New energy power prediction method and system based on electricity price coupling and climbing constraint

Abstract

This invention proposes a method and system for predicting renewable energy power based on electricity price coupling and ramping constraints, belonging to the field of renewable energy power prediction technology. The method includes: calculating the bidding space of thermal power units based on power system data, constructing a bidding space difference, dividing time periods and performing one-hot encoding, introducing time-period features, seasonal ramping correction coefficients, ramping capacity scarcity indicators, and lagging ramping indices to construct multi-dimensional ramping features; fusing the basic features and ramping features and performing multi-dimensional similarity calculations to obtain similar daily datasets, including weighted fusion of Euclidean distance, covariance-reduced distance, and ramping sequence correlation coefficients to obtain a comprehensive similarity; extracting time-series features and inputting them into a random forest model for regression prediction to obtain the predicted renewable energy power value. This invention integrates thermal power unit ramping capacity constraints to improve prediction accuracy.

Description

Technical Field

[0001] This invention belongs to the field of new energy power prediction technology, and in particular relates to a new energy power prediction method and system based on electricity price coupling and ramping constraints. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous increase in the proportion of installed capacity of new energy sources, the prediction of new energy power, such as photovoltaic power, plays a key role in the optimized scheduling of power systems, the consumption of new energy sources, and the safe and stable operation of power grids. Traditional time series forecasting methods (such as autoregressive integrated moving average models and support vector machines) are difficult to capture the nonlinear characteristics and complex time series dependencies between new energy power data and power system constraints, and often suffer from problems such as overfitting, insufficient accuracy, or lagging response to dynamic constraints.

[0004] Predictions for renewable energy power output have gradually shifted from traditional machine learning methods to more refined time-series modeling. In recent years, time-series modeling models (such as Long Short-Term Memory Neural Networks and Random Forests) have made progress in processing time-series data; however, many have not fully integrated core constraints of the power system, particularly the dynamic impact of thermal power unit ramp-up constraints on renewable energy power prediction. This results in insufficient model generalization ability and a lack of mechanistic interpretability in the prediction results. Furthermore, existing models have incomplete feature systems and have not specifically constructed ramp-up related features, making it difficult to capture the differences in the correlation between ramp-up constraints and renewable energy power output across different time periods and seasons. This makes it difficult to meet the needs of refined power system scheduling in scenarios with high renewable energy penetration. Summary of the Invention

[0005] To overcome the shortcomings of the existing technologies, this invention proposes a new energy power prediction method and system based on electricity price coupling and ramping constraints. This method addresses the problems of weak feature extraction capabilities, insufficient consideration of the impact of ramping constraints on thermal power units, and low prediction accuracy in new energy power prediction. It effectively integrates the ramping capacity constraints of thermal power units, improves the model's ability to capture complex time-series dependencies, and enhances prediction accuracy.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, this invention discloses a new energy power prediction method based on electricity price coupling and ramping constraints, comprising: Acquire power system data and construct basic features; Based on the power system data, the bidding space of thermal power units is calculated, the bidding space difference is constructed as a proxy variable for ramping demand, the peak ramping period, low ramping period and afternoon abnormal period are divided and uniquely encoded, and time-segmented features, seasonal ramping correction coefficient, ramping capacity scarcity indicator and lagged ramping index are introduced to construct multi-dimensional ramping features. The basic features and climbing features are fused and multidimensional similarity calculation is performed to obtain a similar day dataset. The similarity calculation includes using Euclidean distance and covariance distance to characterize the numerical differences between features, and obtaining the climbing sequence correlation coefficient based on the bidding space difference. The Euclidean distance, covariance distance and climbing sequence correlation coefficient are weighted and fused to obtain a comprehensive similarity. The Transformer model is used to extract time-series features from similar day datasets, and these time-series features are then input into a random forest model for regression prediction to obtain the predicted value of new energy power.

[0007] Secondly, this invention discloses a new energy power prediction system based on electricity price coupling and ramping constraints, comprising: The data preprocessing module is configured to: acquire power system data and construct basic features; The ramp feature construction module is configured to: calculate the bidding space of thermal power units based on the power system data, construct the bidding space difference as a proxy variable for ramp demand, divide the peak ramp period, low ramp period and afternoon abnormal period and perform unique hot coding, introduce time-segmented features, seasonal ramp correction coefficient, ramp capacity scarcity identifier and lagged ramp index, and construct multi-dimensional ramp features. The similar day filtering module is configured to: fuse the basic features and the climbing features and perform multi-dimensional similarity calculation to obtain a similar day dataset. The similarity calculation includes using Euclidean distance and covariance distance to characterize the numerical differences between features, and obtaining the climbing sequence correlation coefficient based on the bidding space difference. The Euclidean distance, covariance distance and climbing sequence correlation coefficient are weighted and fused to obtain a comprehensive similarity. The time series prediction module is configured to: extract time series features from similar daily datasets using a Transformer model, and input the time series features into a random forest model for regression prediction to obtain the predicted value of new energy power.

[0008] Thirdly, the present invention discloses an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, they complete the steps of the above-mentioned new energy power prediction method based on electricity price coupling and ramping constraints.

[0009] Fourthly, the present invention discloses a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the above-mentioned new energy power prediction method based on electricity price coupling and ramping constraints.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a multi-dimensional climbing feature system, combining time-segmented analysis, seasonal correction, and lag effect capture to accurately characterize the correlation between climbing constraints and photovoltaic power under different scenarios, improving the model's ability to extract features with complex time-series dependencies. Simultaneously, by combining Euclidean distance and covariance constraint distance to characterize numerical differences, the selected historical data better matches the characteristics of the target day, effectively improving the efficiency of model training and prediction accuracy. Regarding the selection of prediction models, a transformer based on attention mechanism freezing and random forest are used as core predictors. These not only possess strong nonlinear fitting capabilities but also quantify the contribution of climbing features, enhancing the mechanism interpretability of the prediction results and solving the problem of insufficient interpretability in traditional model predictions.

[0011] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0013] Figure 1 This is an overall block diagram of the new energy power prediction method based on electricity price coupling and ramping constraints described in Embodiment 1 of the present invention.

[0014] Figure 2 This is a flowchart of the new energy power prediction method based on electricity price coupling and ramping constraints described in Embodiment 1 of the present invention.

[0015] Figure 3 This is a flowchart of the similar day screening mechanism described in Embodiment 1 of the present invention.

[0016] Figure 4 This is a flowchart of the time series prediction process described in Embodiment 1 of the present invention.

[0017] Figure 5 This is a structural block diagram of the new energy power prediction system based on electricity price coupling and ramping constraints as described in Embodiment 2 of the present invention. Detailed Implementation

[0018] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0020] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0021] Example 1 In one or more embodiments, a new energy power prediction method based on electricity price coupling and ramping constraints is disclosed, such as... Figures 1-2 As shown, it includes the following steps: Step S1: Acquire power system data and construct basic features.

[0022] Power system data, including renewable energy output data and related disclosure data, is obtained from the power trading center data platform. Specifically, renewable energy output data includes information on direct-dispatch load, local power plant power generation output, photovoltaic power output, wind power output, interconnection line receiving load, self-owned unit output, and non-market-based nuclear power output. Related disclosure data includes load information, interconnection line information, and reserve information, and an original dataset is constructed.

[0023] By integrating the above-mentioned raw data and output variables, a multidimensional input dataset is formed:

[0024] in, As a multidimensional variable, d This represents the dimensionality of the input features, i.e., the number of features in a single time-series sample. Considering the differences in the dimensions of different features, and to avoid interfering with model training, all features are standardized using a min-max normalization method.

[0025] in, The original data values, , The first The minimum and maximum values ​​of each feature are found among all data points. Missing and outlier values ​​are also handled to ensure data continuity and stability.

[0026] As one implementation method, during the training phase, new energy output data and various disclosed data collected from the power trading center system data platform are processed. All information is collected at one-hour intervals. The specific dataset composition includes time, positive reserve, negative reserve, direct-controlled load, interconnection line load, total wind power load, total photovoltaic power load, total local power plant power generation load, total non-market nuclear power load, total self-owned unit load, total test unit load, North China power inflow, Qingdong DC, Zhaoyi DC, Yindong DC, and Lugu DC, forming 15 columns of input features and 2 output features. The first 15 columns of features are used to train and predict photovoltaic and wind power output features, integrating them into 17 columns of input data, including 15 columns of disclosed data and 2 columns of output features. The data is then divided into training and validation sets according to time series order.

[0027] This completes the data preprocessing and dataset construction, providing foundational data support for subsequent hill-climb feature construction and prediction model training.

[0028] Step S2: Calculate the bidding space of thermal power units based on the power system data, construct the bidding space difference as a proxy variable for ramp demand, divide the peak ramp period, low ramp period and afternoon abnormal period and perform unique thermal coding, introduce time period characteristics, seasonal ramp correction coefficient, ramp capacity scarcity indicator and lagged ramp index to construct multi-dimensional ramp characteristics.

[0029] This embodiment uses the bidding space as a reference variable for the bidding of thermal power units, which achieves both electricity price coupling and thermal power unit ramp-up constraints.

[0030] Step S2-1: Based on the power balance relationship of the power system, define the bidding space for thermal power units as follows:

[0031] in, This indicates the bidding space for thermal power units. For direct load adjustment, For local power plants to generate total electricity, , These are the total photovoltaic power and the total wind power, respectively. For the power receiving load of the tie line, For self-owned generating units, This is for the total addition of non-market-based nuclear power.

[0032] From the above formula, we can see the bidding space. Essentially, this reflects the remaining power demand after meeting system load and the output of various non-thermal power sources, i.e., the adjustable capacity that thermal power units need to handle. Therefore, This can be viewed as an upper limit constraint on the output demand of thermal power units.

[0033] Furthermore, analyzing the regulation behavior of thermal power units from a dynamic operation perspective reveals that its core constraint is the ability to climb slopes, namely:

[0034] in, The maximum ramping rate limit for thermal power units; This represents the output of the thermal power unit. Since the thermal power unit must prioritize meeting the system balance requirements during actual operation, it can be approximated as follows:

[0035] Therefore, the change in the output of thermal power units can be characterized by the change in the bidding space to achieve electricity price coupling, which can be used to characterize the dynamic adjustment needs of thermal power units, i.e.:

[0036] in, This represents the bidding space difference, which serves as the dynamic adjustment ramping constraint for the power system. This is in contrast to the static bidding space. This difference variable can more directly reflect the dynamic dependence of the system on its climbing ability.

[0037] Therefore, This actually reflects the changing demand of the system on the regulation capacity of thermal power units between adjacent time points, and its physical meaning is consistent with the ramp-up / ramp-down behavior of thermal power units. A larger value indicates that the system has placed a higher demand on the thermal power unit for adjustment, corresponding to stronger ramp-up pressure; conversely, a smaller value indicates a more moderate adjustment demand.

[0038] Therefore, the bidding space difference As a proxy variable for the ramp-up demand of thermal power units, it can effectively characterize the dynamic regulation and constraint characteristics of the power system without directly obtaining the unit operating parameters.

[0039] Step S2-2: Expand the climbing information from multiple dimensions such as time structure, periodic characteristics and operating status.

[0040] First, considering the significantly different regulation demands of the power system at different times, the time is divided into peak periods, off-peak periods, and afternoon abnormal fluctuation periods, and time-period characteristics are constructed:

[0041] in, This is a time-segmented feature, where t represents time; this feature is used to enhance the model's ability to identify typical operating scenarios.

[0042] Secondly, to characterize the impact of seasonal variations on the output of new energy sources and the system's regulation capacity, a seasonal ramp-up correction coefficient is constructed:

[0043] in, Indicates month or season category; This is a seasonal climbing correction factor; This is a seasonally corrected mapping function. This feature can reflect the differences in photovoltaic power output fluctuations and thermal power regulation pressures under different seasons. Furthermore, to capture the time dependence of ramp-up behavior, a lag ramp-up index is constructed:

[0044] in, This is a lagged climbing feature; this feature can reflect the impact of historical adjustment processes on the current state, thereby enhancing the model's ability to model time-series dynamics.

[0045] In addition, to characterize the stress level of the system's regulatory capacity, a scarcity marker for climbing ability is constructed. :

[0046] in, A preset threshold is used. When the ramping demand exceeds a certain level, the system is considered to be in a state of strained regulation capacity. Based on this, the ramping constraint shadow price or marginal regulation cost can be further introduced as an auxiliary feature. Specifically, based on the power market clearing model or the unit economic dispatch results, the shadow price (i.e., the constraint dual variable) corresponding to the ramping constraint of thermal power units is calculated, or the marginal cost corresponding to the unit regulation power is calculated. These are quantified as scalar features and added to the model input to reflect the regulation cost and economic signal of the system under different operating conditions, thereby reflecting the regulation cost of the system under different operating conditions.

[0047] Step S2-3: Fuse the above multi-source features to construct a multi-dimensional climbing feature vector:

[0048] Relying solely on a single difference variable is insufficient to fully characterize the complex regulation and constraint characteristics of a power system. This feature system not only includes a direct characterization of the ramp-up demand of thermal power units, but also comprehensively reflects the time structure, seasonal influences, and system operating status, thereby achieving a multi-dimensional expression of the regulation and constraint of the power system.

[0049] Compared with traditional modeling methods that rely solely on meteorological or historical power output characteristics, this invention effectively enhances the model's ability to characterize the power fluctuation mechanism of new energy sources by explicitly transforming the climbing ability constraint of thermal power units into structured features. This provides more physically meaningful input information for subsequent screening of similar days and training of prediction models.

[0050] This invention breaks through the traditional modeling method that relies solely on meteorological and historical power output information. It explicitly introduces the ramp-up capability of thermal power units from implicit operational constraints into the prediction framework, and uses bidding space difference to characterize the system ramp-up demand. By integrating time-period characteristics, seasonal correction coefficients, and lag effects, a multi-dimensional ramp-up characteristic system is constructed. This system characterizes the dynamic coupling relationship between the power system's regulation capability and new energy output from a mechanistic perspective, thus making up for the problem of insufficient consideration of system flexibility constraints in existing methods.

[0051] Step S3: The basic features and climbing features are fused and multi-dimensional similarity calculation is performed to obtain a similar daily dataset to realize the climbing constraint of the thermal power unit. The similarity calculation includes using Euclidean distance and covariance distance to characterize the numerical difference between features, and obtaining the climbing sequence correlation coefficient based on the bidding space difference. The Euclidean distance, covariance distance and climbing sequence correlation coefficient are weighted and fused to obtain the comprehensive similarity.

[0052] Based on the construction of multi-dimensional climbing features, to further improve the consistency between the model training samples and the target prediction scene, this invention introduces a similar day screening mechanism, such as... Figure 3 As shown, historical data is selectively filtered to provide highly relevant input for subsequent prediction models.

[0053] Step S3-1: Directly concatenate and fuse the basic features and the climbing features to construct a unified feature representation vector:

[0054] in, For the first The fusion characteristics of moments Represents a multidimensional input dataset. This represents the climbing characteristics. The two are directly concatenated according to their dimensions to form a high-dimensional fused feature.

[0055] Step S3-2: Construct a similarity assessment method that integrates multi-dimensional information to perform multi-dimensional similarity calculations and obtain a similar day dataset.

[0056] This invention takes into account that traditional similar day screening methods rely on a single distance metric, which makes it difficult to simultaneously characterize the differences in feature values ​​and the temporal changes. Therefore, it comprehensively characterizes sample similarity from two aspects: numerical consistency and consistency of change trends.

[0057] First, from the perspective of feature space, Euclidean distance is used to characterize the overall numerical difference:

[0058] in, Let Euclidean distance be the distance between vectors. with vector The L2 norm, this distance reflects the absolute deviation between the target date and historical dates in multidimensional characteristics. For the fused feature vector of the target day, This is the fusion feature vector of historical days.

[0059] Secondly, to eliminate the impact of correlation between different features on distance calculation, a distance metric based on the covariance matrix is ​​introduced:

[0060] in, Covariance distance; This is the feature covariance matrix. This distance can adaptively weight different dimensions, thereby improving the robustness of the similarity measure.

[0061] Furthermore, from the perspective of time series structure, a correlation index based on the climbing difference sequence is introduced to characterize the consistency of the curve's changing trend. The correlation coefficient between the climbing sequences of the target day and historical days is defined as:

[0062] in, The correlation coefficient of the climbing sequence; This refers to the bidding space difference, i.e., the ramp-up difference sequence. For the target day The bidding space difference at any given moment, The mean of the price difference sequence for the entire day of the target date. For the first time in history The bidding space difference at any given moment, This represents the mean of the historical daily price range difference sequence. Standard deviation of the target day's bidding space difference sequence; This represents the standard deviation of the historical daily bidding spatial difference sequence. This indicator reflects the degree of consistency in the changing trends of two sets of time series data, thus compensating for the inability of simple numerical distance to characterize the curve shape. Based on this, multiple measurement methods are integrated to construct a comprehensive similarity function. :

[0063] in, , , These are the weighting coefficients.

[0064] Meanwhile, to strengthen the dominant role of climbing characteristics in the screening of similar days, a priority screening mechanism is set up to retain historical samples that meet the following conditions:

[0065] in, A preset threshold is used. Based on the constraints, historical samples are ranked according to their overall similarity, and the samples with the highest similarity are selected. A set of similar days is formed by these historical dates:

[0066] The data corresponding to the historical days obtained from the set of similar days constitute the similar day dataset.

[0067] Through the above methods, this invention achieves collaborative screening of historical samples at both the numerical characteristics and variation patterns levels. Furthermore, by introducing a ramp-up differential constraint, the screening results more closely reflect the actual operating conditions of the power system. Compared with traditional similar-day screening methods, this invention not only improves the accuracy of sample matching but also enhances the model's adaptability to complex operating scenarios such as high fluctuations and strong ramp-ups, providing more physically consistent input data for subsequent prediction models.

[0068] The weighted similarity day screening method with climbing constraints proposed in this invention introduces a climbing differential priority matching mechanism based on the collaborative modeling of basic features and climbing features. By setting a similarity threshold, it strengthens the constraint on the consistency of system operating status, effectively improves the targeting and representativeness of sample screening, and enhances the adaptability of the prediction model to complex operating conditions from the data level.

[0069] Step S4: Based on the similar day dataset, construct a time-series input sequence, extract time-series features using a Transformer model, and then input the time-series features into a random forest model for regression prediction to obtain the predicted value of new energy power. The input data for the Transformer model is a fusion feature sequence extracted from the similar day dataset and arranged in chronological order. This sequence contains continuous historical data before the target day and incorporates operating condition features from similar days, thus providing the model with an input that combines temporal continuity and operating condition similarity.

[0070] Specifically, in the time series prediction stage, such as Figure 4 As shown, based on the multidimensional input of fused features, a Transformer feature extraction network with an attention mechanism and parameter freezing is constructed. Its core structure and operation are as follows: (1) Network structure: including input embedding layer, multi-head self-attention layer, and feedforward network layer; wherein, the multi-head self-attention layer is located between the embedding layer and the feedforward network layer, and is used to calculate the correlation weight of features at different times in the sequence and capture key temporal dependencies; (2) Parameter freezing mechanism: During the model training phase, some parameters of the multi-head self-attention layer are frozen to enable it to focus on learning long-term time-series dependency patterns; during the prediction phase, these parameters are unfrozen to enable it to dynamically adapt to short-term output fluctuations. (3) Feature extraction and prediction: After extracting stable time series representations through the above network, the data are input into the random forest model for regression prediction, achieving efficient fitting of nonlinear relationships, quantifying the contribution of climbing features to the prediction results, and outputting photovoltaic and wind power prediction values.

[0071] This embodiment extracts a stable time series representation and inputs it into a random forest model for regression prediction, achieving efficient fitting of nonlinear relationships and quantifying the contribution of climbing features to the prediction results. It then outputs predicted photovoltaic and wind power values, including the following steps: Step S4-1: Temporal feature extraction. After obtaining high-quality samples, a Transformer model based on the attention mechanism is constructed to model the temporal features.

[0072] The fusion features of the fusion feature sequences within consecutive historical time windows extracted from similar day datasets are used as input to the Transformer model:

[0073] in, For temporal embedding features, This represents the fusion characteristics from time period tk to t.

[0074] Specifically, during the model training phase, in order to enhance the model's ability to learn long-term stable patterns, the attention mechanism parameters are frozen, so that the model focuses on learning long-term dependencies.

[0075] In the prediction phase, the latest observation data is introduced and the attention parameter is unfrozen, enabling the model to dynamically adapt to short-cycle fluctuations, thereby achieving collaborative modeling of long-cycle trends and short-cycle changes.

[0076] Step S4-2: After obtaining the time-series features extracted by the transformer (Transformer model), use them as input to the random forest model for regression prediction to obtain the predicted value of new energy power.

[0077] in, Indicates the first The predicted power value at time; This represents the timing embedding characteristics of the transformer output; This represents a random forest regression model.

[0078] In its implementation, a random forest consists of multiple decision trees, and its prediction result is the average of the outputs of each sub-model.

[0079] in, Indicates the number of decision trees; Indicates the first The prediction results of the decision tree.

[0080] To further enhance the model's ability to utilize the original information, the output features of the Transformer and the fused features can be used together as input to the random forest, i.e.:

[0081] in, Indicates the first The fusion characteristics of moments.

[0082] Meanwhile, the random forest model has the ability to evaluate feature importance. By calculating the contribution of each feature in the decision tree splitting process, a feature importance index can be obtained:

[0083] in, Indicates the first The importance of each feature; This indicates that the feature is in the first... Contribution in each decision tree.

[0084] Based on the above-mentioned importance of features, the influence of the climbing features on the prediction results can be quantified, thereby enabling an explanation of the climbing constraint effect of thermal power units.

[0085] The final output includes predicted values ​​of renewable energy power and information on its characteristic contribution, providing decision support for power system operation and dispatch.

[0086] This invention uses a random forest model as the core predictor, focusing on mining the nonlinear mapping relationship between the ramp-up characteristics of thermal power units and the power of new energy sources. This not only improves the prediction accuracy, but also enables quantitative analysis of the degree of ramp-up constraint through feature importance assessment, thereby enhancing the interpretability and physical consistency of the prediction results.

[0087] This invention first processes multi-source data and then achieves a collaborative expression of the prediction results and the impact of ramp-up constraints at the output layer. This allows the prediction results to not only reflect the changing trends of new energy output but also reveal the underlying reasons for its constraint by the system's regulation capacity, thereby enhancing the model's application value and decision support capabilities in the actual power market environment.

[0088] Preferably, this also includes model evaluation and results disclosure: After model training is completed, the model's predictive performance is evaluated using a validation set. By calculating error metrics between predicted and true values, including mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²), the accuracy and stability of the model are comprehensively evaluated.

[0089] This invention, based on the operating mechanism of power systems, integrates the ramp-up capability constraints of thermal power units into the prediction of renewable energy power output. By constructing a multi-dimensional ramp-up feature system, incorporating a multi-scale similar day screening method, and combining it with time-series modeling based on an attention-based freezing and unfreezing strategy, it achieves accurate prediction of renewable energy output. Furthermore, it utilizes random forests to quantify feature contribution, improving the interpretability of the results. This method effectively improves prediction accuracy and adaptability to complex scenarios, demonstrating significant engineering application value.

[0090] The above technical solution, in the feature construction stage, innovatively transforms the ramp-up capacity of thermal power units from an operational constraint into an explicit modeling feature. It uses the bidding space difference as a proxy variable for ramp-up demand, integrating time-period identifiers, seasonal correction coefficients, time-series lag relationships, and scarcity state identifiers to construct a multi-dimensional ramp-up feature system. This system characterizes the dynamic coupling relationship between power system regulation capacity and renewable energy output from a mechanistic perspective. In the similarity day screening stage, a similarity assessment method integrating multi-scale metrics is adopted. Based on the normalization of basic and ramp-up features, it combines Euclidean distance and covariance constraint distance to characterize numerical differences. It also introduces ramp-up difference and sequence correlation to represent time-series changes. Through weight allocation and threshold constraint mechanisms, it prioritizes the screening of historical samples with highly consistent ramp-up characteristics, thus taking into account numerical similarity. To ensure consistency with the curve shape and improve sample matching quality, a Transformer feature extraction network with an attention mechanism and parameter freezing is constructed in the time series prediction stage. This network performs deep time series representation learning on the fused features and uses its output as input to the Random Forest model for regression prediction. This fully leverages the Transformer's ability to model time series dependencies and the Random Forest's advantages in nonlinear mapping and feature importance quantification, achieving a synergistic improvement in prediction accuracy and interpretability. After model training, performance is evaluated by calculating metrics such as MAE, RMSE, and R² on the validation set. The prediction results and the contribution of the climbing constraint are output in conjunction with a visual report and structured disclosure data, achieving a unified expression of prediction results and mechanism attribution.

[0091] Example 2 In one or more embodiments, a new energy power prediction system based on electricity price coupling and ramping constraints is disclosed, such as... Figure 5 As shown, it specifically includes: The data preprocessing module is configured to: acquire power system data and construct basic features; The ramp feature construction module is configured to: calculate the bidding space of thermal power units based on the power system data, construct the bidding space difference as a proxy variable for ramp demand, divide the peak ramp period, low ramp period and afternoon abnormal period and perform unique hot coding, introduce time-segmented features, seasonal ramp correction coefficient, ramp capacity scarcity identifier and lagged ramp index, and construct multi-dimensional ramp features. The similar day filtering module is configured to: fuse the basic features and the climbing features and perform multi-dimensional similarity calculation to obtain a similar day dataset. The similarity calculation includes using Euclidean distance and covariance distance to characterize the numerical differences between features, and obtaining the climbing sequence correlation coefficient based on the bidding space difference. The Euclidean distance, covariance distance and climbing sequence correlation coefficient are weighted and fused to obtain a comprehensive similarity. The time series prediction module is configured to: extract time series features from similar daily datasets using a Transformer model, and input the time series features into a random forest model for regression prediction to obtain the predicted value of new energy power.

[0092] The results disclosure module is configured to generate a visual report that includes time-series change trends, prediction error distribution, feature importance, and analysis of the impact of ramp constraints. It also outputs structured disclosure data containing attribution information of ramp constraints, thus achieving a synergistic presentation of prediction results and regulatory constraints, and providing support for power system dispatching and operation decisions.

[0093] Example 3 This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the above-mentioned new energy power prediction method based on electricity price coupling and ramping constraints.

[0094] Example 4 This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they complete the steps of the above-mentioned new energy power prediction method based on electricity price coupling and ramping constraints.

[0095] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0096] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0098] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A new energy power prediction method based on electricity price coupling and ramping constraints, characterized in that, include: Acquire power system data and construct basic features; Based on the power system data, the bidding space of thermal power units is calculated, the bidding space difference is constructed as a proxy variable for ramping demand, the peak ramping period, low ramping period and afternoon abnormal period are divided and uniquely encoded, and time-segmented features, seasonal ramping correction coefficient, ramping capacity scarcity indicator and lagged ramping index are introduced to construct multi-dimensional ramping features. The basic features and climbing features are fused and multidimensional similarity calculation is performed to obtain a similar day dataset. The similarity calculation includes using Euclidean distance and covariance distance to characterize the numerical differences between features, and obtaining the climbing sequence correlation coefficient based on the bidding space difference. The Euclidean distance, covariance distance and climbing sequence correlation coefficient are weighted and fused to obtain a comprehensive similarity. The Transformer model is used to extract time-series features from similar day datasets, and these time-series features are then input into a random forest model for regression prediction to obtain the predicted value of new energy power.

2. The renewable energy power prediction method based on electricity price coupling and ramping constraints as described in claim 1, characterized in that, The bidding space is: in, This indicates the bidding space for thermal power units. For direct load adjustment, For local power plants to generate total electricity, , These are the total photovoltaic power and the total wind power, respectively. For the power receiving load of the tie line, For self-owned generating units, For non-market-based nuclear power total additions; Construct the bid space difference as a proxy variable for ramp-up demand: in, This represents the bidding space difference, which serves as the dynamic adjustment ramping constraint for the power system.

3. The renewable energy power prediction method based on electricity price coupling and ramping constraints as described in claim 1, characterized in that, The scarce climbing ability is identified as follows: in, The preset threshold; The lagging climb index is: in, It exhibits a delayed climbing characteristic. This represents the bidding space difference, where t is time t.

4. The renewable energy power prediction method based on electricity price coupling and ramping constraints as described in claim 1, characterized in that, Specifically, the use of Euclidean distance and covariance distance to characterize the numerical differences between features is as follows: The overall numerical differences are characterized using Euclidean distance: in, For Euclidean distance, For the fused feature vector of the target day, This represents the fused feature vector for historical days; Distance metrics based on covariance matrix: in, Covariance distance; Let be the characteristic covariance matrix.

5. The renewable energy power prediction method based on electricity price coupling and ramping constraints as described in claim 1, characterized in that, The method for obtaining the correlation coefficient of the climbing sequence based on the bidding space difference includes: The correlation coefficient between the target date and the historical date climbing sequence is defined as follows: in, The correlation coefficient of the climbing sequence; For the bidding space difference, For the target day The bidding space difference at any given moment, The mean of the price difference sequence for the entire day of the target date. For the first time in history The bidding space difference at any given moment, This represents the mean of the historical daily price range difference sequence. Standard deviation of the target day's bidding space difference sequence; The standard deviation of the historical daily bidding space difference sequence.

6. The renewable energy power prediction method based on electricity price coupling and ramping constraints as described in claim 5, characterized in that, Based on the premise that the correlation coefficient of the climbing sequence is greater than a preset threshold, historical samples are sorted according to comprehensive similarity, and the top samples with the best similarity are selected. A set of similar days is formed from historical days, resulting in a similar day dataset.

7. The renewable energy power prediction method based on electricity price coupling and ramping constraints as described in claim 1, characterized in that, The Transformer model includes an input embedding layer, a multi-head self-attention layer, and a feedforward network layer; wherein, the multi-head self-attention layer is located between the embedding layer and the feedforward network layer, and is used to calculate the correlation weights of features at different time points within the sequence to capture key temporal dependencies; The parameters of the multi-head self-attention layer are frozen during the model training phase and unfrozen during the prediction phase.

8. A new energy power prediction system based on electricity price coupling and ramping constraints, characterized in that, include: The data preprocessing module is configured to: acquire power system data and construct basic features; The ramp feature construction module is configured to: calculate the bidding space of thermal power units based on the power system data, construct the bidding space difference as a proxy variable for ramp demand, divide the peak ramp period, low ramp period and afternoon abnormal period and perform unique hot coding, introduce time-segmented features, seasonal ramp correction coefficient, ramp capacity scarcity identifier and lagged ramp index, and construct multi-dimensional ramp features. The similar day filtering module is configured to: fuse the basic features and the climbing features and perform multi-dimensional similarity calculation to obtain a similar day dataset. The similarity calculation includes using Euclidean distance and covariance distance to characterize the numerical differences between features, and obtaining the climbing sequence correlation coefficient based on the bidding space difference. The Euclidean distance, covariance distance and climbing sequence correlation coefficient are weighted and fused to obtain a comprehensive similarity. The time series prediction module is configured to: extract time series features from similar daily datasets using a Transformer model, and input the time series features into a random forest model for regression prediction to obtain the predicted value of new energy power.

9. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it performs the new energy power prediction method based on electricity price coupling and ramping constraints as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the new energy power prediction method based on electricity price coupling and ramping constraints as described in any one of claims 1-7.

Images