Construction method of wind power-load joint scenario generation model for persistent weather process

Abstract

The application discloses a kind of wind power-load joint scene generation model construction methods for persistent weather process, belong to wind power generation technical field, it will persistent weather process condition input initial scene generation model and train until model converges;Initial scene generation model includes first, second stage condition generation model;First stage condition generation model is trained with persistent weather process condition as input, with historical structured scene index as output;Second stage condition generation model is trained with historical structured scene index as input, with historical wind power-load curve as output.The scene generation model of the application will structured scene index as intermediate representation, will complex regional wind power-load hourly curve generation problem be decomposed into weather process condition to structured index generation, structured index to wind power-load joint curve generation, can guarantee the authenticity and stability of generated scene under different weather process conditions.

Description

Technical Field

[0001] This application belongs to the field of wind power generation technology, and more specifically, relates to a method for constructing a wind power-load joint scenario generation model for persistent weather processes. Background Technology

[0002] With the continuous increase in the installed capacity of new energy sources, the natural fluctuation characteristics of wind power output and the dynamic changes in electricity load are increasingly impacting power planning, energy storage capacity configuration, and power system risk assessment. In actual operation scenarios, persistent weather events can systematically alter the overall output level of regional wind power over several consecutive days, reshaping key characteristics such as wind power ramp-up rate and sustained output range. Simultaneously, they will also change the intraday variation trend and inter-day fluctuation pattern of electricity load.

[0003] The provincial-level wind power-load combined curve is the result of aggregating the output of numerous spatially heterogeneous wind farms and dispersed load users. Its overall shape is influenced by both local, small-scale meteorological factors and is highly correlated with the large-scale regional weather background. Regionally averaged persistent weather processes can reflect the overall meteorological conditions within a province at a macro level, providing a basic meteorological constraint for the long-term trends of wind power and load. However, in actual operation, the same type of regionally averaged persistent weather process often corresponds to multiple sets of significantly different provincial total wind power output sequences and total load change sequences. The relationship between the two is not a simple one-to-one correspondence, but rather exhibits obvious polymorphic mapping characteristics.

[0004] The traditional scene construction methods widely used in the industry mainly rely on the direct reuse of historical samples or random generation for a single time scale. When faced with different types of persistent weather processes, these methods often struggle to balance constraints across multiple time scales: either in order to approximate the statistical characteristics of historical data, the generated scenes deviate from the typical patterns of the corresponding weather processes in long-term trends, resulting in insufficient stability; or in order to strictly match the overall trend of a specified weather process, the generated hourly and daily detailed features deviate significantly from actual operational data, making it difficult to guarantee authenticity. In other words, current technologies cannot simultaneously ensure the authenticity and stability of generated scenes under different weather conditions. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, this application provides a method for constructing a wind power-load joint scenario generation model for persistent weather processes. The purpose is to solve the technical problem that the existing technology is unable to simultaneously ensure the authenticity and stability of the generated scenario under different weather process conditions.

[0006] To achieve the above objectives, according to one aspect of this application, a method for constructing a wind power-load joint scenario generation model for persistent weather processes is provided, comprising: S1: Obtain the historical available wind power output time series, historical load output time series, and persistent weather conditions for the preset area within the preset time period; S2: Convert the historical available wind power output time series and its corresponding historical load output time series into historical wind power-load curves; S3: Calculate historical structured scenario indicators at various time scales using the historical wind power-load curves; S4: Input the persistent weather process conditions into the initial scene generation model for training until the model converges to obtain the target scene generation model; wherein, the initial scene generation model includes: a first-stage condition generation model and a second-stage condition generation model; the first-stage condition generation model is trained with the persistent weather process conditions as input and the historical structured scene indicators as output; the second-stage condition generation model is trained with the historical structured scene indicators as input and the historical wind power-load curve as output.

[0007] Furthermore, the persistent weather process conditions include the category information and intensity description of the regional average persistent weather process, in order to distinguish the statistical tendency, intra-week trend and risk characteristics of wind power and load scenarios under different persistent weather processes.

[0008] Furthermore, the historical structured scene indicators include daily-scale structured scene indicators and weekly-scale structured scene indicators; The daily-scale structured scenario indicators include: daily average wind power output, daily wind power uphill intensity, daily wind power downhill intensity, daily average load output, and daily load peak-valley difference over multiple consecutive days. The weekly-scale structured scenario indicators include: the lowest average wind power output within a multi-hour sliding window, the highest average wind power output within a multi-hour sliding window, and the maximum average wind power ramp intensity within a multi-hour sliding window.

[0009] Furthermore, the first-stage condition generation model is a structured condition diffusion model; persistent weather process conditions and diffusion step After being embedded, the historical structured scene indicators are used as global conditional inputs. This includes: the daily-scale wind power-load trend component corresponding to the daily-scale structured scenario indicators, and the weekly-scale wind power sustainability component corresponding to the weekly-scale structured scenario indicators.

[0010] Furthermore, the daily-scale wind power-load trend section includes a daily wind power trend branch and a daily load trend branch; The wind power daily trend branch is used to characterize the daily average wind power, uphill and downhill wind power, and the corresponding cross-day changes. The load daily trend branch is used to characterize the daily average load and the peak-valley difference, etc., cross-day changes. The wind power daily trend branch, the load daily trend branch, and the weekly wind power persistence part are bidirectionally fused to form a structured indicator output that takes into account both daily trend and weekly persistence risk.

[0011] Furthermore, the first-stage conditional generation model introduces soft constraints on physical structure during the training process; The physical structure soft constraints include weekly-scale index sequence constraints and daily-scale at most-day window proxy consistency constraints. The weekly-scale index sequence constraint is used to constrain the minimum average wind power output over multiple days to not be higher than the maximum average wind power output over multiple days, and to constrain the wind power ramping intensity to be within a physically reasonable range. The daily-scale to multi-day window proxy consistency constraint utilizes the sliding window proxy formed by the daily average wind power values ​​over multiple consecutive days to ensure that the generated multi-day minimum average wind power index, multi-day maximum average wind power index, and weekly daily-scale wind power trend maintain a structural correlation.

[0012] According to another aspect of this application, a construction apparatus for a wind power-load joint scenario generation model oriented towards persistent weather processes is provided, comprising: The acquisition module is used to acquire the historical available wind power output time series, historical load output time series, and persistent weather conditions for a preset area within a preset time period. The conversion module is used to convert the historical available wind power output time series and its corresponding historical load output time series into historical wind power-load curves. The calculation module is used to calculate historical structured scenario indicators at various time scales using the historical wind power-load curves. The training module is used to input the persistent weather process conditions into the initial scene generation model for training until the model converges, thereby obtaining the target scene generation model. The initial scene generation model includes a first-stage condition generation model and a second-stage condition generation model. The first-stage condition generation model is trained using the persistent weather process conditions as input and the historical structured scene indicators as output. The second-stage condition generation model is trained using the historical structured scene indicators as input and the historical wind power-load curve as output.

[0013] According to another aspect of this application, a method for generating wind power-load joint scenarios for persistent weather processes is provided, comprising: inputting the current weather process conditions into the target scenario generation model so that the first-stage condition generation model outputs the current structured scenario index, and then inputting the current structured scenario index into the second-stage condition generation model so that it outputs the target wind power-load joint curve, and using it as the target wind power-load joint scenario.

[0014] Furthermore, the method for generating wind power-load joint scenarios for persistent weather processes also includes: The target wind power-load combined scenario is evaluated on a multi-scale basis. The multi-scale evaluation includes hourly-scale output distribution and autocorrelation evaluation, daily-scale structured index evaluation, weekly-scale persistent risk index evaluation, and three-dimensional joint structure evaluation of key weekly-scale wind power indicators. The three-dimensional joint structure evaluation of key weekly-scale wind power indicators uses multiple hours of maximum average ramp intensity, multiple hours of minimum average wind power output, and multiple hours of maximum average wind power output as three-dimensional coordinates to form a real sample point set and a generated sample point set. The difference between the two types of point sets in the joint space of key weekly-scale risk indicators is quantified by the three-dimensional optimal transmission distance.

[0015] According to another aspect of this application, a system for generating wind power-load combined scenarios is provided, comprising: a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for generating wind power-load combined scenarios oriented towards persistent weather processes.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived in this application can achieve the following beneficial effects: (1) This invention provides a method for constructing a wind power-load joint scenario generation model for persistent weather processes. The persistent weather process conditions are input into an initial scenario generation model for training until the model converges, resulting in a target scenario generation model. The initial scenario generation model includes a first-stage condition generation model and a second-stage condition generation model. The first-stage condition generation model is trained using the persistent weather process conditions as input and the historical structured scenario indicators as output. The second-stage condition generation model is trained using the historical structured scenario indicators as input and the historical wind power-load curve as output. The target scenario generation model constructed in this way uses structured scenario indicators as intermediate representations to decompose the complex problem of generating regional wind power-load hourly curves into the generation of structured indicators from weather process conditions and the generation of wind power-load joint curves from structured indicators. This method can simultaneously ensure the authenticity and stability of the generated scenarios under different weather process conditions.

[0017] (2) The persistent weather process conditions described in this scheme include the category information and intensity description of the regional average persistent weather process, so as to distinguish the statistical tendency, intra-week trend and risk characteristics of wind power and load scenarios under different persistent weather processes. This design takes into account that different types and intensities of weather processes have different impacts on wind power and load. The advantage is that it can distinguish different persistent weather processes and their degree of impact, and realize differentiated meteorological condition guidance for the scenario generation process.

[0018] (3) The historical structured scenario indicators mentioned in this scheme include daily-scale structured scenario indicators and weekly-scale structured scenario indicators. This design takes into account that wind power and load have both daily-scale variation and weekly-scale continuity characteristics. The advantage is that it can transform complex hourly curves into multi-time-scale intermediate representations with clear physical meaning, and realize the unified expression of daily-scale trends and weekly-scale risk characteristics.

[0019] (4) In this scheme, the first stage condition generation model is a structured condition diffusion model. This design takes into account that the same weather process may correspond to multiple reasonable structured scenario indicators. The advantage is that it can learn the conditional probability distribution of structured scenario indicators under weather conditions, and realize the probabilistic and diversified generation of structured scenario indicators.

[0020] (5) The daily-scale wind power-load trend section in this scheme includes the wind power daily trend branch and the load daily trend branch. This design takes into account that wind power and load have different change patterns but also have a synergistic relationship. The advantage is that the daily-scale features of the two can be extracted separately and then fused together, realizing the differentiated modeling and synergistic expression of the heterogeneous features of wind power and load.

[0021] (6) In this scheme, the first-stage conditional generation model introduces physical structure soft constraints during the training process; the physical structure soft constraints include weekly scale index order constraints and daily scale to multi-day window proxy consistency constraints; this design takes into account the problem that the generated index may have unreasonable numerical order or contradictory features at different time scales. The advantage is that it can use physical relationships to flexibly guide the model training, realizing the physical feasibility of weekly scale indices and the structural consistency of daily scale trends and multi-day window indices.

[0022] (7) This invention provides a method for generating wind power-load joint scenarios for persistent weather processes. The current weather process conditions are input into the target scenario generation model, so that the first-stage condition generation model outputs the current structured scenario index. Then, the current structured scenario index is input into the second-stage condition generation model, so that it outputs the target wind power-load joint curve, which is used as the target wind power-load joint scenario. This design takes into account the complex mapping relationship of directly generating high-dimensional hourly curves from weather process conditions. The advantage is that the generation process can be decomposed into two interconnected stages using structured scenario indexes, realizing the hierarchical generation from weather conditions to multi-timescale indexes, and then to hourly wind power-load joint curves.

[0023] (8) In this scheme, the target wind power-load joint scenario is evaluated on a multi-scale basis. The multi-scale evaluation includes a three-dimensional joint structure evaluation of key weekly wind power indicators. This design takes into account that a single scale or a single indicator is difficult to fully reflect the authenticity of the generated scenario. Its advantage is that it can comprehensively evaluate the hourly, daily and weekly scale characteristics and measure the joint relationship between multiple key weekly scale risk indicators, thereby achieving a comprehensive evaluation of the ability to maintain the multi-scale characteristics of the generated scenario and the joint risk structure. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for constructing a wind power-load joint scenario generation model for persistent weather processes, provided in an embodiment of this application. Figure 2 This is a schematic diagram illustrating the relationship between persistent weather process conditions, structured indicators, and hourly curves provided in one embodiment of this application.

[0025] Figure 3 This is a schematic diagram of the first-stage physical structure-guided structured index generation process provided in an embodiment of this application.

[0026] Figure 4 This is a schematic diagram of the hourly joint curve generation process under the second-stage structured index conditions provided in an embodiment of this application.

[0027] Figure 5 This is a flowchart of the method for generating wind power-load joint scenarios for persistent weather processes, provided in one embodiment of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Furthermore, the technical features involved in the various embodiments described below can be combined with each other as long as they do not conflict with each other.

[0029] Example 1 This embodiment provides a method for constructing a wind power-load joint scenario generation model for persistent weather processes, such as... Figure 1 As shown, it includes: S1-S4.

[0030] S1: Obtain the historical available wind power output time series, historical load output time series, and persistent weather conditions for the preset area within the preset time period.

[0031] Specifically, it can obtain the per-unit time series of available wind power output, the per-unit time series of load output, and the persistent weather conditions corresponding to each weekly window. A weekly window is defined as 168 consecutive hours, with a sample duration of 168 hours. The hourly wind power-load combined curve includes both wind power output and load output channels, mathematically expressed as follows:

[0032] Construct a sample that includes weather process conditions, structured scenario indicators, and hourly wind power-load joint curves.

[0033] As an optional implementation, persistent weather process conditions include category information and intensity description of regional average persistent weather processes to distinguish the statistical tendencies, intra-weekly trends and risk characteristics of wind power and load scenarios under different persistent weather processes.

[0034] Specifically, Figure 2 This is a schematic diagram illustrating the relationship between persistent weather process conditions, structured indicators, and hourly curves according to an embodiment of this application. Persistent weather process conditions include category information and intensity descriptions of regional average persistent weather processes. These conditions are used to distinguish the statistical trends, intra-weekly trends, and risk characteristics of wind power and load scenarios under different persistent weather processes. In one embodiment, the conditions can be expressed as: The superscript indicates a dimension of 30. A real number vector of 1 can be simply understood as a set of features consisting of 30 numbers. Specifically, the 30 numbers include 4 sets of 7-day meteorological features (7-day average temperature, 7-day precipitation, 7-day average wind speed, and 7 types of weather labels) and 2 time codes (indicating which week of the year the current week falls in).

[0035] S2: Convert the historical available wind power output time series and its corresponding historical load output time series into historical wind power-load curves.

[0036] S3: Calculate historical structured scenario indicators for various time scales using historical wind power-load curves.

[0037] As an optional implementation method, Figure 3 This is a schematic diagram of the first-stage physical structure-guided structured index generation process provided in an embodiment of this application. Historical structured scenario indicators include daily-scale and weekly-scale structured scenario indicators. Daily-scale structured scenario indicators include: daily average wind power output, daily wind power uphill intensity, daily wind power downhill intensity, daily average load output, and daily load peak-valley difference over multiple consecutive days. Weekly-scale structured scenario indicators include: the lowest average wind power output within multiple hourly sliding windows, the highest average wind power output within multiple hourly sliding windows, and the maximum average wind power uphill intensity within multiple hourly sliding windows.

[0038] Specifically, structured scenario indicators are calculated based on the historical wind power-load curves within each weekly window. These structured scenario indicators include daily and weekly scale indicators, and can be expressed as follows:

[0039] in, Indicates a daily-scale index, The indicators are displayed on a weekly scale. Daily scale indicators are calculated over 7 days and include daily average wind power output, daily wind power uphill intensity, daily wind power downhill intensity, daily average load output, and daily load peak-to-valley difference. Weekly scale indicators are used to describe persistent risk and long-term window characteristics, including the minimum average wind power output within a 72-hour sliding window, the maximum average wind power output within a 72-hour sliding window, and the maximum average wind power uphill intensity within a 72-hour sliding window.

[0040] To characterize the impact of persistent weather events on weekly-scale wind power patterns, this implementation further employs a 72-hour sliding window to extract weekly-scale wind power indices. Let the hourly wind power output sequence over one week be the superscript of W for P, and the starting set and corresponding window of the 72-hour window be defined as follows:

[0041] Based on the above window, the 72-hour minimum average wind power output, the 72-hour maximum average wind power output, and the 72-hour maximum average climbing intensity are respectively expressed as:

[0042]

[0043]

[0044] In the formula This refers to wind power output during time period t. The above indicators... This describes the average uphill intensity during periods of continuous low winds, continuous high winds, and persistent weather. Indicators such as the proportion of hours with high winds, the longest duration of high winds, and the longest duration of low winds can be used as optional extended indicators, incorporated into the structured scenario indicators according to specific application requirements.

[0045] Furthermore, the dimensions of the structured scenario indicators are normalized to ensure that indicators with different dimensions and ranges are on a unified numerical scale. The hourly wind power-load curve is in a per-unit format to facilitate subsequent condition generation and multi-scale evaluation.

[0046] S4: Input the persistent weather conditions into the initial scene generation model and train it until the model converges to obtain the target scene generation model. The initial scene generation model includes: a first-stage condition generation model and a second-stage condition generation model. The first-stage condition generation model is trained using persistent weather conditions as input and historical structured scene indicators as output. The second-stage condition generation model is trained using historical structured scene indicators as input and historical wind power-load curves as output. Figure 4 This is a schematic diagram of the hourly joint curve generation process under the second-stage structured index conditions provided in an embodiment of this application.

[0047] As an optional implementation, the first-stage condition generation model is a structured condition diffusion model. (Conditions for persistent weather processes) and diffusion step After being embedded, it serves as a global conditional input, historical structured scene indicators. This includes: the daily-scale wind power-load trend component corresponding to the daily-scale structured scenario indicators, and the weekly-scale wind power sustainability component corresponding to the weekly-scale structured scenario indicators.

[0048] The first-stage conditional generation model is used to learn the following conditional distributions:

[0049] That is, under given persistent weather conditions Generate structured scene metrics The first-stage conditional generative model introduces soft constraints on physical structure during training to improve the physical rationality of the generated indices and the consistency of diurnal and periodic scale structures. Specifically, the first phase of training will be based on weather conditions. As a condition, normalized structured scenario indicators To generate objects, a first-stage conditional generation model is trained. The first-stage training uses conditional diffusion to model the structured index. Given the zero index of the true structured index Z, the weather condition vector R, and the diffusion step t, the first-stage conditional generation model takes the noisy structured index, the diffusion step, and the weather conditions as input and learns to predict the noise term. Its basic diffusion loss is:

[0050] Considering the comprehensive diffusion reconstruction, weekly-scale sequence relationships, surrogate consistency from the daily scale to the 72-hour window, and the low-temperature load underestimation constraint, the overall training objective for the first stage is written as:

[0051] in, This indicates the weekly-scale index order constraint. This indicates a daily-scale to 72-hour window proxy consistency constraint. Due to the constraint of underestimation of low temperature load, , and For the corresponding weights.

[0052] First-stage conditional generation model and The embedding is used as a global condition for noisy structured metrics. Noise prediction and progressive denoising are performed. The structured decomposition relationship can be represented as: ;in, The section representing the daily-scale wind power / load trend is... This represents the weekly scale of wind power sustainability. The daily scale is organized according to 7 days and various daily indicators, including the daily average wind power, wind power upslope, wind power downslope, daily average load, and load peak-valley difference; the weekly scale includes indicators such as the minimum, maximum, and maximum average upslope intensity of wind power output within a 72-hour sliding window.

[0053] To enable the generation of weekly-scale indicators in the first stage to explicitly perceive the multi-day window effect in the continuous weather process, this implementation method sets up a rolling indicator attention module on the daily-scale hidden representation. Let... Let the joint hidden representation of wind power and load on the j-th natural day be represented by the encoder, which takes the daily-scale hidden features of three consecutive days and feeds them into an encoder to obtain the window features corresponding to these three days. U j :

[0054] For different weekly scale metrics, the k-index of the metric-related query vector q is set, and attention weights are calculated for the contexts of the five scrolling windows:

[0055] This yields a window representation associated with the k-th weekly scale index:

[0056] In the formula, For the first k Weekly wind power index for the first j The weight of each scroll window, Subsequently, the weekly-scale baseline state and window representations of different indicators are jointly input into the weekly-scale output network to aggregate the relationship between the weather label R learned by the model and the weekly-scale wind power indicators, and then output as a mapping. The weekly-scale output network is used to generate predictive noise for indicators such as the minimum average wind power, the maximum average wind power, and the maximum average slope over 72 hours. 。 This module adaptively selects the most explanatory 3-day window through indicator queries, without limiting the specific number of channels, hidden layers, or implementation framework.

[0057] As an optional implementation, the daily-scale wind power-load trend component includes a daily wind power trend branch and a daily load trend branch. The daily wind power trend branch is used to characterize the daily average wind power, uphill and downhill wind power, and the corresponding inter-day changes, while the daily load trend branch is used to characterize the daily average load and peak-valley differences, etc., inter-day changes. The daily wind power trend branch, the daily load trend branch, and the weekly-scale wind power persistence component are bidirectionally fused to form a structured indicator output that takes into account both daily-scale trends and weekly-scale persistence risks.

[0058] As an optional implementation, the first-stage conditional generation model introduces soft constraints on physical structure during training. These constraints include weekly-scale index order constraints and daily-scale multi-day window proxy consistency constraints. The weekly-scale index order constraints ensure that the minimum average wind power output over multiple days does not exceed the maximum average wind power output over multiple days, and that the wind power ramp-up intensity remains within a physically reasonable range. The daily-scale multi-day window proxy consistency constraints utilize a sliding window proxy formed by the daily average wind power values ​​over consecutive days to maintain a structural correlation between the generated minimum and maximum average wind power indices and the daily-scale wind power trend within the week.

[0059] Optionally, the weekly-scale index sequence constraint is used to penalize situations where the 72-hour minimum average wind power output is higher than the 72-hour maximum average wind power output, and to suppress unreasonable values ​​for wind power ramp-up intensity. The daily-scale to 72-hour window proxy consistency constraint constructs a sliding window proxy based on the daily average wind power values ​​over multiple consecutive days, ensuring that the generated 72-hour minimum and maximum average wind power outputs are consistent with the daily-scale wind power trend. These constraints are soft constraints and do not require the daily average proxy to be completely equivalent to the hourly 72-hour window index.

[0060] To address the issue that the upper tail of the daily average load is easily underestimated by the generative model during periods of sustained low temperatures, this implementation introduces a low-temperature load underestimation constraint in the first stage. The temperature threshold can be determined by the low-temperature quantile of historical samples. An asymmetric penalty is applied to the underestimated portion of the daily average load during low temperatures. This constraint only penalizes the underestimated portion of the daily average load during low temperatures, without globally boosting all samples, thereby enhancing the interpretability of the generative model for low-temperature, high-load scenarios.

[0061] In the first phase of training, physical structure soft constraints are only activated during the low-to-medium noise stage. This allows the model to receive structural guidance as the sample morphology gradually becomes clear, avoiding the impact on generation stability caused by forcibly imposing structural constraints during the high-noise stage. For example, physical structure soft constraints are activated during the low-to-medium noise stage, allowing structural constraints to act on the generation stage where the sample morphology gradually becomes clear. In one implementation, the weekly scale index sequence constraint weight is 0.02, the daily scale to 72-hour window surrogate consistency constraint weight is 0.05, and the upper limit of the time step ratio for activating physical structure soft constraints is 0.5.

[0062] With normalized structured scenario indicators As a condition, the combined wind power-load curve of the dual-channel system was calculated over a period of 168 hours per week. To generate objects, a second-stage conditional generation model is trained. This second-stage model employs a conditional diffusion-based hourly curve generation structure, with input including noisy hourly curves. diffusion step and condition indicators Wind power and load are generated jointly through dual channels.

[0063] The second stage uses the structured trend index Z generated in the first stage or given historically as a condition to learn the conditional distribution of the hourly wind power-load joint curve Y. The second-stage conditional diffusion network takes the noisy hourly curve, diffusion step, and structured trend index as input, and the training objective is the noise prediction loss. 。 In this model, structured trend indicators serve as conditional information to guide the joint generation of hourly wind power and load curves. The consistency constraints of the second-stage sub-indicators will be explained in detail later; the loss formula will not be elaborated here. Specifically, the second-stage conditional generation model encodes and compresses the 168-hour dual-channel curve to a shorter time scale, and sets a shared temporal convolution module in the compressed joint feature space. This module uses dilated temporal convolution to extract long-term dependencies, cross-day transitions, and local continuity features, and then restores the 168-hour wind power-load joint curve through decoding.

[0064] Given the conditions of a persistent weather process Structured scenario indicators are generated through the first-stage condition generation model. The generated indicators are then subject to reasonableness and scope constraints; and the structured scenario indicators are then... Input the second-stage conditional generation model to generate hourly wind power-load combined curves. This process yields a scene sample set containing weather conditions, structured scene indicators, wind power hourly curves, and load hourly curves. It should be noted that the scene generation model in this application can be represented as: ;in, For persistent weather conditions, For structured scenario metrics, This represents the hourly wind power-load combined curve. Through this two-stage relationship, weather conditions first influence the daily and weekly structured indices, and then these structured indices constrain the shape, trend, and risk characteristics of the hourly wind power-load curve. The second-stage condition generation model is used to learn the following condition distributions: That is, in a given structured scenario index Generating hourly wind power-load combined curves The second-stage conditional generation model adopts a conditional generation structure oriented towards sequence modeling and introduces a time-series feature extraction mechanism to enhance the local continuity, time-series correlation, and cross-hour trend expression capabilities of hourly curves.

[0065] The second-stage conditional generation model adopts a dual-channel joint generation structure for wind power and load, and sets a shared temporal convolution module on the time scale after encoding and compression. The shared temporal convolution module uses dilated temporal convolution to extract cross-day dependence and long-term trend features, and its output is then decoded to restore the 168-hour wind power-load joint curve.

[0066] The first stage generates a 168-hour wind power-load joint curve based on the structured scenario indicators generated by the conditional generation model. The second stage utilizes a sequence-oriented modeling-based conditional generation structure and a time-series feature extraction mechanism to maintain the local continuity, temporal correlation, and cross-hourly trend of the hourly curves. Simultaneously, it introduces consistency constraints on the sub-indicators of key weekly-scale wind power indicators to ensure consistency between the weekly-scale risk indicators obtained from the generated curves and the conditional structured indicators. Furthermore, the second stage employs a dual-channel joint generation method for wind power-load, transforming the noisy hourly curves... diffusion step and structured scenario metrics A common input condition is used to generate the structure. A shared temporal convolution module is set on the time scale of the encoded and compressed data. By dilating the temporal convolution, cross-day dependence and long-term trend features are extracted, and then decoded to restore the 168-hour wind power-load joint curve. In one implementation, the temporal convolutional module in the second stage is located at the bottleneck feature after encoding and compression, and is shared by the wind power channel and the load channel. Because this module operates on the joint feature space, its output can simultaneously constrain the shape of the wind power curve, the shape of the load curve, and the synergistic changes between the two, rather than generating wind power and load separately and then splicing them together.

[0067] Example 2 This embodiment provides a device for constructing a wind power-load joint scenario generation model for persistent weather processes, including: an acquisition module, a conversion module, a calculation module, and a training module. The acquisition module acquires historical available wind power output time series, historical load output time series, and persistent weather process conditions for a preset area within a preset time period. The conversion module converts the historical available wind power output time series and its corresponding historical load output time series into historical wind power-load curves. The calculation module calculates historical structured scenario indicators at various time scales using the historical wind power-load curves. The training module inputs the persistent weather process conditions into the initial scenario generation model for training until the model converges, resulting in the target scenario generation model. The initial scenario generation model includes: a first-stage condition generation model and a second-stage condition generation model. The first-stage condition generation model is trained using persistent weather process conditions as input and historical structured scenario indicators as output. The second-stage condition generation model is trained using historical structured scenario indicators as input and historical wind power-load curves as output.

[0068] Example 3 This embodiment provides a method for generating wind power-load joint scenarios for persistent weather events, such as... Figure 5 As shown, it includes: inputting the current weather process conditions into the target scenario generation model, so that the first-stage condition generation model outputs the current structured scenario index, and then inputting the current structured scenario index into the second-stage condition generation model so that it outputs the target wind power-load joint curve, and uses it as the target wind power-load joint scenario.

[0069] For example, the condition dimension of persistent weather processes is 30, and the dimension of structured scenario indicators is 38; the first stage adopts a structured conditional diffusion model with soft constraints on physical structure, and applies noisy structured indicators. The first stage involves modeling daily-scale wind power / load trend branches and weekly-scale wind power persistence branches. The second stage employs an hourly-scale wind power-load joint curve generation model with a shared temporal convolution module. The wind power-load joint scenario generated in this embodiment within a designated area can be used for power planning, energy storage capacity configuration, hydrogen energy storage or long-term energy storage simulation, low-output, high-output, and ramp-up risk analysis of wind power under persistent weather processes, as well as the construction of conditional probabilistic scenarios in power systems with a high proportion of renewable energy.

[0070] As an optional implementation, the above-mentioned method for generating wind power-load joint scenarios for persistent weather processes further includes: performing multi-scale evaluation of the target wind power-load joint scenario. The multi-scale evaluation includes hourly-scale output distribution and autocorrelation evaluation, daily-scale structured index evaluation, weekly-scale persistent risk index evaluation, and a three-dimensional joint structure evaluation of key weekly-scale wind power indices. The three-dimensional joint structure evaluation of key weekly-scale wind power indices uses multiple hours of maximum average ramp intensity, multiple hours of minimum average wind power output, and multiple hours of maximum average wind power output as three-dimensional coordinates to form a real sample point set and a generated sample point set. The difference between the two types of point sets in the joint space of key weekly-scale risk indices is quantified by the three-dimensional optimal transmission distance.

[0071] Specifically, the generated hourly wind power-load curves are evaluated at multiple scales, assessing the ability of the generated scenario to retain features at hourly, daily, and weekly scales. The hourly evaluation includes hourly wind power output distribution, hourly load output distribution, hourly wind power autocorrelation function, and hourly load autocorrelation function. The daily evaluation inversely calculates the generated curves into indicators such as daily average wind power output, intraday wind power uphill intensity, intraday wind power downhill intensity, daily average load output, and load peak-valley difference, and compares these with historical samples. The weekly evaluation inversely calculates the generated curves into key features such as sustained high winds, sustained low winds, and 72-hour uphill characteristics, and compares these with historical samples.

[0072] Using the 72-hour maximum average slope intensity, 72-hour minimum average wind power output, and 72-hour maximum average wind power output of each sample as three-dimensional coordinates, a real sample point set and a generated sample point set are constructed, respectively. The minimum average transport cost between the two types of point sets is calculated using the three-dimensional optimal transmission distance, which is used to quantify the maintenance capability of the joint structure of key weekly-scale wind power risk indicators. The 72-hour maximum average slope intensity, 72-hour minimum average wind power output, and 72-hour maximum average wind power output can be used to construct a three-dimensional joint feature space, and the three-dimensional optimal transmission distance is used to measure the difference between the real sample point set and the generated sample point set. The three-dimensional optimal transmission distance can be expressed as:

[0073] in and These are the empirical distributions corresponding to the real sample point set and the generated sample point set, respectively. and These are three-dimensional feature points.

[0074] Example 4 This embodiment provides a system for generating wind power-load combined scenarios, including: a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for generating wind power-load combined scenarios for persistent weather processes.

[0075] The system can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or models. The processor implements various functions of the system by running or executing the computer programs and / or models stored in the memory, and by accessing data stored in the memory.

[0076] Example 5 This application also relates to a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method for constructing a wind power-load joint scenario generation model for persistent weather processes or the method for generating wind power-load joint scenarios for persistent weather processes provided in the above embodiments.

[0077] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0078] Example 6 This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the method for constructing a wind power-load joint scenario generation model for persistent weather processes or the method for generating wind power-load joint scenarios for persistent weather processes provided in the above embodiments.

[0079] The technical features of the embodiments described above can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. It should be noted that the terms "in one embodiment," "for example," and "as in another example" in this application are intended to illustrate the application and are not intended to limit the application.

[0080] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A method for constructing a wind power-load joint scenario generation model for persistent weather processes, characterized in that, include: S1: Obtain the historical available wind power output time series, historical load output time series, and persistent weather conditions for the preset area within the preset time period; S2: Convert the historical available wind power output time series and its corresponding historical load output time series into historical wind power-load curves; S3: Calculate historical structured scenario indicators at various time scales using the historical wind power-load curves; S4: Input the persistent weather process conditions into the initial scene generation model for training until the model converges to obtain the target scene generation model; wherein, the initial scene generation model includes: a first-stage condition generation model and a second-stage condition generation model; the first-stage condition generation model is trained with the persistent weather process conditions as input and the historical structured scene indicators as output; the second-stage condition generation model is trained with the historical structured scene indicators as input and the historical wind power-load curve as output.

2. The method for constructing a wind power-load joint scenario generation model for persistent weather processes as described in claim 1, characterized in that, The persistent weather conditions include the category information and intensity description of regional average persistent weather processes, in order to distinguish the statistical tendencies, intra-weekly trends and risk characteristics of wind power and load scenarios under different persistent weather processes.

3. The method for constructing a wind power-load joint scenario generation model for persistent weather processes as described in claim 1, characterized in that, The historical structured scene indicators include daily-scale structured scene indicators and weekly-scale structured scene indicators; The daily-scale structured scenario indicators include: daily average wind power output, daily wind power uphill intensity, daily wind power downhill intensity, daily average load output, and daily load peak-valley difference over multiple consecutive days. The weekly-scale structured scenario indicators include: the lowest average wind power output within a multi-hour sliding window, the highest average wind power output within a multi-hour sliding window, and the maximum average wind power ramp intensity within a multi-hour sliding window.

4. The method for constructing a wind power-load joint scenario generation model for persistent weather processes as described in claim 3, characterized in that, The first-stage condition generation model is a structured condition diffusion model; persistent weather process conditions. and diffusion step After being embedded, the historical structured scene indicators are used as global conditional inputs. This includes: the daily-scale wind power-load trend component corresponding to the daily-scale structured scenario indicators, and the weekly-scale wind power sustainability component corresponding to the weekly-scale structured scenario indicators.

5. The method for constructing a wind power-load joint scenario generation model for persistent weather processes as described in claim 4, characterized in that, The daily-scale wind power-load trend section includes a wind power daily trend branch and a load daily trend branch; The wind power daily trend branch is used to characterize the daily average wind power, uphill and downhill wind power, and the corresponding cross-day changes. The load daily trend branch is used to characterize the daily average load and the peak-valley difference, etc., cross-day changes. The wind power daily trend branch, the load daily trend branch, and the weekly wind power persistence part are bidirectionally fused to form a structured indicator output that takes into account both daily trend and weekly persistence risk.

6. The method for constructing a wind power-load joint scenario generation model for persistent weather processes as described in claim 4, characterized in that, The first-stage conditional generation model introduces soft constraints on physical structure during training; The physical structure soft constraints include weekly-scale index sequence constraints and daily-scale at most-day window proxy consistency constraints. The weekly-scale index sequence constraint is used to constrain the minimum average wind power output over multiple days to not be higher than the maximum average wind power output over multiple days, and to constrain the wind power ramping intensity to be within a physically reasonable range. The daily-scale to multi-day window proxy consistency constraint utilizes the sliding window proxy formed by the daily average wind power values ​​over multiple consecutive days to ensure that the generated multi-day minimum average wind power index, multi-day maximum average wind power index, and weekly daily-scale wind power trend maintain a structural correlation.

7. A device for constructing a wind power-load joint scenario generation model for persistent weather processes, characterized in that, include: The acquisition module is used to acquire the historical available wind power output time series, historical load output time series, and persistent weather conditions for a preset area within a preset time period. The conversion module is used to convert the historical available wind power output time series and its corresponding historical load output time series into historical wind power-load curves. The calculation module is used to calculate historical structured scenario indicators at various time scales using the historical wind power-load curves. The training module is used to input the persistent weather process conditions into the initial scene generation model for training until the model converges, thereby obtaining the target scene generation model. The initial scene generation model includes a first-stage condition generation model and a second-stage condition generation model. The first-stage condition generation model is trained using the persistent weather process conditions as input and the historical structured scene indicators as output. The second-stage condition generation model is trained using the historical structured scene indicators as input and the historical wind power-load curve as output.

8. A method for generating wind power-load joint scenarios for persistent weather processes, characterized in that, include: The current weather conditions are input into the target scenario generation model according to any one of claims 1-6, so that the first-stage condition generation model outputs the current structured scenario index, and then the current structured scenario index is input into the second-stage condition generation model so that it outputs the target wind power-load joint curve, and uses it as the target wind power-load joint scenario.

9. The method for generating wind power-load joint scenarios for persistent weather processes as described in claim 8, characterized in that, Also includes: The target wind power-load combined scenario is evaluated on a multi-scale basis. The multi-scale evaluation includes hourly-scale output distribution and autocorrelation evaluation, daily-scale structured index evaluation, weekly-scale persistent risk index evaluation, and three-dimensional joint structure evaluation of key weekly-scale wind power indicators. The three-dimensional joint structure evaluation of key weekly-scale wind power indicators uses multiple hours of maximum average ramp intensity, multiple hours of minimum average wind power output, and multiple hours of maximum average wind power output as three-dimensional coordinates to form a real sample point set and a generated sample point set. The difference between the two types of point sets in the joint space of key weekly-scale risk indicators is quantified by the three-dimensional optimal transmission distance.

10. A system for generating wind power-load combined scenarios, characterized in that, include: A memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method for generating wind power-load combined scenarios for persistent weather processes as described in claim 7 or 8.

Images