieved. This not only improves prediction accuracy and adaptability but also enhances the ability to cope with uncertainties and external disturbances, providing scientific, robust, and efficient data support for long-term planning, scheduling, and policy formulation in the power system, and significantly improving the practicality and reliability of new energy power generation capacity prediction.

[0066] In this embodiment of the invention, step S100 includes the following sub-steps A1-A3;

[0067] In A1: Data quality improvement processing is performed on the multidimensional data to obtain the first multidimensional data;

[0068] In A2: Align the first multidimensional data and construct a time-series feature mapping table using a multi-granularity aggregation method;

[0069] In A3: Feature variables are selected from the time-series feature mapping table, and higher-order features are constructed through feature combination and derivation methods to obtain the time-series feature vector.

[0070] Specifically, acquiring multidimensional data on new energy power generation includes designing differentiated access strategies based on the characteristics of different data sources to ensure the integrity of three core types of information: meteorological driving factors, power system operating status, and macroeconomic policy guidance.

[0071] The meteorological observation system prioritizes access to data from national-level ground meteorological stations, high-density distributed sensors, and satellite remote sensing inversion data, with a data acquisition frequency ranging from minutes to hours. Key parameters include hourly wind speed (10 m / 2 m altitude), total solar radiation on the horizontal / tilted surface (W / m²), ambient temperature (°C), relative humidity (%), air pressure (hPa), and cloud cover (%). For example, data acquisition for a centralized photovoltaic power station requires simultaneous acquisition of the direct and diffuse radiation components from the meteorological stations in its area to correct for the actual radiation received by the modules on the tilted surface.

[0072] The database is then reanalyzed, incorporating high-precision global / regional reanalysis products to supplement the spatiotemporal continuity data of sparse weather stations. The reanalysis data is typically provided with a 0.25°×0.25° grid resolution and a time resolution of 1 hour or 3 hours, containing three-dimensional atmospheric variables and surface flux parameters. Spatial interpolation is required to match the geographic coordinates of the renewable energy power stations.

[0073] The power dispatch center connects to provincial and higher-level power grid dispatch automation systems to extract real-time / historical power operation indicators, including: load-side data; renewable energy generation-side data; and power grid topology parameters. For example, the dispatch data for a regional power grid needs to include the output curves of a cluster of wind farms and the charging and discharging status of their supporting energy storage systems.

[0074] Policy planning documents, consisting of unstructured text data, utilize Natural Language Processing (NLP) technology to extract key information, including: quantitative objectives; policy constraints; and timelines. This type of data requires entity identification and numerical correlation analysis to transform it into structured policy driving factors.

[0075] In one alternative embodiment, the data quality enhancement process can reduce data noise by decomposing the data into sub-bands of different frequencies through wavelet transform, then performing thresholding on the high-frequency sub-bands, and finally reconstructing the data through inverse wavelet transform.

[0076] In another alternative embodiment, the data quality improvement process can be data smoothing, which smooths the time series data by calculating the average value within a certain time window to reduce short-term fluctuations, giving higher weight to recent data, and smoothing the data through exponential decay.

[0077] In this embodiment of the invention, data quality improvement processing includes cleaning, completion, and standardized quality assurance.

[0078] Specifically, the raw data often contains missing values, outliers, and dimensional differences, requiring quality improvement through standardized processes.

[0079] Data cleaning involves identifying outliers based on statistical thresholds and business rules. For example, solar radiation exceeding the 99th percentile of historical data for the same period or falling below the theoretical minimum of 0 is marked as an anomaly; wind speeds exceeding 50 m / s momentarily for less than 1 minute are considered noise interference. Outlier handling employs either a moving average of nearby time periods or business logic corrections.

[0080] For interpolation completion, whether periodic or continuous, the interpolation method is selected based on the data type: continuous variables are filled using cubic spline interpolation or linear regression; discrete variables are filled using mode or manual annotation. For example, if a photovoltaic power station has missing radiation data for 3 consecutive hours, and the cloud cover data for the same period is normal, the missing value is estimated using an empirical model of radiation-cloud cover (non-linear fitting); if there are no related variables, the mean of the same hour in the previous 24 hours (daily periodic characteristics) is used for completion.

[0081] Normalization eliminates the influence of dimensions and compresses the numerical range. This is achieved using (min-maximum) Min-Max standardization (mapping the data to the [0,1] interval) or Z-Score standardization (converting to a distribution with a mean of 0 and a standard deviation of 1). For example, the original wind speed range is 0~30m / s, which, after normalization, becomes (current wind speed - historical minimum wind speed) / (historical maximum wind speed - historical minimum wind speed). The proportion of wind and solar power curtailment (typically 0%~50%) is directly processed using Min-Max to avoid extreme values ​​dominating model training.

[0082] Specifically, constructing a time-series feature mapping table using a multi-granularity aggregation method involves addressing the potential resolution differences in the original timestamps of multi-source data. This necessitates aligning with a unified time benchmark and constructing the time-series feature mapping table through multi-granularity aggregation. Using minute-level data as the smallest granularity, high-frequency data is aggregated according to target intervals, such as hourly averages, maximum values, and cumulative values. Low-frequency data is extended using timestamps and marked as static parameters. For example, minute-level wind speed data from a wind farm is averaged hourly to generate an hourly wind speed sequence, while the annual grid connection target is used as a fixed value to fill the timestamps for each hour throughout 2024. The data is stored in a two-dimensional table format, with row indices based on a unified timestamp. Column fields include: meteorological characteristics; power operation characteristics; policy characteristics; and derived characteristics. Each cell in the table corresponds to a standardized value at a specific time point, ensuring all features are aligned on the same time axis.

[0083] Constructing higher-order features through feature combination and derivation methods involves integrating key variables from the temporal feature mapping table into the final feature vector, considering spatiotemporal dependencies and business logic relevance, to form a high-dimensional sparse or dense vector. The construction logic is as follows:

[0084] The selection of basic features involves screening variables with strong relevance based on the prediction target. For example, short-term power forecasts need to include hourly wind speed, temperature, and current turbine output for the current and previous 6 hours, while long-term absorption capacity assessments need to include macroeconomic factors such as annual policy targets and regional load growth rates.

[0085] Feature combination and derivation: generating higher-order features through mathematical transformations or business rules, such as: interaction terms; sliding window statistics; periodic coding; policy impact weights.

[0086] The vectors are standardized and the resulting feature vectors are multidimensional arrays of fixed length. Each element corresponds to a normalized value at a specific timestamp, and all vectors share the same time base.

[0087] It should be noted that comprehensive data preprocessing and feature engineering effectively improved data quality, enhanced the accuracy and reliability of the model's input data, provided a solid foundation for accurate prediction of the medium- and long-term power generation capacity of new energy sources, and significantly improved prediction accuracy and model stability.

[0088] In this embodiment of the invention, step S200 includes the following sub-steps B1-B3;

[0089] In B1: The first clustering algorithm is used to divide the new energy resources into multiple resource subsets, and the differences of each subset are statistically verified.

[0090] In B2: external driving factors are acquired and quantified into prediction parameters, and prediction parameters are assigned to each resource subset;

[0091] In B3: Resource subsets are coupled with prediction parameters to construct a multi-scenario hierarchical framework for medium- and long-term predictions.

[0092] In an alternative embodiment, the first clustering algorithm can be spectral clustering, which calculates the similarity between data points, typically using a Gaussian kernel function or other similarity measures, constructs a Laplacian matrix of the graph based on the similarity matrix, performs eigenvalue decomposition on the Laplacian matrix, extracts its eigenvectors, and uses the K-means algorithm to cluster in the eigenvector space.

[0093] In another alternative embodiment, the first clustering algorithm can also be autoencoder clustering, which uses an autoencoder network to encode the data, learns a low-dimensional representation of the data, extracts low-dimensional features of the data in the encoding layer of the autoencoder, and performs clustering using a conventional clustering algorithm in the low-dimensional feature space.

[0094] In this embodiment of the invention, the first clustering algorithm includes selecting a clustering method based on data distribution characteristics;

[0095] Specifically, key features for pattern recognition are extracted from the time-series feature mapping table, including but not limited to: resource endowment features, regional average annual wind speed, annual total horizontal radiation, topographic relief, and latitude / altitude.

[0096] Fluctuation characteristics, standard deviation of historical power / output, daily maximum / minimum output ratio (reflecting intraday fluctuation range), and percentage of consecutive low output periods;

[0097] System correlation characteristics, time-series correlation coefficient with regional load, and blocking frequency of grid access points;

[0098] Static attributes include power source type, average unit capacity, and years of operation.

[0099] To reduce redundancy in high-dimensional feature spaces, principal component analysis (PCA) or t-SNE nonlinear dimensionality reduction is used to compress features. For example, PCA calculates the eigenvectors of the feature covariance matrix and selects principal components with a cumulative contribution rate > 85%, mapping the original tens-dimensional features to a low-dimensional space while retaining the main information entropy.

[0100] If the subset boundaries are clear, hard clustering methods such as K-means++ or GMM (Gaussian Mixture Model) can be used. Taking K-means++ as an example, the calculation logic is as follows: the squared Euclidean distance based on the feature vectors is expressed as:

[0101]

[0102] in, For the first The feature vector of each sample, For the first The feature vectors of each cluster center;

[0103] Initialize cluster centers, iteratively update the category of each sample and recalculate the cluster center position until the center point moves less than a threshold (e.g., 1e-5) or reaches the maximum number of iterations (e.g., 100 times). Output K cluster subsets (e.g., K=3, corresponding to the wind power areas of "high resources-low volatility", "medium resources-medium volatility", and "low resources-high volatility").

[0104] If the subset boundaries are ambiguous, DBSCAN or hierarchical clustering can be used. DBSCAN defines the neighborhood radius ( The minimum number of samples (MinPts) is used to group density-connected samples into the same cluster, while noise points are marked as a special subset.

[0105] After the partitioning is completed, statistical tests are needed to verify the significant differences in key features among the subsets. For example, the average annual wind speed of subset A is significantly higher than that of subset B. And assign a business-interpretable label to each subset. For example:

[0106] Wind data collection: "Coastal high wind speed stable zone" (annual average wind speed > 7 m / s, standard deviation of fluctuation < 15%), "Inland low wind speed intermittent zone" (annual average wind speed < 5 m / s, standard deviation of fluctuation > 25%).

[0107] Photovoltaic subsets: "Western Plateau direct radiation type" (annual total radiation > 1800 kWh / m², daily fluctuation peak occurs at noon) and "Eastern coastal diffuse type" (annual total radiation < 1400 kWh / m², daily fluctuation is significantly affected by cloud cover).

[0108] Specifically, acquiring and quantifying external driving factors into predictive parameters includes policy guidance, which involves analyzing quantitative targets and constraints from policy planning documents and converting them into subset-related parameters. For example, policy parameters for a subset of "coastal high-wind-speed stable areas" include "the upper limit of the rate of new installed capacity during the 14th Five-Year Plan period is 15% / year" and "the minimum guaranteed utilization hours are 3000 hours".

[0109] The installed capacity expansion curve is based on historical installed capacity data to fit the growth trend, and the future expansion rate range is adjusted in combination with policy objectives. For example, if historical data shows that the average annual installed capacity growth rate of a certain photovoltaic subset is 20%, but the policy requires that it "slow down to within 10% after 2025", then its installed capacity growth rate range is set as [5%, 15%] (considering market fluctuations).

[0110] Climate change scenarios, referencing representative concentration pathways or regional climate models published by the IPCC, are used to obtain the trends in wind speed / radiation over the next 30 years. This information is then converted into resource endowment correction coefficients.

[0111] Assigning prediction parameters to each resource subset involves considering both its historical characteristics and the interaction with external drivers when assigning parameters to each subset. The specific logic is as follows:

[0112] The installed capacity growth rate range. Calculated based on the slope of the historical expansion curve, the upper limit of policy targets, and the lower limit of resource constraints, and expressed as:

[0113]

[0114]

[0115] in, The slope of the historical extension curve. As the upper limit of policy objectives, As the lower limit of resource constraints, This is the deceleration factor caused by land / grid capacity limitations (e.g., 0.2 indicates a 20% reduction in growth rate due to insufficient absorption capacity). , This represents the technical feasibility boundary (e.g., the minimum growth rate of offshore wind power is limited by construction capacity and should not be less than 3% / year). For example, if the historical growth rate of a certain subset of "inland low wind speed areas" is 8% / year, the policy requires it not to exceed 12%, but land resources can only support a growth rate of 5%, then its rate range is [5%, 12%].

[0116] The technological progress correction factor, reflecting the gain of unit efficiency improvement on actual output, is extrapolated from historical technological iteration data and expressed as:

[0117]

[0118] in, The rate of technological progress (e.g., 0.005 / year, meaning an annual increase of 0.5%). The difference between the predicted year and the base year (e.g., the year corresponding to 2030). This coefficient serves as a correction factor for the upper limit of output of the subset (e.g., if the original maximum output is 100MW, the correction is 100MW × (1 + 0.005 × 7)≈ 103.5MW).

[0119] Under the upper limit of availability, the proportion of reliable operating time of a subset of devices is defined, affected by extreme weather events (such as typhoon damage to wind power and sandstorm shading of photovoltaics) and maintenance capabilities. This is expressed as:

[0120] in, This is the historical average availability rate (e.g., 95% for wind power). This is due to the decline in availability caused by climate change (e.g., under the RCP8.5 scenario, the increased frequency of typhoons reduces the lower limit of wind power availability to 93%). Maintenance optimization brought about by technological upgrades (such as intelligent inspection that increases the upper limit of photovoltaic availability to 96%).

[0121] The coupling of resource subsets with forecast parameters results in a multi-scenario hierarchical framework that couples resource subsets with differentiated parameters to construct a forecasting system covering multiple dimensions of assumptions, including the "Baseline Activated Scenario (BAU)," "Optimistic Scenario (Technology Acceleration + Policy Incentives)," and "Pessimistic Scenario (Resource Degradation + Policy Contraction)." The core structure of the framework includes:

[0122] The scenario dimension is defined, and the key driving assumptions for each scenario are clarified (such as the baseline scenario continuing the current policy and technology trends, the optimistic scenario assuming that the rate of technological progress doubles and the policy support is strengthened, and the pessimistic scenario assuming that wind speed / radiation decreases by 3% due to climate change and that absorption policies are weakened).

[0123] The subset-scenario mapping table assigns specific parameter values ​​to each resource subset under different scenarios (e.g., in the optimistic scenario, the upper limit of installed capacity growth in the "coastal high wind speed stable area" is 20% / year, and the technological progress correction coefficient is 1.01 / year; in the pessimistic scenario, the lower limit of growth rate drops to 3% / year, and the lower limit of availability is 90%).

[0124] The framework features a dynamic update mechanism, enabling real-time parameter calibration and subset re-division as new policies are released, technological breakthroughs occur, or climate monitoring data is updated.

[0125] It should be noted that by accurately dividing resource subsets and allocating differentiated forecast parameters, and comprehensively considering factors such as resource endowment, policy guidance, and climate change, a multi-scenario forecasting framework is constructed, which significantly improves the scientificity, adaptability, and accuracy of the forecasting of medium- and long-term power generation capacity of new energy sources, and provides strong support for power system planning and dispatch.

[0126] In this embodiment of the invention, step S300 includes the following sub-steps C1-C3;

[0127] In C1: Select at least one prediction model to predict the power generation capacity of each resource subset, and use the model fusion method to fuse the output results of at least one prediction model to obtain the preliminary prediction value;

[0128] In C2: External constraints are introduced to limit the preliminary predicted values.

[0129] In one alternative embodiment, the prediction model can be a Bayesian network, which constructs a complex relationship between meteorological conditions, equipment status, and policy factors. A Bayesian network is constructed, taking meteorological data as input, and learning the probabilistic relationship between meteorological conditions and power generation capacity through historical data, thereby predicting future power generation capacity.

[0130] In another alternative embodiment, the prediction model can also be a deep belief network, which takes meteorological data and power operation data as input, and learns the deep features of these data through the deep belief network to predict future power generation capacity.

[0131] In this embodiment of the invention, the prediction model can be a time series prediction model, a machine learning regression model, or a physical simulation model;

[0132] Specifically, time series forecasting models, targeting subsets of historical power output data exhibiting significant periodic characteristics, employ either SARIMA (Seasonal Autoregressive Integral Moving Average) or Prophet (an additive model open-sourced by Facebook). Taking SARIMA as an example, its input features include historical power generation sequences (…). The model determines hyperparameters (e.g., p=2, d=1, q=1, P=1, D=1, Q=1, s=24) by identifying the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the sequence, and then estimates the power value at future times. These hyperparameters include seasonal periodic parameters, autoregressive order (p, reflecting the impact of power at historical time points tp on the current time), moving average order (q, capturing the smoothing effect of random noise), and seasonal autoregressive / moving average order (P / Q, handling yearly / monthly periodicity). (where k is the prediction step size).

[0133] Machine learning regression models, when predicting targets influenced by multiple factors, employ gradient boosting decision trees (GBDT, such as XGBoost / LightGBM) or long short-term memory networks (LSTM). Taking XGBoost as an example, input features include current and lagging meteorological factors, system-related variables, subset attributes, and external driving parameters. The model iteratively constructs a decision tree and optimizes the objective function to fit the input features to the power generation capacity. The nonlinear relationship.

[0134] The physical simulation model targets the underlying physical processes of wind / solar power, employing a parametric model based on power curves or a simplified version of CFD (Computational Fluid Dynamics). For example, the input to the wind power physical model is the wind speed at time t (…). ), air density ( Affected by temperature / altitude), rated capacity of the wind turbine ( ) and cut-in / cut-out wind speed ( / The theoretical output force, calculated using a piecewise function, is expressed as follows:

[0135] ( if ; if ; if ; if )

[0136] The output is adjusted in conjunction with a correction factor for technological advancements (such as a 5% increase in blade efficiency).

[0137] Specifically, fusing the outputs of at least one prediction model includes integrating the outputs of different models using weighted integration or stacking strategies.

[0138] For example, for the prediction of monthly power generation capacity of the same subset, if the output of the time series model (SARIMA) is... The output of the machine learning model (XGBoost) is The physical model output is The initial predicted value after fusion ( The weighted calculation based on business rules is expressed as follows:

[0139]

[0140] Among them, the weighting coefficient , , The α value is assigned inversely based on the mean absolute error (MAE) of the historical validation period (e.g., if SARIMA has the lowest MAE, then α is chosen to be higher). For hourly high-precision forecasts, LSTM or XGBoost is preferred; for annual planning forecasts, a trend correction term (such as the linear increment brought by the increase in installed capacity) is superimposed on the physical model.

[0141] Specifically, external constraints include explicitly introducing seasonal features and planning demand constraints into the model input layer. For example, the input features of the photovoltaic model include "the season at time t (spring / summer / autumn / winter)" (encoding periodicity through sine / cosine functions) and "the number of hours between time t and sunset" (reflecting the stage of sunlight decay). Constraints are used to adjust the output upper limit (e.g., if planning requires wind power to account for a minimum of 15% of output during the winter heating season, the initial forecast must meet this requirement). ).

[0142] Preliminary forecasts only reflect resource potential. Actual power generation capacity is subject to system constraints such as grid transmission capacity, energy storage regulation potential, and consumption policies. Therefore, it is necessary to construct an Ability Allocation Matrix (AAM) to dynamically adjust the power allocation of subsets.

[0143] The quantification of system constraints includes determining the power grid transmission capacity, based on the regional main grid structure (such as the rated capacity of 500kV transmission lines and the short-circuit capacity of substation nodes) and real-time power flow calculation results, to determine the maximum power that can be transmitted to each subset. For example, if a wind power cluster is connected to a hub substation via a single 500kV line with a thermal stability limit of 2000MW, then the total output of the cluster must not exceed 2000MW (actually 1800MW after considering a 10% safety margin).

[0144] Energy storage regulation potential, based on the rated capacity of the supporting energy storage system. Charging and discharging power Given the state of charge (SOC, current capacity / rated capacity), calculate its net adjustment during the forecast period. The logic is as follows: if the current SOC is low (e.g., <20%), then charge first (to absorb excess power generation), otherwise discharge (to make up for the power gap).

[0145] The upper limit of the adjustment amount is (Charging scenario) or (Discharge scenario), in which The time step (e.g., 1 hour).

[0146] The renewable energy consumption ratio, stipulated by policy or market mechanisms, translates into minimum / maximum constraints on the output of a subset of power sources. For example, the total monthly output of wind power + solar power must meet 0.3 × ≤ ≤0.5× ).

[0147] It should be noted that by integrating multiple prediction models and introducing external constraints, and comprehensively considering resource potential and system operation constraints, accurate prediction and dynamic adjustment are achieved, significantly improving the scientificity, adaptability and reliability of new energy power generation capacity prediction, and providing strong support for power system planning and dispatch.

[0148] In this embodiment of the invention, step S400 includes the following sub-steps D1-D6;

[0149] In D1: the objective function is to maximize the total absorption benefit or minimize the total abandoned electricity.

[0150] In D2: Constraints include subset output limits, matching of total power system demand, dynamic balance of energy storage, and nonnegativity and normalization;

[0151] In D3: The objective function is solved using an optimization algorithm to obtain the optimal solution for each element in the capability allocation matrix, and the capability allocation matrix is ​​constructed using the optimal solution for each element.

[0152] In D4: The capacity allocation matrix is ​​applied to the initial prediction values ​​to obtain the adjusted power generation capacity trajectory of each subset;

[0153] In D5: The adjusted power generation capacity trajectories of each subset are used as new observation data to update the time series feature mapping table and the corresponding feature vectors.

[0154] In D6: Based on the update results, trigger parameter retraining of the prediction model or iteration of prediction parameters of the resource subset.

[0155] Specifically, the capability allocation matrix is ​​an N×M weight matrix (N is the number of subsets, M is the number of prediction periods), and its elements... This represents the power allocation ratio of the i-th subset in the j-th time period (0 ≤ (≤1), the solution is obtained through an optimization algorithm.

[0156] The objective function includes maximizing the overall system absorption efficiency (such as renewable energy utilization rate) or minimizing the amount of power curtailed, with the following logic: (The goal is to maximize total power generation) while satisfying the constraints.

[0157] The set of constraints includes:

[0158] The upper limit of subset output is expressed as:

[0159]

[0160] in, For subset Rated capacity, Set its maximum availability (e.g., 95%).

[0161] The overall system requirements matching is represented as:

[0162] Minimum capacity for waste disposal (ensuring minimum disposal requirements)

[0163] The dynamic balance of energy storage is expressed as:

[0164] (SOC is always within the safe range of 20%~80%)

[0165] Nonnegativity and normalization are expressed as:

[0166] Furthermore, the distribution ratio of the same subset in adjacent time periods does not exceed the threshold (e.g., ±0.2, to avoid power abrupt changes).

[0167] In one alternative embodiment, the optimization algorithm can be particle swarm optimization, where each particle is represented as a possible power allocation scheme, and these schemes are gradually optimized through swarm cooperation to find the optimal power allocation scheme that satisfies all constraints.

[0168] In another alternative embodiment, the optimization algorithm can also be a simulated annealing algorithm, which treats the current power allocation scheme as a state and gradually adjusts these states through the simulated annealing process to find the optimal power allocation scheme that satisfies all constraints.

[0169] In this embodiment of the invention, the optimization algorithm includes linear programming (LP) (if the objective and constraints are linear) or genetic algorithm (GA) (if there is nonlinear coupling, such as energy storage lifetime loss being related to charge and discharge depth).

[0170] For example, LP uses the simplex method to iteratively solve for the optimal allocation ratio that satisfies all constraints. The final capability allocation matrix AAM is generated. The value of is the actual power allocation coefficient of subset i in time period j.

[0171] The output of the capability allocation matrix needs to be used to correct the input features of the prediction model, forming a dynamic closed loop of "prediction-optimization-update" to improve the accuracy of subsequent predictions.

[0172] The allocated actual power values ​​are used as new observation data to update the power generation capacity trajectory of the subset. Simultaneously, the derived features of the subset are corrected, and the corresponding fields in the time-series feature mapping table are updated synchronously.

[0173] The updated capability trajectory is fed back to the feature update module, triggering the following operations:

[0174] If the retraining of model parameters reveals that the original model underestimated the impact of grid constraints, including multiple instances of power curtailment due to insufficient transmission capacity, then the input features of the machine learning model should be adjusted or the seasonal parameters of the time series model should be refitted.

[0175] Subset parameter iteration: Based on actual running data, the differentiated parameters of the subset are corrected.

[0176] It should be noted that by using dynamic optimization and feedback mechanisms to adjust the prediction model and parameters in real time, the accuracy and adaptability of new energy power generation capacity prediction are significantly improved, ensuring that the prediction results closely match the actual operating conditions.

[0177] In this embodiment of the invention, step S500 includes the following sub-steps E1-E5;

[0178] In E1: Calculate at least one deviation index between the predicted and actual power generation capacity;

[0179] In E2: Deviation indicators are decomposed into multiple dimensions, and corresponding correction processes are triggered for different dimensions;

[0180] In E3: Deviation indicators are compared with multi-level thresholds, which include at least warning thresholds, intervention thresholds, and emergency thresholds;

[0181] In E4: When the deviation indicator triggers the intervention threshold and the emergency threshold, the corresponding re-optimization strategy is activated;

[0182] In E5: Re-optimization strategies include retraining the prediction model, reconstructing the scheduling matrix, and calibrating subset parameters.

[0183] Specifically, deviation indicators include absolute deviation, which directly reflects the absolute difference between the predicted and actual values, expressed as: ;

[0184] Relative deviation eliminates the influence of dimensions and measures proportional deviation (e.g., wind power forecast output is 1596 higher than the actual value), expressed as: ;

[0185] Cumulative bias, identifying long-term systematic biases (such as a model overestimating photovoltaic output for a week consecutively), is expressed as: T represents the rolling window period, such as 7 days;

[0186] The fluctuation deviation, reflecting the degree of dispersion in prediction stability, is expressed as: ( This represents the average deviation within the window.

[0187] Deviations are decomposed into multiple dimensions, including subsets, time scales, and prediction targets, to pinpoint the specific source of the problem. For example, if the daily relative deviation of a coastal wind farm exceeds 10% for three consecutive days, while the monthly cumulative deviation of an inland photovoltaic cluster exceeds 20%, corresponding correction processes need to be triggered separately.

[0188] The system has preset dynamic thresholds, which are divided into three levels: early warning threshold, intervention threshold, and emergency threshold.

[0189] The warning threshold is when the relative deviation r > 5% or the absolute deviation e > 5% of the rated capacity, triggering a data quality check or local feature update.

[0190] When the intervention threshold is r > 15% or E > 10%, the retraining of the prediction model parameters or the local optimization of the scheduling matrix is ​​automatically triggered.

[0191] The emergency threshold is r > 30% or when the deviation exceeds 20% for three consecutive time periods, which will force a full model reconstruction and notify the operation and maintenance personnel to intervene.

[0192] Threshold settings are based on the quantiles of historical deviation distributions (e.g., the 95th percentile corresponds to common extreme deviations) and the acceptable range for business operations. For example, a regional power grid stipulates that backup power dispatch must be initiated when the wind power forecast deviation exceeds 20%, therefore the emergency threshold is set to r > 20%.

[0193] When a deviation triggers a threshold, the system selects a targeted re-optimization strategy based on the type of deviation. The core objective is to correct model parameters or scheduling rules to adapt to the latest data distribution.

[0194] For time series models or machine learning models, incremental learning or window sliding retraining strategies can be used:

[0195] Incremental learning allows the model to gradually adjust weights using an online algorithm, eliminating the need for recalculation of the entire dataset. For example, when the SARIMA model detects an increase in wind speed prediction bias, it recursively estimates new autoregressive coefficients. ) and moving average coefficient ( Only update the data window for the most recent N time periods (e.g., 30 days).

[0196] The sliding window retraining method retains high-quality data from the most recent M time periods (e.g., 6 months) and re-executes the entire modeling process. For example, if the seasonal pattern of photovoltaic output shifts due to climate change, the sliding window removes data from older years and incorporates radiation-output correlation features from the latest 12 months to retrain the XGBoost model.

[0197] When the deviation originates from changes in system constraints or drift in subset characteristics, the capability allocation matrix is ​​reconstructed.

[0198] The constraints are updated, and the real-time transmission capacity of the power grid, the latest state of charge of the energy storage system, and the consumption ratio target after the policy adjustment are obtained again.

[0199] The optimization objectives and constraints are reconstructed, and the objective function and constraint set of the capability allocation matrix are redefined based on the updated parameters;

[0200] The algorithm is adapted to the solution. If the constraints are highly linearized, linear programming (LP) is used for fast solution. If there is nonlinear coupling, genetic algorithm (GA) or particle swarm optimization (PSO) is used to find an approximate optimal solution.

[0201] For differentiated parameters of resource subsets, their value logic is adjusted based on the latest actual data. For example, if the actual annual equivalent utilization hours of a photovoltaic subset are 5% lower than the predicted value for two consecutive years, its technological progress correction coefficient is reduced from 1.005 / year to 1.002 / year; if the equipment failure rate of a wind farm increases, causing the availability rate to drop from 95% to 92%, its operating upper limit parameter is updated. The figure is 92%.

[0202] In this embodiment of the invention, step S600 includes the following sub-steps F1-F3;

[0203] In F1: Obtain the disturbance index, input the disturbance index into the security assessment model, and output the trend adjustment coefficient and boundary condition correction coefficient;

[0204] In F2: The long-term power generation capacity prediction trend of the resource subset is corrected by the trend adjustment coefficient, and the system constraints in the capacity allocation matrix are updated by the boundary condition correction coefficient to obtain the corrected prediction curve and boundary conditions.

[0205] In F3: Based on the corrected prediction curve and boundary conditions, the medium- and long-term power supply security indicators are recalculated and the security margin threshold is updated to realize the prediction of the medium- and long-term power generation capacity of new energy sources.

[0206] Specifically, the acquisition of disturbance indicators includes meteorological extreme signals, and the extraction of key indicators based on long-term climate models and real-time monitoring data. For example, when the ENSO index indicates that a strong La Niña event will occur in the next three months, the system marks it as "high risk of typhoon disturbance".

[0207] Policy / demand shock factors are extracted by mining policy texts and macroeconomic data to identify signals of changes in external demand.

[0208] The safety assessment model employs scenario simulation and sensitivity analysis methods, outputting a correction factor for the prediction curve after inputting disturbance indicators. This is the trend adjustment coefficient. (These are boundary condition correction coefficients);

[0209] The trend adjustment is as follows: if extreme weather events are expected to cause an overall decrease in wind speed of 1096, then the future power forecast trend of wind turbines will be multiplied by the correction factor. If the policy requires the growth rate of photovoltaic installations to increase to 25%, then the slope of the installation expansion curve for the photovoltaic subset will be revised to... .

[0210] The boundary conditions are modified to address the system limits under extreme scenarios, adjusting the capacity allocation matrix accordingly. The constraint is updated to 800MW, and the safety margin is recalculated; if the policy mandates that "the wind and solar power consumption ratio shall not be less than 35%", the lower limit constraint of the consumption ratio will be adjusted to 35%; the minimum feasible output boundary of the forecast curve is revised.

[0211] Based on the revised forecast curves and boundary conditions, the system recalculates medium- and long-term power supply security indicators and dynamically updates the security margin thresholds. For example, after a typhoon warning is triggered, the peak output boundary of the wind power forecast curve in a certain regional power grid is lowered from 90% to 80% of the rated capacity. At the same time, the supporting energy storage system is required to provide an additional 50MW of regulation capacity during the typhoon to ensure the safe operation of the system.

[0212] It should be noted that through real-time monitoring and dynamic adjustments, the system can effectively respond to extreme events and policy changes, significantly improving the adaptability of forecasts and the security of system operation.

[0213] The above is an illustrative scheme of a software data processing method for medium- and long-term power generation capacity prediction according to this embodiment. It should be noted that the technical solution of this software data processing system for medium- and long-term power generation capacity prediction belongs to the same concept as the technical solution of the software data processing method for medium- and long-term power generation capacity prediction described above. Details not described in detail in the technical solution of the software data processing system for medium- and long-term power generation capacity prediction in this embodiment can be found in the description of the technical solution of the software data processing method for medium- and long-term power generation capacity prediction described above.

[0214] The software data processing system for medium- and long-term power generation capacity prediction in this embodiment includes:

[0215] The preprocessing module is used to acquire multidimensional data of new energy power generation, preprocess the multidimensional data, and obtain time-series feature vectors.

[0216] The partitioning module is used to partition new energy resources based on the time-series feature vector to obtain multiple resource subsets, assign prediction parameters to each resource subset, and establish a multi-scenario hierarchical framework for medium- and long-term prediction.

[0217] The prediction module is used to select a prediction model based on the multi-scenario hierarchical framework to predict the power generation capacity of each resource subset and obtain preliminary prediction values.

[0218] The adjustment module is used to introduce the constraints of the power system to construct a capacity allocation matrix, adjust the preliminary prediction value, and obtain the power generation capacity prediction result.

[0219] The monitoring module is used to monitor the deviation between the predicted power generation capacity and the actual results, and trigger corresponding correction and optimization mechanisms based on the deviation characteristics.

[0220] The calibration module is used to perform safety boundary calibration and update the prediction curve when an external disturbance event is detected, so as to realize the prediction of the medium and long-term power generation capacity of new energy.

[0221] This embodiment also provides a computer device suitable for software data processing in medium- and long-term power generation capacity forecasting, including:

[0222] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a software data processing method for predicting medium- and long-term power generation capacity as proposed in the above embodiments.

[0223] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a software data processing method for predicting medium- and long-term power generation capacity as proposed in the above embodiments.

[0224] The storage medium proposed in this embodiment and the software data processing method for predicting medium- and long-term power generation capacity proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0225] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computing device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0226] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 software data processing method for mid-long term power generation capacity prediction, characterized in that, The method comprises the following steps: acquiring multi-dimensional data of new energy power generation, pre-processing the multi-dimensional data to obtain a time series feature vector; dividing new energy resources based on the time series feature vector to obtain a plurality of resource subsets, assigning a prediction parameter to each resource subset, and establishing a multi-scenario hierarchical framework for medium and long-term prediction; selecting a prediction model to predict the power generation capacity of each resource subset based on the multi-scenario hierarchical framework to obtain a preliminary prediction value; introducing a constraint condition of a power system to construct a capacity allocation matrix, and adjusting the preliminary prediction value to obtain a power generation capacity prediction result; monitoring the deviation between the power generation capacity prediction result and an actual result, triggering a corresponding correction and optimization mechanism according to the deviation characteristics; when an external disturbance event is detected, calibrating a safety boundary and updating a prediction curve to realize the prediction of the medium and long-term power generation capacity of new energy.

2. The software data processing method for medium and long term power generation capacity prediction according to claim 1, characterized in that, The pre-processing of the multi-dimensional data comprises the following steps: improving the data quality of the multi-dimensional data to obtain first multi-dimensional data; aligning the first multi-dimensional data, and constructing a time series feature mapping table through a multi-granularity aggregation method; screening feature variables from the time series feature mapping table, constructing high-order features through feature combination and derivation methods, and obtaining a time series feature vector.

3. A software data processing method for medium and long term power generation capacity prediction according to claim 1 or 2, characterized in that, The establishment of the multi-scenario hierarchical framework for medium and long-term prediction comprises the following steps: dividing new energy resources into a plurality of resource subsets by using a first clustering algorithm, and statistically verifying the difference of each subset; obtaining external driving factors and quantifying them as prediction parameters, and assigning a prediction parameter to each resource subset; coupling the resource subset and the prediction parameter to construct a multi-scenario hierarchical framework for medium and long-term prediction.

4. A software data processing method for medium and long term power generation capacity prediction according to claim 3, characterized in that, The selection of the prediction model to predict the power generation capacity of each resource subset comprises the following steps: selecting at least one prediction model to predict the power generation capacity of each resource subset, and fusing the output results of the at least one prediction model by using a model fusion method to obtain a preliminary prediction value; introducing external constraint conditions to limit the preliminary prediction value.

5. A software data processing method for medium and long term power generation capacity prediction according to claim 4, characterized in that, The construction of the capacity allocation matrix comprises the following steps: taking the maximization of total consumption benefit or the minimization of total curtailment amount as an objective function; the constraint conditions include the upper limit of the subset output, the total demand matching of the power system, the dynamic balance of the energy storage, and the non-negativity and normalization; solving the objective function by using an optimization algorithm to obtain the optimal solution of each element in the capacity allocation matrix, and constructing the capacity allocation matrix by using the optimal solution of each element; applying the capacity allocation matrix to the preliminary prediction value to obtain the adjusted power generation capacity trajectory of each subset; using the adjusted power generation capacity trajectory of each subset as new observation data to update the time series feature mapping table and the corresponding feature vector; according to the update result, triggering the parameter retraining of the prediction model or the iteration of the prediction parameter of the resource subset.

6. A software data processing method for medium and long term power generation capacity prediction according to claim 5, characterized in that, The triggering of the corresponding correction and optimization mechanism comprises the following steps: calculating at least one deviation index between the power generation capacity prediction result and the actual result; performing multi-dimensional decomposition on the deviation index, and triggering a corresponding correction process for different dimensions; comparing the deviation index with a multi-level threshold value, the multi-level threshold value at least including a warning threshold value, an intervention threshold value and an emergency threshold value; when the deviation index triggers the intervention threshold value and the emergency threshold value, starting a corresponding re-optimization strategy; The re-optimization strategy includes re-training of the prediction model, reconstruction of the dispatch matrix, and calibration of the subset parameters.

7. A software data processing method for medium and long term power generation capacity prediction according to claim 6, characterized in that, The security boundary calibration includes: Obtaining a disturbance index, inputting the disturbance index into a security assessment model, and outputting a trend adjustment coefficient and a boundary condition correction coefficient; Using the trend adjustment coefficient to correct the long-term power generation capacity prediction trend of the resource subset, and using the boundary condition correction coefficient to update the system constraint conditions in the capacity allocation matrix to obtain a corrected prediction curve and boundary conditions; Based on the corrected prediction curve and boundary conditions, the medium and long-term power supply security index is recalculated, and the safety margin threshold is updated to realize the prediction of the medium and long-term power generation capacity of new energy.

8. A software data processing system for mid-long term power generation capacity prediction, applying a software data processing method for mid-long term power generation capacity prediction according to any one of claims 1 to 7, characterized in that, It includes: A preprocessing module for obtaining multi-dimensional data of new energy power generation, preprocessing the multi-dimensional data to obtain a time series feature vector; A division module for dividing new energy resources based on the time series feature vector to obtain a plurality of resource subsets, assigning prediction parameters to each resource subset, and establishing a multi-scenario hierarchical framework for medium and long-term prediction; A prediction module for selecting a prediction model based on the multi-scenario hierarchical framework to predict the power generation capacity of each resource subset to obtain a preliminary prediction value; An adjustment module for introducing the constraint conditions of the power system to construct a capacity allocation matrix and adjusting the preliminary prediction value to obtain a power generation capacity prediction result; A monitoring module for monitoring the deviation between the power generation capacity prediction result and the actual result, and triggering the corresponding correction and optimization mechanism according to the deviation characteristics; A calibration module for performing security boundary calibration and updating the prediction curve when an external disturbance event is detected to realize the prediction of the medium and long-term power generation capacity of new energy.

9. A computer device, comprising: It includes: A memory and a processor, the memory stores a computer program, and the processor realizes the steps of the software data processing method for medium and long-term power generation capacity prediction according to any one of claims 1-7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and the computer program is executed by a processor to realize the steps of the software data processing method for medium and long-term power generation capacity prediction according to any one of claims 1-7.

Images