New energy output probability prediction method and device and electronic equipment

Abstract

The invention discloses a new energy output probability prediction method and device and electronic equipment, which are applied to a probability prediction model and comprise a feature extraction module, a feature fusion module, a probability prediction module and a result output module. The method comprises the following steps: acquiring day multi-source data; extracting meteorological element features of the weather forecast data in a first preset time period through a feature extraction module, extracting image features of the remote sensing image data, and extracting time sequence features of the power data; fusing the meteorological element features, the image features and the time sequence features through a feature fusion module to obtain target fusion feature data; processing the target fusion feature data with the time sequence through a probability prediction module to obtain probability distribution of new energy power generation power; and analyzing the probability distribution through a result output module, and outputting a prediction result. According to the method, the complete probability distribution of the future output potential can be directly output, and a more comprehensive decision basis is provided for market participants.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method, device and electronic equipment for predicting the probability of new energy output. Background Technology

[0002] With the global energy structure transformation, the installed capacity of new energy power generation, represented by wind and solar power, has continued to grow rapidly. However, wind and solar power generation are characterized by strong intermittency, volatility, and randomness, posing serious challenges to traditional power systems that rely on controllable power sources, including the time mismatch between power generation and consumption within a day and seasonally. In the electricity spot market, power generation companies need to declare their day-ahead, intraday, and real-time trading volumes based on forecasts of future output. In this market environment, the accuracy of forecasting is no longer merely a technical requirement for grid dispatch, but a core element directly related to the economic benefits of market participants.

[0003] In the spot market, forecasting errors can lead to severe economic consequences. If a power plant's actual output deviates from its declared trading plan, it will face punitive settlements. Therefore, high-precision output forecasting is crucial for new energy power generation companies to reduce financial risk, improve economic efficiency, and enhance market competitiveness in the spot market. Traditional, single-point forecasting methods cannot provide information about the uncertainty of forecast results, making it difficult for market participants to assess risk and formulate optimal trading strategies. Summary of the Invention

[0004] This disclosure provides a method, apparatus, equipment, and storage medium for predicting the probability of new energy output, in order to at least solve the above-mentioned technical problems existing in the prior art.

[0005] According to a first aspect of this application, a method for probabilistic prediction of new energy output is provided, applied to a probabilistic prediction model, wherein the probabilistic prediction model includes a feature extraction module, a feature fusion module, a probabilistic prediction module, and a result output module; the method includes: The acquired weather forecast data, remote sensing image data, and power data are input into the probabilistic prediction model; The feature extraction module extracts meteorological features from the weather forecast data in a first preset time period, image features from the remote sensing image data, and time-series features from the power data. The feature fusion module fuses the meteorological element features, image features, and temporal features to obtain target fused feature data with temporal sequence. The probability prediction module processes the time-series target fusion feature data to obtain the probability distribution of new energy power generation. The probability distribution is analyzed by the result output module, and the prediction result is output.

[0006] In one possible implementation, the weather forecast data is data within a first preset time period; the weather forecast data is high-resolution numerical weather forecast data. The remote sensing data includes satellite remote sensing data and ground-based remote sensing data; The power data includes historical power output data and historical market data.

[0007] In one embodiment, the feature extraction module includes a time-series branch unit and an image branch unit; the step of extracting meteorological feature features of the weather forecast data in a first preset time period, extracting image features of the remote sensing image data, and extracting time-series features of the power data through the feature extraction module includes: The time-series branch unit extracts meteorological features of the weather forecast data in the first preset time period and captures the time-series dependencies of the power data to extract time-series features; wherein, the meteorological features include: wind speed, wind direction, solar irradiance, air pressure, temperature and humidity; The image branching unit analyzes the movement, density, and morphology of clouds in the satellite remote sensing data to predict the irradiance change within a second preset time period, thereby obtaining a first image feature; the first image feature characterizes the fluctuation of photovoltaic power generation within the second preset time period. The image branching unit analyzes the near-surface wind speed changes and low-level jet streams in the ground-based remote sensing data to obtain a second image feature; the second image feature characterizes the power output changes of wind power.

[0008] In one possible implementation, the step of fusing the meteorological element features, image features, and temporal features through the feature fusion module to obtain target fused feature data with temporal sequence includes: Linear mapping is performed on the meteorological element features, image features, and time series features respectively to generate corresponding query vectors, key vectors, and value vectors; Using a cross-attention mechanism, attention weights are obtained by calculating the attention of the key vector of another modality using the query vector of any modality among the meteorological feature features, image features, and time series features. The attention weights are applied to the value vector of the other modality feature to determine the dependency between any modality feature and the other modality feature, thereby obtaining the first fused feature data; By using a cross-attention mechanism, the dependency relationship between the first fused feature data and the remaining modal features is determined, and the target fused feature data is obtained.

[0009] In one possible implementation, the dependency between any modal feature and another modal feature is determined in the following manner:

[0010] in, This represents the dependency between image features and temporal features. The query vector represents the image features. The key vector represents the temporal features. This is a value vector of time-series features. The dimension of the key vector.

[0011] In one possible implementation, the step of processing the time-series target fusion feature data through the probability prediction module to obtain the probability distribution of new energy power generation includes: Calculate the approximate distribution of each parameter in the target fused feature data; Randomly select a set of parameter values ​​from the approximate distribution of each parameter; Forward calculation is performed on any set of parameter values ​​to obtain a single-point predicted value; Repeatedly extract any set of parameter values ​​to obtain a set of single-point predicted values; The probability distribution is obtained by fitting the set of single-point predicted values.

[0012] In one possible implementation, the step of analyzing the probability distribution and outputting prediction results through the result output module includes: Based on the probability distribution, the mean of the probability distribution is calculated, and the mean is used as the point prediction value; Determine the confidence interval of the probability distribution, and calculate the probability index based on the probability distribution; The probability index is calculated using the following method.

[0013] in, This is the probability threshold; For probability indicators; Φ( () is the cumulative distribution function of the standard normal distribution; To predict the mean; For the predicted standard deviation; It is a probability index.

[0014] In one possible implementation, it further includes: Based on the forecast results, determine the day-ahead trading strategy and intraday / real-time scheduling.

[0015] According to a second aspect of this application, a new energy output probability prediction device is provided, the device comprising: The data acquisition module is used to acquire weather forecast data, remote sensing image data, and power data; The feature extraction module is used to extract meteorological features of the weather forecast data in a first preset time period, extract image features of the remote sensing image data, and extract time-series features of the power data. The feature fusion module is used to fuse the meteorological element features, image features and time-series features to obtain target fused feature data with time sequence. The probability distribution module is used to process the time-series target fusion feature data through the probability prediction module to obtain the probability distribution of new energy power generation. The probability analysis module is used to analyze the probability distribution through the result output module and output the prediction result.

[0016] According to a third aspect of this application, an electronic device is provided, comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in this application.

[0017] According to a fourth aspect of this application, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the methods described in this application.

[0018] According to a fifth aspect of this application, a computer program product is provided, comprising a computer program or instructions that, when executed by a processor, implement the method described in this application.

[0019] Using the technical solution of this application, the complete probability distribution of future output potential can be directly output, providing market participants with a more comprehensive basis for decision-making.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0021] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0022] Figure 1 A schematic diagram illustrating the implementation process of the new energy output probability prediction method in the embodiments of this application is shown; Figure 2 A schematic diagram of the probability distribution in an embodiment of this application is shown; Figure 3 This paper shows a block diagram illustrating the implementation of the new energy output probability prediction device in an embodiment of this application. Figure 4 A schematic diagram of the composition structure of the electronic device in an embodiment of this application is shown. Detailed Implementation

[0023] To make the objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0025] Unless otherwise defined, 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 application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0026] The following description, in conjunction with the accompanying drawings, introduces a method, apparatus, and electronic device for predicting the probability of new energy output provided in this application.

[0027] like Figure 1 As shown, this application provides a method for predicting the probability of new energy output, characterized in that it is applied to a probability prediction model, which includes a feature extraction module, a feature fusion module, a probability prediction module, and a result output module. It should be noted that the probabilistic prediction model in this application consists of a feature extraction module, a feature fusion module, a probabilistic prediction module, and a result output module. The probabilistic prediction model is trained using training data to obtain a trained probabilistic prediction model. The probabilistic prediction model can predict the probability of processing new energy sources. The method includes: S101 inputs the acquired weather forecast data, remote sensing image data, and power data into the probabilistic prediction model.

[0028] In some embodiments, the weather forecast data is data within a first preset time period; the weather forecast data is high-resolution numerical weather forecast data. The remote sensing data includes satellite remote sensing data and ground-based remote sensing data; The power data includes historical power output data and historical market data.

[0029] The weather forecast data in this application is high-resolution numerical weather prediction data. High-resolution numerical weather prediction data provides forecasts of meteorological elements with high spatiotemporal resolution (e.g., a temporal resolution of 15 minutes and a spatial resolution of 5 km x 5 km) for the next 72 hours. These meteorological elements include wind speed, wind direction, solar irradiance, air pressure, temperature, humidity, etc.

[0030] Satellite remote sensing data can be satellite cloud images. It's important to note that satellite cloud images are crucial for ultra-short-term forecasting of photovoltaic (PV) power generation. This application utilizes remote sensing image data from meteorological satellites, and can analyze cloud movement, density, and morphology using convolutional neural networks (CNNs). Furthermore, dense optical flow algorithms or generative models can be used to predict cloud images, effectively capturing short-term changes in irradiance, making it suitable for predicting sharp fluctuations in PV power generation.

[0031] Ground-based remote sensing data, such as ground-based lidar and weather radar, can provide extremely detailed meteorological data for specific areas. For example, lidar wind measurement technology can detect near-surface wind speed variations and low-level jet streams. Near-surface wind speed variations and low-level jet streams are crucial for identifying extreme weather conditions that may lead to wind power output "climbing events" (sudden increases or decreases in power). This application can use these high-resolution local data as important supplementary inputs.

[0032] Electricity data can be historical output data and historical market data. Examples include historical output data, grid load data, and historical market clearing prices. Using this electricity data as input features and processing date factors through methods such as one-hot encoding can capture specific date effects, such as holidays.

[0033] S102, the feature extraction module extracts meteorological features of the weather forecast data in the first preset time period, extracts image features of the remote sensing image data, and extracts time-series features of the power data.

[0034] The feature extraction module in this application can employ multiple branch units, with each branch unit performing feature extraction for a specific modality of data. For example, one branch unit can be used to extract meteorological features from weather forecast data, another branch unit can be used to extract image features from remote sensing image data, and other branch units can be used to extract time-series features from power data, etc.

[0035] S103, the meteorological element features, image features and time series features are fused by the feature fusion module to obtain target fusion feature data with time series.

[0036] Before fusing different modalities, feature standardization can be performed on the feature data of each modality. For example, normalization or standardization can be performed on the three types of features separately to eliminate differences in numerical ranges. For example, the magnitude difference between temperature (°C) and wind speed (m / s) in meteorological elements, and the difference between pixel values ​​(0-255) in image features, affects the fusion effect. Temporal alignment can also be performed on different modalities. This is because the goal is to obtain fused features with temporal sequence, requiring the temporal granularity of the three types of modal features to be unified, such as adjusting them to one time step every 15 minutes, and ensuring that the timestamps match perfectly. For example, the meteorological data at a certain time step, the satellite cloud image features at the corresponding time, and the power temporal features for that period should correspond one-to-one.

[0037] After feature extraction is completed in each branch, the feature fusion module is used to deeply fuse the feature vectors of different modalities. Since there are multiple modalities in this application, the features of two modalities can be fused first, and then the fused features can be fused with a third modality, thus obtaining the target fused feature data after fusing all modalities.

[0038] S104, The probability prediction module processes the time-series target fusion feature data to obtain the probability distribution of new energy power generation.

[0039] This application utilizes the parameter distribution characteristics of Bayesian Neural Networks (BNNs) to perform deep modeling of temporal fusion features, capturing the nonlinear correlation and uncertainty between features and power. Then, based on the distribution characteristics of the BNN parameters, multiple sets of power prediction values ​​are generated through repeated sampling and prediction cycles. The probability distribution includes the power distribution at each time step and the overall temporal distribution characteristics.

[0040] S105, the probability distribution is analyzed by the result output module, and the prediction result is output.

[0041] Finally, the probability distribution is analyzed through the results output module to obtain the final prediction result.

[0042] The new energy output probability prediction method provided in this application pre-constructs a probability prediction model capable of predicting new energy output. Upon receiving weather forecast data, remote sensing image data, and power data, the probability prediction model extracts meteorological features from the weather forecast data for a first preset time period, image features from the remote sensing image data, and time-series features from the power data using a feature extraction module. Then, a feature fusion module fuses the meteorological features, image features, and time-series features to obtain time-series target fused feature data. Next, the probability prediction module processes this time-series target fused feature data to obtain the probability distribution of new energy power generation. Finally, a result output module analyzes the probability distribution and outputs the prediction results. This application can directly output a complete probability distribution of future output potential, providing market participants with a more comprehensive decision-making basis.

[0043] In some embodiments, the feature extraction module includes a temporal branch unit and an image branch unit; the step of extracting meteorological feature features of the weather forecast data in a first preset time period, extracting image features of the remote sensing image data, and extracting temporal features of the power data through the feature extraction module includes: The time-series branch unit extracts meteorological features of the weather forecast data in the first preset time period and captures the time-series dependencies of the power data to extract time-series features; wherein, the meteorological features include: wind speed, wind direction, solar irradiance, air pressure, temperature and humidity; The image branching unit analyzes the movement, density, and morphology of clouds in the satellite remote sensing data to predict the irradiance change within a second preset time period, thereby obtaining a first image feature; the first image feature characterizes the fluctuation of photovoltaic power generation within the second preset time period. The image branching unit analyzes the near-surface wind speed changes and low-level jet streams in the ground-based remote sensing data to obtain a second image feature; the second image feature characterizes the power output changes of wind power.

[0044] In this application, the temporal branch unit focuses on processing structured data with a time dimension (weather forecast data, power data), capturing dynamic dependencies through temporal modeling. Specifically, a "first preset time period" (e.g., the next 72 hours) is defined, and the aforementioned six types of meteorological indicators for this period are extracted from the weather forecast data to form raw meteorological time-series data. Through standardization, such as eliminating dimensional differences and correcting outliers (e.g., removing values ​​exceeding physically reasonable ranges), the data is directly used as meteorological feature data, while retaining time format and dimensions. The temporal branch unit is also used to extract temporal features from power data, recording inputs such as historical photovoltaic power generation, wind power output, and load data in a time series. Temporal modeling algorithms, such as LSTM, GRU, or temporal convolutional networks, capture dependencies in the time dimension; these dependencies can be intraday periodicity, trends, or correlations between adjacent times. For example, the gating mechanism of LSTM is used to filter key historical information (e.g., power fluctuation patterns in the same time period of the previous three days), outputting a compressed high-dimensional feature vector as the temporal feature. The format of time series features can be (time steps, feature dimension).

[0045] Image branch units can analyze cloud movement trajectories in satellite remote sensing data using optical flow and target tracking methods, such as movement direction, speed, density, and morphology. Based on cloud analysis results, they predict the trend of solar irradiance changes within a second preset time period (e.g., the next 24 hours). This prediction result and cloud characteristics (e.g., movement speed, density level) are encoded into a high-dimensional vector as the "first image feature," which characterizes the fluctuation pattern of photovoltaic power generation within the second preset time period. Image branch units analyze the spatiotemporal variations of near-surface wind speed and low-level jet characteristics in ground-based remote sensing data using wind field inversion algorithms. Wind speed variation patterns (e.g., peak intraday wind speed periods) and low-level jet parameters (e.g., intensity, frequency) are encoded into high-dimensional vectors as the "second image feature"—its core function is to characterize the power output changes of wind power (because wind speed is a key influencing factor for wind power output, and low-level jets directly cause sudden changes in wind speed, thus triggering fluctuations in wind power).

[0046] In this application, the time-series branch unit focuses on structured time-series data, directly extracting meteorological elements and mining the time dependence of power data to provide basic dynamic features for subsequent fusion. The image branch unit targets two types of remote sensing images, using image analysis to correlate the core driving factors of new energy power generation and generate image features with clear physical meaning.

[0047] In some embodiments, the step of fusing the meteorological element features, image features, and temporal features through the feature fusion module to obtain target fused feature data with temporal sequence includes: Linear mapping is performed on the meteorological element features, image features, and time series features respectively to generate corresponding query vectors, key vectors, and value vectors; Using a cross-attention mechanism, attention weights are obtained by calculating the attention of the key vector of another modality using the query vector of any modality among the meteorological feature features, image features, and time series features. The attention weights are applied to the value vector of the other modality feature to determine the dependency between any modality feature and the other modality feature, thereby obtaining the first fused feature data; By using a cross-attention mechanism, the dependency relationship between the first fused feature data and the remaining modal features is determined, and the target fused feature data is obtained.

[0048] Meteorological features, image features, and time-series features are considered as three modalities. After feature extraction is completed in each branch unit, this application utilizes a feature fusion module to deeply fuse the feature vectors from different modalities. This fusion process is achieved through a cross-attention mechanism. Given features from one modality as a query vector and features from another modality as keys and values, attention calculation is used to achieve dynamic alignment and fusion of information, thereby learning the potential correlations and mutual influences between different data modalities.

[0049] For example, using image modal features Temporal modal features For example, its fusion process can be represented as: First, a linear mapping is performed on the features of each modality to generate a query vector. Key vector Sum value vector Taking image modalities as an example, we obtain the following through linear mapping:

[0050] in, , , These are the learnable weight matrices for the image modalities.

[0051] Temporal modality features are processed in a similar manner. To achieve cross-modal fusion, this application uses a cross-attention mechanism to allow features from one modality to serve as a query vector for querying the key vector of another modality (such as temporal). This yields attention weights, which are then applied to the value vector of the other modality. This process represents the updating of image features based on information from the temporal modality, thereby explicitly capturing the dependency between the two types of data. This process can be represented as:

[0052] in, The key vector dimension. This operation assigns weights to image features on the time-series data, enabling the model to more accurately understand time-series information (such as changes in irradiance) based on image information (such as cloud movement). This allows image modalities to supplement and guide time-series modalities, thereby generating a unified, high-dimensional target fusion feature data that can comprehensively characterize future power generation potential.

[0053] In some embodiments, the step of processing the time-series target fusion feature data through the probability prediction module to obtain the probability distribution of new energy power generation includes: Calculate the approximate distribution of each parameter in the target fused feature data; Randomly select a set of parameter values ​​from the approximate distribution of each parameter; Forward calculation is performed on any set of parameter values ​​to obtain a single-point predicted value; Repeatedly extract any set of parameter values ​​to obtain a set of single-point predicted values; The probability distribution is obtained by fitting the set of single-point predicted values.

[0054] In this application, the probability prediction module employs a Bayesian Neural Network (BNN). The BNN learns the probability distribution of its weights. The BNN can pass the uncertainty of the input data (e.g., the inherent error of NWP prediction) through the network structure layer by layer, ultimately reflecting it in the output result. In this application, the input of the Bayesian Neural Network is the target fused feature data output by the data fusion module.

[0055] In this application, variational inference is used to approximate the posterior distribution, specifically by introducing an easily tractable approximate distribution. (such as a Gaussian distribution), and optimize its parameters so that it approximates the true posterior distribution. Minimize the KL divergence between them.

[0056]

[0057] Each weight and bias in a BNN is determined by a distribution (e.g., a mean). and variance The distribution is represented by ). During prediction, the model randomly selects multiple sets of weights from these distributions, performs multiple forward propagations, and obtains a series of prediction results. The set of these prediction results together forms a complete probability distribution.

[0058] In some embodiments, the step of analyzing the probability distribution and outputting prediction results through the result output module includes: Based on the probability distribution, the mean of the probability distribution is calculated, and the mean is used as the point prediction value; Determine the confidence interval of the probability distribution, and calculate the probability index based on the probability distribution; The probability index is calculated using the following method.

[0059] in, This is the probability threshold; For probability indicators; Φ( () is the cumulative distribution function of the standard normal distribution; To predict the mean; For the predicted standard deviation; It is a probability index.

[0060] In this application, such as Figure 2 As shown, the output of a BNN is no longer a single deterministic value. It is not a complete probability distribution. For example, a value derived from the mean and variance Determined Gaussian distribution .include: Point prediction (P50): Take the mean of the probability distribution. , representing the most likely output value.

[0061] Confidence intervals (e.g., P5-P95): provide the upper and lower bounds of the prediction results, that is, there is a 90% probability that the actual output will fall within this interval. The upper and lower bounds can be derived from the cumulative distribution function (CDF) of the probability distribution.

[0062] Probability index: It can directly calculate the probability that the output will exceed or fall below a certain value at a future point in time. This application is fundamentally superior to quantile regression because it only requires training one model to obtain the complete probability distribution, and there is an inherent theoretical consistency between each quantile point, avoiding the problem of unsmooth or inconsistent curves caused by training independent models.

[0063] In some embodiments, the new energy output probability prediction device provided in this application further includes: Based on the forecast results, determine the day-ahead trading strategy and intraday / real-time scheduling.

[0064] Based on the forecast results, this application allows for the formulation of day-ahead trading strategies. Specifically, new energy power generation companies can use the P50 (50% confidence level) value as the core electricity volume they declare in the day-ahead market. Simultaneously, based on the confidence interval output by the BNN model, the risk of penalties due to output deviations is assessed. The larger the variance of the probability distribution output by the BNN, the higher the forecast uncertainty and the greater the risk. In this situation, companies can strategically purchase or sell ancillary services (such as reserve capacity) to hedge against risk.

[0065] Based on the forecast results, intraday / real-time scheduling can also be implemented. Specifically, for intraday and real-time transactions, the model can provide ultra-short-term (e.g., 15-minute) probabilistic predictions. When the BNN model predicts a potential "climbing event" based on feature vectors that fuse real-time satellite cloud imagery and ground-based radar data, it will output a probability distribution that drops sharply and increases in uncertainty. Grid dispatchers can use this information to adjust reserve capacity in advance, ensuring real-time grid balance and avoiding system security issues caused by a sudden drop in renewable energy output.

[0066] This application improves prediction accuracy: By effectively fusing high-resolution numerical weather prediction, satellite remote sensing, and ground-based lidar data from multiple sources, this application can better capture various complex factors affecting new energy output, especially under extreme weather conditions, thereby significantly reducing prediction errors (e.g., Normalized Mean Absolute Error (NMAE) and Root Mean Square Error (RMSE)). It also comprehensively quantifies uncertainty: The use of a Bayesian deep learning model enables this method to directly output a complete probability distribution, rather than just a single point prediction or discrete quantile. This makes the prediction results more reliable and provides richer, more valuable information for downstream applications. Furthermore, it demonstrates powerful data fusion capabilities: The proposed multi-branch, multi-modal fusion architecture effectively solves the problem of deep fusion of different types of heterogeneous data, fully utilizing the complementary information of image data, time-series data, and discrete data, enhancing the model's generalization ability and robustness.

[0067] In addition, this application can reduce market transaction risks: high-precision probabilistic forecasts enable renewable energy power plants to submit day-ahead and intraday trading plans more accurately. By utilizing the confidence intervals of the forecast results, power plants can adjust their trading strategies with a basis, for example, purchasing standby services when forecast uncertainty is high, thereby effectively avoiding economic penalties caused by excessive deviations between actual output and the submitted plan. It also enhances the profitability of power generation companies: accurate forecasts enable power generation companies to bid for better prices and quantities in the spot market, avoiding additional costs on unnecessary standby capacity. This allows companies to shift from the traditional "consumption-oriented" model to a proactive market-participation, profit-optimizing model, improving their market competitiveness. Furthermore, it ensures the safe and stable operation of the power grid: the accurate forecasts provided by this application, especially the probabilistic early warning of short-term output surges ("ramp events"), enable power grid dispatching agencies to anticipate and respond to potential system imbalance risks earlier, reducing the demand for standby capacity, thereby improving the overall operating efficiency and reliability of the power grid.

[0068] As an example, wind power output probability forecasting for the day-ahead market.

[0069] This embodiment aims to illustrate how this application can be applied to trading decisions in the day-ahead market for wind farms. According to the rules of the electricity spot market, wind farms are typically required to submit their output plans for the following day, every 15 minutes or hour, to the market operator before 15:00 on the day before the operation date.

[0070] (1) Data input and feature fusion: At 15:00 the day before the operation date, the latest information is automatically obtained from high-resolution NWP data, historical power output data of wind farms, etc. These multi-source data are input into the multimodal data fusion module. This module performs feature extraction and deep fusion on the data, and finally generates a high-dimensional, unified feature vector, which can comprehensively represent the meteorological and operating conditions in the next 72 hours.

[0071] (2) Bayesian deep learning model processing and prediction: ① High-dimensional feature vectors are used as inputs to a Bayesian neural network (BNN).

[0072] ②BNN utilizes its ability to model the probability distribution of model weight parameters. Through multiple forward propagations and sampling, it generates a complete output probability distribution for each 15-minute time point within the next 24 hours, rather than a single deterministic value. This distribution is presented in the form of mean (P50) and variance, with the magnitude of the variance directly reflecting the uncertainty of the prediction.

[0073] (3) Market application: ① Power traders use the mean (P50) of the probability distribution output by BNN as the core electricity volume to be reported in their day-ahead market.

[0074] ② Power traders can refer to the predicted confidence intervals (e.g., P5 and P95) to assess risk. If the predicted confidence interval is wide at a certain moment, it indicates that the BNN model considers the uncertainty at that moment to be high. Traders may choose to declare electricity slightly lower than P50, or strategically purchase ancillary services (such as reserve capacity) to reduce the risk of being penalized for actual output being lower than the declared amount.

[0075] As another specific example, distributed photovoltaic power output probability prediction is aimed at the intraday / real-time market.

[0076] This embodiment aims to illustrate how this application can be applied to the intraday and real-time markets of distributed photovoltaic (PV) power generation. Due to the rapid movement of clouds, PV output can fluctuate drastically in a short period of time, posing a threat to grid stability.

[0077] (1) Data Input and Feature Fusion: The system continuously receives and processes high-frequency real-time data streams, including satellite cloud images updated every 5-15 minutes, local sensor data, and ultra-short-term NWP data. The image branch (CNN) of the multimodal data fusion module analyzes the satellite cloud images in real time and extracts spatial features such as cloud motion and density. The fusion layer uses a cross-attention mechanism to deeply fuse these image features with time-series data, generating a high-dimensional feature vector containing real-time dynamic information.

[0078] (2) Bayesian deep learning model processing and prediction: ① Input the high-dimensional feature vector that incorporates real-time dynamic information into the BNN model.

[0079] ②The BNN model processes the vector and uses its ability to quantify uncertainty to generate a probability distribution for each time point within the next 15-60 minutes.

[0080] ③ When the model detects a "climbing event" that may be caused by rapidly moving clouds based on the fused image features, its output probability distribution will exhibit distinct characteristics: the mean (P50) drops sharply, and the variance increases significantly. For example, the model may output information such as "there is an 85% probability that output will decrease by more than 30% in the next 10 minutes."

[0081] (3) Power grid application: Power grid dispatching agencies can use this ultra-short-term probabilistic early warning information to activate standby units or demand-side response resources in advance to cope with the impending sharp drop in photovoltaic output. This proactive early warning mechanism based on uncertainty quantification allows dispatchers more time to make decisions and deployments before the event occurs, thereby significantly improving the real-time balance and safe operation level of the power grid.

[0082] As shown in Figure 3, this application provides a new energy output probability prediction device, applied to a probability prediction model, wherein the probability prediction model includes a feature extraction module, a feature fusion module, a probability prediction module, and a result output module; the device includes: Data acquisition module 301 is used to acquire weather forecast data, remote sensing image data, and power data; Feature extraction module 302 is used to extract meteorological features of the weather forecast data in a first preset time period, extract image features of the remote sensing image data, and extract time-series features of the power data. The feature fusion module 303 is used to fuse the meteorological element features, image features and time series features through the feature fusion module to obtain target fused feature data with time series. The probability distribution module 304 is used to process the time-series target fusion feature data through the probability prediction module to obtain the probability distribution of new energy power generation. The probability analysis module 305 is used to analyze the probability distribution through the result output module and output the prediction result.

[0083] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0084] The electronic device includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enable the at least one processor to perform the new energy output probability prediction method described in this application. The computer instructions are used to cause the computer to execute the new energy output probability prediction method described in this application.

[0085] This application also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implements the new energy output probability prediction method of this application.

[0086] Figure 4 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0087] like Figure 4 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0088] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0089] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the renewable energy output probability prediction method. For example, in some embodiments, the renewable energy output probability prediction method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the renewable energy output probability prediction method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the renewable energy output probability prediction method by any other suitable means (e.g., by means of firmware).

[0090] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0091] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0092] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0093] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0094] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0095] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0096] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the probability of new energy power output, characterized in that, The method is applied to a probabilistic prediction model, which includes a feature extraction module, a feature fusion module, a probabilistic prediction module, and a result output module; the method includes: The acquired weather forecast data, remote sensing image data, and power data are input into the probabilistic prediction model; The feature extraction module extracts meteorological features from the weather forecast data in a first preset time period, image features from the remote sensing image data, and time-series features from the power data. The feature fusion module fuses the meteorological element features, image features, and temporal features to obtain target fused feature data with temporal sequence. The probability prediction module processes the time-series target fusion feature data to obtain the probability distribution of new energy power generation. The probability distribution is analyzed by the result output module, and the prediction result is output.

2. The method according to claim 1, characterized in that, The weather forecast data is data within a first preset time period; the weather forecast data is high-resolution numerical weather forecast data; The remote sensing data includes satellite remote sensing data and ground-based remote sensing data; The power data includes historical power output data and historical market data.

3. The method according to claim 2, characterized in that, The feature extraction module includes a time-series branch unit and an image branch unit; the step of extracting meteorological features of the weather forecast data in a first preset time period, extracting image features of the remote sensing image data, and extracting time-series features of the power data through the feature extraction module includes: The time-series branch unit extracts meteorological features of the weather forecast data in the first preset time period and captures the time-series dependencies of the power data to extract time-series features; wherein, the meteorological features include: wind speed, wind direction, solar irradiance, air pressure, temperature and humidity; The image branching unit analyzes the movement, density, and morphology of clouds in the satellite remote sensing data to predict the irradiance change within a second preset time period, thereby obtaining a first image feature; the first image feature characterizes the fluctuation of photovoltaic power generation within the second preset time period. The image branching unit analyzes the near-surface wind speed changes and low-level jet streams in the ground-based remote sensing data to obtain a second image feature; the second image feature characterizes the power output changes of wind power.

4. The method according to claim 1, characterized in that, The process of fusing meteorological element features, image features, and temporal features through the feature fusion module to obtain target fused feature data with temporal sequence includes: Linear mapping is performed on the meteorological element features, image features, and time series features respectively to generate corresponding query vectors, key vectors, and value vectors; Using a cross-attention mechanism, attention weights are obtained by calculating the attention of the key vector of another modality using the query vector of any modality among the meteorological feature features, image features, and time series features. The attention weights are applied to the value vector of the other modality feature to determine the dependency between any modality feature and the other modality feature, thereby obtaining the first fused feature data; By using a cross-attention mechanism, the dependency relationship between the first fused feature data and the remaining modal features is determined, and the target fused feature data is obtained.

5. The method according to claim 4, characterized in that, The dependency between any modal feature and another modal feature is determined in the following ways: in, This represents the dependency between image features and temporal features. The query vector represents the image features. The key vector represents the temporal features. This is a value vector of time-series features. The dimension of the key vector.

6. The method according to claim 1, characterized in that, The step of processing the time-series target fusion feature data through the probability prediction module to obtain the probability distribution of new energy power generation includes: Calculate the approximate distribution of each parameter in the target fused feature data; Randomly select a set of parameter values ​​from the approximate distribution of each parameter; Forward calculation is performed on any set of parameter values ​​to obtain a single-point predicted value; Repeatedly extract any set of parameter values ​​to obtain a set of single-point predicted values; The probability distribution is obtained by fitting the set of single-point predicted values.

7. The method according to claim 1, characterized in that, The step of analyzing the probability distribution and outputting prediction results through the result output module includes: Based on the probability distribution, the mean of the probability distribution is calculated, and the mean is used as the point prediction value; Determine the confidence interval of the probability distribution, and calculate the probability index based on the probability distribution; The probability index is calculated using the following method. in, This is the probability threshold; For probability indicators; Φ( () is the cumulative distribution function of the standard normal distribution; To predict the mean; For the predicted standard deviation; It is a probability index.

8. The method according to claim 1, characterized in that, Also includes: Based on the forecast results, determine the day-ahead trading strategy and intraday / real-time scheduling.

9. A new energy output probability prediction device, characterized in that, The device is applied to a probabilistic prediction model, which includes a feature extraction module, a feature fusion module, a probabilistic prediction module, and a result output module; the device includes: The data acquisition module is used to acquire weather forecast data, remote sensing image data, and power data; The feature extraction module is used to extract meteorological features of the weather forecast data in a first preset time period, extract image features of the remote sensing image data, and extract time-series features of the power data. The feature fusion module is used to fuse the meteorological element features, image features and time-series features to obtain target fused feature data with time sequence. The probability distribution module is used to process the time-series target fusion feature data through the probability prediction module to obtain the probability distribution of new energy power generation. The probability analysis module is used to analyze the probability distribution through the result output module and output the prediction result.

10. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

Images