AI-based new energy power generation prediction system and method

Abstract

The present application relates to the field of power generation prediction, and specifically to an AI-based new energy power generation prediction system and method. The method includes: dividing the photovoltaic power supply cycle, and determining the predicted radiation curve of the next cycle based on predicted meteorological data and the location of the photovoltaic power supply; correcting the predicted radiation curve to obtain a corrected predicted radiation curve; obtaining a temperature prediction change curve, and obtaining a predicted photovoltaic power supply feed-in power curve based on the temperature prediction change curve and the corrected predicted radiation curve; calculating the predicted feed-in power of the wind power supply at each moment to obtain a predicted wind power supply feed-in power curve; and adjusting the hydrogen energy storage strategy based on the predicted photovoltaic power supply feed-in power curve, the predicted wind power supply feed-in power curve, and the real-time feed-in power.

Description

Technical Field

[0001] The present application relates to the field of power generation prediction, and specifically to an AI-based new energy power generation prediction system and method. Background Art

[0002] As the world actively promotes sustainable energy development, renewable energy generation is increasingly becoming a significant contributor to the energy mix. However, renewable energy sources like solar and wind power are characterized by significant intermittency and volatility, creating numerous challenges for the stable operation and efficient management of power systems. Therefore, accurate power generation forecasting is crucial to ensuring the safe and stable operation of power grids.

[0003] Patent CN115441447B proposes a method for predicting renewable energy power generation. This approach combines weather forecasts with real-time cloud image analysis to periodically predict the radiation output of photovoltaic power sources and adjust the radiation output curve to accurately predict the feed-in power. Furthermore, wind power generation curves are predicted by forecasting wind conditions and temperature fluctuations. Ultimately, hydrogen storage strategies are optimized based on the photovoltaic and wind power forecasts and actual feed-in power, improving the accuracy and reliability of renewable energy power generation forecasts.

[0004] The patent predicts wind power curves using common machine learning algorithms, such as convolutional neural networks. CNNs are primarily used to recognize local patterns in two- or three-dimensional space. When processing multidimensional time series data such as wind speed and direction, they may not be able to effectively capture the complex interactions between different dimensions, and struggle to establish effective temporal dependencies between different time points, thus affecting prediction accuracy.

[0005] Therefore, an optimized new energy power generation power prediction scheme is desired. Summary of the Invention

[0006] This application is made in consideration of the above problems. One purpose of this application is to provide an AI-based new energy power generation power prediction system and method.

[0007] An embodiment of the present application provides an AI-based new energy power generation power prediction method, comprising: dividing a photovoltaic power supply cycle, and determining a predicted radiation curve for the next cycle based on predicted meteorological data and the location of the photovoltaic power supply; correcting the predicted radiation curve to obtain a corrected predicted radiation curve; obtaining a temperature prediction change curve, and obtaining a predicted photovoltaic power supply feed-in power curve based on the temperature prediction change curve and the corrected predicted radiation curve; calculating the predicted feed-in power of a wind power supply at each moment to obtain a predicted wind power supply feed-in power curve; and adjusting a hydrogen energy storage strategy based on the predicted photovoltaic power supply feed-in power curve, the predicted wind power supply feed-in power curve, and real-time feed-in power; wherein calculating the predicted feed-in power of a wind power supply at each moment to obtain a predicted wind power supply feed-in power curve comprises:

[0008] Obtaining predicted wind condition time series parameters at the location of the wind power source, and obtaining predicted temperature time series parameters at the location of the wind power source, wherein the predicted wind condition time series parameters include wind speed time series parameters and wind direction angle time series parameters;

[0009] performing direction field calculation and feature extraction on the wind speed time series parameter, the wind direction angle time series parameter, and the predicted temperature time series parameter respectively to obtain wind speed time series direction field features, wind direction angle time series direction field features, and predicted temperature time series direction field features;

[0010] Performing interactive fusion of wind condition multi-dimensional time series features on the wind speed time series direction field features and the wind direction angle time series direction field features to obtain an interactive representation of wind condition multi-dimensional time series features;

[0011] The principal component-guided feature compensation interaction is performed on the multi-dimensional time series feature interaction representation of the wind condition and the predicted temperature time series directional field feature to obtain the wind condition-temperature time series directional field compensation interaction feature, and based on the wind condition-temperature time series directional field compensation interaction feature, the wind power feed-in power value at the next moment is obtained.

[0012] For example, according to the AI-based new energy power generation prediction method of an embodiment of the present application, the wind speed time series parameters, the wind direction angle time series parameters, and the predicted temperature time series parameters are respectively subjected to direction field calculation and feature extraction to obtain wind speed time series direction field features, wind direction angle time series direction field features, and predicted temperature time series direction field features, including:

[0013] Calculating the direction fields of the wind speed time series parameter, the wind direction angle time series parameter, and the predicted temperature time series parameter respectively to obtain a wind speed time series direction field diagram representation, a wind direction angle time series direction field diagram representation, and a predicted temperature time series direction field diagram representation;

[0014] The wind speed time series directional field map representation, the wind direction angle time series directional field map representation and the predicted temperature time series directional field map representation are respectively passed through a directional field feature extractor based on an expanded convolutional neural network to obtain a wind speed time series directional field feature vector as the wind speed time series directional field feature, a wind direction angle time series directional field feature vector as the wind direction angle time series directional field feature and a predicted temperature time series directional field feature vector as the predicted temperature time series directional field feature.

[0015] For example, according to the AI-based new energy power generation prediction method of an embodiment of the present application, the wind speed time series direction field characteristics and the wind direction angle time series direction field characteristics are interactively fused to obtain a wind condition multi-dimensional time series feature interactive representation, including: passing the wind speed time series direction field feature vector and the wind direction angle time series direction field feature vector through a wind condition multi-dimensional time series feature interactive fusion device based on an X-linear attention bilinear pooling unit to obtain a wind condition multi-dimensional time series feature interactive representation vector as the wind condition multi-dimensional time series feature interactive representation.

[0016] For example, according to an embodiment of the present application, an AI-based renewable energy power generation prediction method is provided, wherein the principal component-guided feature compensation interaction is performed on the wind condition multidimensional time series feature interaction representation and the predicted temperature time series direction field feature to obtain a wind condition-temperature time series direction field compensation interaction feature, and based on the wind condition-temperature time series direction field compensation interaction feature, the wind power feed-in power value at the next moment is obtained, including:

[0017] Performing principal component-guided feature compensation interaction on the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series directional field feature vector to obtain a wind condition-temperature time series directional field compensation interaction feature vector as the wind condition-temperature time series directional field compensation interaction feature;

[0018] Based on the wind condition-temperature time series direction field compensation interaction characteristic vector, the wind power feed-in power value at the next moment is obtained.

[0019] For example, according to an AI-based new energy power generation prediction method according to an embodiment of the present application, the principal component-guided feature compensation interaction is performed on the wind condition multi-dimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain a wind condition-temperature time series direction field compensation interaction feature vector, including:

[0020] Performing principal component analysis on the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain a set of wind condition multidimensional time series feature principal component encoding vectors and a set of predicted temperature time series direction field feature principal component encoding vectors;

[0021] Constructing the set of principal component coding vectors of the multidimensional time series characteristics of wind conditions and the set of principal component coding vectors of the predicted temperature time series direction field characteristics into a principal component aggregation coding feature map of the multidimensional time series characteristics of wind conditions and a principal component aggregation coding feature map of the predicted temperature time series characteristics;

[0022] Calculating a wind condition-predicted temperature time series difference embedding compensation coding weight vector between the principal component aggregation coding feature map of the wind condition multi-dimensional time series feature and the principal component aggregation coding feature map of the predicted temperature time series feature;

[0023] Based on the wind condition-predicted temperature time series difference embedded compensation coding weight vector, the set of principal component coding vectors of the wind condition multidimensional time series characteristics and the set of principal component coding vectors of the predicted temperature time series directional field characteristics are compensated and aggregated to obtain the wind condition-temperature time series directional field compensation interaction feature vector.

[0024] For example, according to the AI-based new energy power generation prediction method of an embodiment of the present application, the calculation of the wind condition-predicted temperature time series difference embedding compensation coding weight vector between the principal component aggregation coding feature map of the wind condition multidimensional time series feature and the principal component aggregation coding feature map of the predicted temperature time series feature includes:

[0025] Performing differential embedding processing on the principal component aggregation coding feature map of the wind condition multidimensional time series feature and the principal component aggregation coding feature map of the predicted temperature time series feature to obtain a wind condition multidimensional time series branch weight vector and a predicted temperature time series branch weight vector;

[0026] The wind condition-predicted temperature time series difference embedding compensation coding weight vector is calculated between the wind condition multi-dimensional time series branch weight vector and the predicted temperature time series branch weight vector.

[0027] For example, according to an AI-based new energy power generation prediction method according to an embodiment of the present application, a compensation coding weight vector is embedded based on the wind condition-predicted temperature time series difference, and a set of principal component coding vectors of the wind condition multidimensional time series characteristics and a set of principal component coding vectors of the predicted temperature time series direction field characteristics are compensated and aggregated to obtain the wind condition-temperature time series direction field compensation interaction feature vector, including:

[0028] Calculating the positional mean vector of the set of principal component coding vectors of the wind condition multidimensional time series characteristics and the set of principal component coding vectors of the predicted temperature time series directional field characteristics to obtain the wind condition multidimensional time series principal component representation coding vector and the predicted temperature time series principal component representation coding vector;

[0029] Based on the wind condition-predicted temperature time series difference embedding compensation coding weight vector, the wind condition multidimensional time series principal component characterization coding vector and the predicted temperature time series principal component characterization coding vector are aggregated and interacted to obtain the wind condition-temperature time series directional field compensation interaction feature vector.

[0030] For example, according to the AI-based new energy power generation prediction method of an embodiment of the present application, the wind power feed-in power value at the next moment is obtained based on the wind condition-temperature time series direction field compensation interaction characteristic vector, including: passing the wind condition-temperature time series direction field compensation interaction characteristic vector through a decoder-based wind power predictor to obtain a wind power feed-in power decoding value as the wind power feed-in power value at the next moment.

[0031] The embodiments of the present application also provide an AI-based new energy power generation prediction system, including:

[0032] The predicted radiation curve acquisition module is used to divide the photovoltaic power supply cycle and determine the predicted radiation curve of the next cycle based on the predicted meteorological data and the location of the photovoltaic power supply;

[0033] A correction module, configured to correct the predicted radiation amount curve to obtain a corrected predicted radiation amount curve;

[0034] A module for obtaining a predicted photovoltaic power supply feed-in power curve is configured to obtain a temperature prediction change curve and obtain a predicted photovoltaic power supply feed-in power curve based on the temperature prediction change curve and the corrected predicted radiation amount curve;

[0035] The predicted feed-in power calculation module is used to calculate the predicted feed-in power of the wind power source at each moment to obtain a predicted wind power source feed-in power curve;

[0036] A strategy adjustment module is configured to adjust the hydrogen energy storage strategy based on the predicted photovoltaic power supply feed-in power curve, the predicted wind power supply feed-in power curve, and the real-time feed-in power; wherein the strategy adjustment module includes:

[0037] a timing parameter acquisition unit, configured to acquire a predicted wind condition timing parameter at the location where the wind power source is located, and acquire a predicted temperature timing parameter at the location where the wind power source is located, wherein the predicted wind condition timing parameter includes a wind speed timing parameter and a wind direction angle timing parameter;

[0038] a feature extraction unit, configured to perform direction field calculation and feature extraction on the wind speed time series parameter, the wind direction angle time series parameter, and the predicted temperature time series parameter, respectively, to obtain wind speed time series direction field features, wind direction angle time series direction field features, and predicted temperature time series direction field features;

[0039] a multi-dimensional time series feature interactive fusion unit, configured to perform wind condition multi-dimensional time series feature interactive fusion on the wind speed time series directional field feature and the wind direction angle time series directional field feature to obtain a wind condition multi-dimensional time series feature interactive representation;

[0040] The wind power feed-in power value acquisition unit at the next moment is used to perform principal component-guided feature compensation interaction on the multi-dimensional time series feature interaction representation of the wind condition and the predicted temperature time series directional field feature to obtain the wind condition-temperature time series directional field compensation interaction feature, and obtain the wind power feed-in power value at the next moment based on the wind condition-temperature time series directional field compensation interaction feature.

[0041] For example, according to an AI-based new energy power generation prediction system according to an embodiment of the present application, the feature extraction unit is used to:

[0042] Calculating the direction fields of the wind speed time series parameter, the wind direction angle time series parameter, and the predicted temperature time series parameter respectively to obtain a wind speed time series direction field diagram representation, a wind direction angle time series direction field diagram representation, and a predicted temperature time series direction field diagram representation;

[0043] The wind speed time series directional field map representation, the wind direction angle time series directional field map representation and the predicted temperature time series directional field map representation are respectively passed through a directional field feature extractor based on an expanded convolutional neural network to obtain a wind speed time series directional field feature vector as the wind speed time series directional field feature, a wind direction angle time series directional field feature vector as the wind direction angle time series directional field feature and a predicted temperature time series directional field feature vector as the predicted temperature time series directional field feature.

[0044] According to the AI-based new energy power generation prediction method of the embodiment of the present application, it uses AI-based data analysis and prediction technology to perform directional field calculation and time series feature extraction on the predicted wind condition time series parameters and predicted temperature time series parameters, and then calculates the interactive representation information between the wind speed time series directional field characteristics and the wind direction angle time series directional field characteristics, so as to intelligently predict the wind power feed-in power value at the next moment based on the principal component-guided interactive representation between the wind condition multi-dimensional time series feature interactive representation and the predicted temperature time series directional field feature. Compared with traditional convolutional neural networks, the present application can better reflect the dynamic changes of wind conditions through directional field calculation and time series feature extraction. At the same time, it can establish a closer and more reasonable association between different time points, which helps to capture long-term historical dependencies and short-term dynamic changes, thereby improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings of the embodiments of the present application. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0046] Figure 1 A flowchart of a method for predicting renewable energy power generation based on AI in an embodiment of the present application is shown;

[0047] Figure 2 A flowchart of sub-step S140 of the AI-based new energy power generation prediction method in an embodiment of the present application is shown;

[0048] Figure 3 A flowchart of sub-step S142 of the AI-based new energy power generation prediction method in an embodiment of the present application is shown;

[0049] Figure 4 A flowchart of sub-step S144 of the AI-based new energy power generation prediction method in an embodiment of the present application is shown;

[0050] Figure 5 A flowchart showing sub-step S1441 of the AI-based new energy power generation prediction method in an embodiment of the present application is shown; and

[0051] Figure 6 A structural diagram of an AI-based new energy power generation prediction system in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0052] The terms used in this specification are those commonly used in the art currently in consideration of the functions of the present application, but these terms may vary according to the intentions of those skilled in the art, precedents, or new technologies in the art. In addition, specific terms may be selected, and in such cases, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but rather as the meaning of the terms and the overall description of the present application.

[0053] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0054] Flowcharts are used throughout this application to illustrate the operations performed by the systems of the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of this application.

[0056] refer to Figure 1 As shown, the present application proposes an AI-based new energy power generation power prediction method, including: S110, dividing the photovoltaic power supply cycle, and determining the predicted radiation curve of the next cycle based on the predicted meteorological data and the location of the photovoltaic power supply; S120, correcting the predicted radiation curve to obtain a corrected predicted radiation curve; S130, obtaining a temperature prediction change curve, and obtaining a predicted photovoltaic power supply feed-in power curve based on the temperature prediction change curve and the corrected predicted radiation curve; S140, calculating the predicted feed-in power of the wind power supply at each moment to obtain a predicted wind power supply feed-in power curve; S150, adjusting the hydrogen energy storage strategy based on the predicted photovoltaic power supply feed-in power curve, the predicted wind power supply feed-in power curve and the real-time feed-in power. In particular, in a specific example of the present application, the corrected predicted radiation curve can be corrected by a first cloud layer image obtained at the beginning of the next cycle and a second cloud layer image obtained after a set time interval.

[0057] It is worth mentioning that in the field of new energy, especially photovoltaic and wind power generation, accurate power forecasting is crucial to the stable operation of the power system. It helps grid operators optimize scheduling, improve the utilization rate of renewable energy and reduce the demand for backup capacity. The AI-based new energy power generation power forecasting method proposed in this application aims to provide a more accurate and intelligent forecasting method through a series of steps, so as to better adapt to the characteristics of new energy with greater volatility. The premise of this forecasting method is that the output power of photovoltaic power generation and wind power generation is closely related to natural conditions. The intensity and temperature of solar radiation affect the efficiency of photovoltaic panels, while the wind speed directly determines the power generation of wind turbines. Therefore, before formulating any forecasting model, the influence of these factors must be considered.

[0058] Specifically, step S110 is to divide the day into several time periods, each corresponding to a specific lighting condition. For example, one can choose to use an hourly period as a cycle, and determine the amount of solar radiation likely to be received in each hour based on historical data, current seasonal changes, and predicted meteorological information. The "location" mentioned here refers not only to geographic coordinates, but also includes factors such as the angle and orientation of the photovoltaic panels, as these factors also affect the final amount of radiation received.

[0059] After obtaining the preliminary predicted radiation curve, in step S120, the predicted radiation curve is corrected. The reason for the correction is that the original prediction may be affected by various uncertain factors, such as cloud cover or other transient meteorological phenomena. In order to make the prediction closer to the actual situation, a machine learning algorithm can be introduced to automatically adjust the predicted value by using similar situations in historical data as training samples. For example, if there is a brief period of dark cloud cover at noon on a certain day, but then the sky returns to clear, the system will learn this pattern and make appropriate corrections when encountering similar situations in the future. This method can significantly improve the accuracy of short-term predictions.

[0060] Once the revised predicted radiation profile is ready, in step S130, it is combined with the predicted temperature variation profile to calculate the expected photovoltaic power feed-in power profile. Temperature has a direct impact on photovoltaic cell efficiency: rising temperatures lead to decreased conversion efficiency. Therefore, before generating the final predicted power profile, temperature trends must be considered. In practice, linear regression or other statistical methods can be used to model the relationship between temperature and power output. Once the temperature prediction curve is in place, the expected power output at different temperatures can be calculated based on this model, resulting in a complete predicted photovoltaic power feed-in power profile.

[0061] At the same time, for wind power generation, it is also necessary to predict its feed-in power at each moment. Accordingly, in step S140, the predicted feed-in power of the wind power source at each moment is calculated. Unlike photovoltaic power generation, which depends on solar radiation and temperature, wind power generation mainly depends on wind speed. Therefore, the focus of the work at this stage is to collect historical records about local wind speeds and future forecast information. Modern meteorological services can usually provide relatively reliable short-term wind speed forecasts, which are very valuable resources for wind farms. By analyzing these data and combining them with the performance parameters of the wind turbine itself, a prediction model can be constructed to estimate wind power output at different times. It is worth noting that with the advancement of wind turbine technology and control strategies, it is now possible to further improve the prediction accuracy by performing deep learning on the real-time monitoring data of a single wind turbine.

[0062] Finally, based on all the above prediction results, namely the predicted feed-in power curves for photovoltaic and wind power sources, combined with the actual feed-in power monitored in real time, the hydrogen energy storage strategy is adjusted in step S150. Hydrogen energy storage, as an emerging energy storage method, can convert and store electrical energy into hydrogen energy during periods of excess power, releasing it to replenish the grid during periods of power shortage. Therefore, rationally planning the charging and discharging of hydrogen energy storage is crucial for balancing supply and demand. Specifically, a set of rules can be established or advanced algorithms such as reinforcement learning can be used to ensure that energy storage is utilized at the appropriate time. For example, during daytime periods with abundant sunshine and high wind power, excess electricity can be prioritized for hydrogen production; at night or during periods of calm wind, stored hydrogen energy can be released to maintain power supply stability. Furthermore, consideration can be given to combining this with other types of energy storage facilities, such as lithium batteries, to build a multi-level energy management system.

[0063] It should be understood that in calculating the predicted feed-in power of a wind power source at each moment to obtain a predicted wind power feed-in power curve, the technical concept of this application is to obtain the time series parameters of the predicted wind conditions (wind speed and wind direction angle) at the location of the wind power source, and obtain the predicted temperature time series parameters at the location of the wind power source, and use AI-based data analysis and prediction technology to perform directional field calculation and time series feature extraction on the predicted wind condition time series parameters and the predicted temperature time series parameters. Then, the interactive representation information between the wind speed time series directional field features and the wind direction angle time series directional field features is calculated. In this way, the wind power feed-in power value at the next moment is intelligently predicted based on the principal component-guided interactive representation between the multi-dimensional time series feature interactive representation of the wind condition and the predicted temperature time series directional field features. Compared with traditional CNN, this application can better reflect the dynamic changes of wind conditions through directional field calculation and time series feature extraction. At the same time, it can establish a closer and more reasonable association between different time points, which helps to capture long-term historical dependencies and short-term dynamic changes, thereby improving prediction accuracy.

[0064] In one example, if Figure 2As shown, in step S140, the predicted feed-in power of the wind power source at each moment is calculated to obtain the predicted wind power source feed-in power curve, including: S141, obtaining the predicted wind condition time series parameters of the location of the wind power source, and obtaining the predicted temperature time series parameters of the location of the wind power source, wherein the predicted wind condition time series parameters include wind speed time series parameters and wind direction angle time series parameters; S142, performing direction field calculation and feature extraction on the wind speed time series parameters, the wind direction angle time series parameters and the predicted temperature time series parameters respectively to obtain wind speed time series direction field features, wind direction angle time series parameters and wind direction angle time series parameters. Angle time series direction field characteristics and predicted temperature time series direction field characteristics; S143, interactively fuse the wind speed time series direction field characteristics and the wind direction angle time series direction field characteristics with the wind condition multi-dimensional time series characteristics to obtain a wind condition multi-dimensional time series characteristic interactive representation; S144, perform principal component-guided feature compensation interaction on the wind condition multi-dimensional time series characteristic interactive representation and the predicted temperature time series direction field characteristics to obtain a wind condition-temperature time series direction field compensation interactive characteristic, and based on the wind condition-temperature time series direction field compensation interactive characteristic, obtain the wind power feed-in power value at the next moment.

[0065] Specifically, in the technical solution of the present application, first, the predicted wind condition time series parameters of the location of the wind power source are obtained, wherein the predicted wind condition time series parameters include wind speed time series parameters and wind direction angle time series parameters.

[0066] Next, the predicted temperature time series parameters for the location of the wind turbine are obtained. In a specific example, the required temperature information type is first determined. For wind power generation, the most important is the ground temperature, which is the air temperature at a height of two meters above the ground. Next, an appropriate data source is selected to obtain this information, such as short-term weather forecasts from the National Meteorological Agency (CMA). Short-term weather forecasts can include hourly temperature forecasts. A request is then sent to the selected service provider via an API or web interface, specifying the geographic coordinates (latitude and longitude), the forecast period (e.g., the next three days), and the time resolution (e.g., hourly). Upon successful request, data containing hourly temperature forecasts for the next 72 hours is received. After receiving the raw data, necessary cleansing and conversions are performed to ensure suitability for subsequent analysis. This may involve unit conversion (e.g., converting Fahrenheit to Celsius), imputing missing values, and detecting outliers. Finally, the cleaned temperature data is organized into an ordered time series list, with each element representing a temperature measurement at a specific point in time. This series is then used as an input into the subsequent wind turbine power prediction model.

[0067] Considering that renewable energy power generation is dynamically influenced by multiple factors, and that these factors exhibit complex temporal variations and flow trends, in order to more deeply understand and capture the dynamic characteristics of these variables over time, this application calculates the directional fields of the wind speed time series parameters, the wind direction angle time series parameters, and the predicted temperature time series parameters to intuitively display the flow and variation trends of the parameters, reflecting parameter information at different moments. This results in wind speed time series directional field diagrams, wind direction angle time series directional field diagrams, and predicted temperature time series directional field diagrams. It is understood that directional field diagrams (or vector field diagrams) are typically used to display the direction and intensity of changes in a physical quantity over time and / or space. By calculating directional diagrams for these time series parameters, the variation trends of each parameter at different moments can be reflected. For example, a wind speed time series directional field diagram can display the direction and magnitude of wind speed changes over time at a point (e.g., the location of a wind turbine). The wind speed at each point in time can be represented by an arrow, with the length of the arrow representing the wind speed magnitude, and the direction of the arrow indicating the trend of wind speed increase or decrease.

[0068] Next, considering that each time series directional field map representation contains rich time series feature information, and there are inherent patterns and trends between the data at each time point, the dilated convolutional neural network can increase the receptive field without increasing the size of the convolution kernel by introducing holes in the convolution kernel, and can better process data with spatial continuity and long-range correlation, such as macro trends and features that may exist in the directional field map. Based on this, the present application obtains the wind speed time series directional field map representation, the wind direction angle time series directional field map representation and the predicted temperature time series directional field map representation by respectively passing the directional field feature extractor based on the dilated convolutional neural network to capture the subtle features and complex patterns in the directional field map, including local turbulence characteristics, macro wind field distribution characteristics, etc., to obtain the wind speed time series directional field feature vector, the wind direction angle time series directional field feature vector and the predicted temperature time series directional field feature vector.

[0069] Accordingly, if Figure 3As shown, in step S142, the direction field calculation and feature extraction are performed on the wind speed time series parameters, the wind direction angle time series parameters and the predicted temperature time series parameters respectively to obtain wind speed time series direction field features, wind direction angle time series direction field features and predicted temperature time series direction field features, including: S1421, respectively calculating the direction fields of the wind speed time series parameters, the wind direction angle time series parameters and the predicted temperature time series parameters to obtain wind speed time series direction field map representation, wind direction angle time series direction field map representation and predicted temperature time series direction field map representation; S1422, respectively passing the wind speed time series direction field map representation, the wind direction angle time series direction field map representation and the predicted temperature time series direction field map representation through a direction field feature extractor based on an expanded convolutional neural network to obtain a wind speed time series direction field feature vector as the wind speed time series direction field feature, a wind direction angle time series direction field feature vector as the wind direction angle time series direction field feature and a predicted temperature time series direction field feature vector as the predicted temperature time series direction field feature.

[0070] Then, considering that the wind speed time series direction field feature vector and the wind direction angle time series direction field feature vector represent the important features of wind speed and wind direction angle in different time and space dimensions respectively. However, in the actual wind power generation process, wind speed and wind direction angle do not affect power independently, but are interrelated and influence each other. For example, under different wind direction angles, the same wind speed may produce different power generation. Based on this, in the technical solution of the present application, the wind speed time series direction field feature vector and the wind direction angle time series direction field feature vector are passed through a wind condition multi-dimensional time series feature interaction fusion device based on an X-linear attention bilinear pooling unit to obtain a wind condition multi-dimensional time series feature interaction representation vector. Specifically, the X-linear attention mechanism can assign different weights according to the importance of the features, and the bilinear pooling unit can effectively handle the nonlinear relationship between features. The combination of the two can deeply explore the interactive information between different feature vectors, so that the generated vector integrates the multi-dimensional time series features of wind speed and wind direction angle and the interactive information between them, and can describe the wind condition more comprehensively and accurately.

[0071] Correspondingly, in step S143, the wind speed time series direction field feature and the wind direction angle time series direction field feature are interactively fused to obtain a wind condition multi-dimensional time series feature interactive representation, including: passing the wind speed time series direction field feature vector and the wind direction angle time series direction field feature vector through a wind condition multi-dimensional time series feature interactive fusion device based on an X-linear attention bilinear pooling unit to obtain a wind condition multi-dimensional time series feature interactive representation vector as the wind condition multi-dimensional time series feature interactive representation.

[0072] Accordingly, if Figure 4As shown, in step S144, the principal component-guided feature compensation interaction is performed on the wind condition multi-dimensional time series feature interaction representation and the predicted temperature time series direction field feature to obtain the wind condition-temperature time series direction field compensation interaction feature, and based on the wind condition-temperature time series direction field compensation interaction feature, the wind power feed-in power value at the next moment is obtained, including: S1441, the principal component-guided feature compensation interaction is performed on the wind condition multi-dimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain the wind condition-temperature time series direction field compensation interaction feature vector as the wind condition-temperature time series direction field compensation interaction feature; S1442, based on the wind condition-temperature time series direction field compensation interaction feature vector, the wind power feed-in power value at the next moment is obtained.

[0073] Specifically, considering that wind power generation is not only affected by the interaction of wind speed and wind direction, temperature is also an important indirect influencing factor. The previously obtained wind condition multi-dimensional time series feature interaction representation vector only integrates the time series interaction information of wind speed and wind direction, while the predicted temperature time series direction field feature vector represents the important features related to temperature time series. Therefore, in order to fully consider the comprehensive impact of wind and temperature factors on power generation, and to explore these potential key complex relationships, the present application performs principal component-guided feature compensation interaction on the wind condition multi-dimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to identify the main component that best represents its characteristics, and dynamically compensates and interacts their respective characteristics based on the difference between the two, to obtain the wind condition-temperature time series direction field compensation interaction feature vector.

[0074] Accordingly, if Figure 5As shown, in step S1441, the principal component-guided feature compensation interaction is performed on the wind condition multi-dimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain the wind condition-temperature time series direction field compensation interaction feature vector, including: S14411, feature principal component analysis is performed on the wind condition multi-dimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain a set of wind condition multi-dimensional time series feature principal component coding vectors and a set of predicted temperature time series direction field feature principal component coding vectors; S14412, the set of wind condition multi-dimensional time series feature principal component coding vectors and the set of predicted temperature time series direction field feature principal component coding vectors are combined. The set is constructed as a principal component aggregation coding feature map of the multi-dimensional time series characteristics of the wind condition and a principal component aggregation coding feature map of the predicted temperature time series characteristics; S14413, calculating the wind condition-predicted temperature time series difference embedding compensation coding weight vector between the principal component aggregation coding feature map of the multi-dimensional time series characteristics of the wind condition and the principal component aggregation coding feature map of the predicted temperature time series characteristics; S14414, based on the wind condition-predicted temperature time series difference embedding compensation coding weight vector, compensating and aggregating the set of the principal component coding vectors of the multi-dimensional time series characteristics of the wind condition and the set of the principal component coding vectors of the predicted temperature time series direction field characteristics to obtain the wind condition-temperature time series direction field compensation interaction feature vector.

[0075] Among them, in step S14413, the wind condition-prediction temperature time series difference embedding compensation coding weight vector between the principal component aggregation coding feature map of the wind condition multi-dimensional time series characteristics and the principal component aggregation coding feature map of the predicted temperature time series characteristics is calculated, including: performing difference embedding processing on the principal component aggregation coding feature map of the wind condition multi-dimensional time series characteristics and the principal component aggregation coding feature map of the predicted temperature time series characteristics to obtain the wind condition multi-dimensional time series branch weight vector and the predicted temperature time series branch weight vector; calculating the wind condition-prediction temperature time series difference embedding compensation coding weight vector between the wind condition multi-dimensional time series branch weight vector and the predicted temperature time series branch weight vector.

[0076] Wherein, in step S14414, based on the wind condition-predicted temperature time series difference embedding compensation coding weight vector, the set of the wind condition multidimensional time series feature principal component coding vector and the set of the predicted temperature time series direction field feature principal component coding vector are compensated and aggregated to obtain the wind condition-temperature time series direction field compensation interaction feature vector, including: calculating the positional mean vector of the set of the wind condition multidimensional time series feature principal component coding vector and the set of the predicted temperature time series direction field feature principal component coding vector to obtain the wind condition multidimensional time series principal component representation coding vector and the predicted temperature time series principal component representation coding vector; based on the wind condition-predicted temperature time series difference embedding compensation coding weight vector, the wind condition multidimensional time series principal component representation coding vector and the predicted temperature time series principal component representation coding vector are aggregated and interacted to obtain the wind condition-temperature time series direction field compensation interaction feature vector.

[0077] More specifically, principal component analysis is first performed on the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain a set of wind condition multidimensional time series feature principal component encoding vectors and a set of predicted temperature time series direction field feature principal component encoding vectors. Those skilled in the art should be aware that principal component analysis (PCA), as an unsupervised learning method, transforms a multidimensional dataset into a new coordinate system in which the first principal component has the maximum variance, and each subsequent component is orthogonal to all previous components and captures as much of the remaining variance as possible. In this technical solution, performing principal component analysis on the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector means finding the main directions in these two vector spaces to achieve feature dimensionality reduction and feature sparsification.

[0078] The process can be expressed as follows: ;in, is the interactive representation vector of the multi-dimensional time series characteristics of the wind conditions, is the characteristic principal component analysis operation, It is through The calculated covariance matrix of the multidimensional time series samples of wind conditions is: is the principal component orthogonal matrix of the multidimensional time series characteristics of wind conditions, is the principal component coding vector of each wind condition multi-dimensional time series feature in the set of principal component coding vectors of wind condition multi-dimensional time series features, is the diagonal matrix of multidimensional time series characteristics of wind conditions, They are and The corresponding weight value, for The transposed matrix of is the predicted temperature time series direction field feature vector, It is through The calculated predicted temperature time series sample covariance matrix, is the orthogonal matrix of the principal components of the predicted temperature time series features, is each predicted temperature time series direction field feature principal component coding vector in the set of predicted temperature time series direction field feature principal component coding vectors, is the diagonal matrix of predicted temperature time series features, They are and The corresponding weight value, for The transposed matrix of .

[0079] After obtaining the set of principal component encoding vectors of the multi-dimensional time series characteristics of the wind conditions and the set of principal component encoding vectors of the predicted temperature time series directional field characteristics, the set of principal component encoding vectors of the multi-dimensional time series characteristics of the wind conditions and the set of principal component encoding vectors of the predicted temperature time series directional field characteristics are constructed into a principal component aggregate encoding feature map of the multi-dimensional time series characteristics of the wind conditions and a principal component aggregate encoding feature map of the predicted temperature time series characteristics. It should be understood that each set of principal component encoding vectors can be regarded as a point cloud in a high-dimensional space, and constructing a feature map is to map these point clouds into a two-dimensional or three-dimensional space so that local neighborhood relationships are maintained. This mapping is crucial for subsequent spatial information processing because it determines which information will be retained for the next step.

[0080] The process can be expressed as follows: ;in, For the reshape operation, It is the principal component aggregation coding feature map of the multi-dimensional time series characteristics of wind conditions. It is the principal component aggregation coding feature map of the predicted temperature time series features. is the principal component coding vector of each wind condition multi-dimensional time series feature in the set of principal component coding vectors of wind condition multi-dimensional time series features, It is each predicted temperature time series direction field feature principal component coding vector in the set of predicted temperature time series direction field feature principal component coding vectors.

[0081] Next, the principal component aggregation coding feature map of the multi-dimensional time series features of the wind conditions and the principal component aggregation coding feature map of the predicted temperature time series features are input into the difference embedding unit to obtain the wind condition multi-dimensional time series branch weight vector and the predicted temperature time series branch weight vector. That is, as the feature map is formed, the design of the difference embedding unit is intended to capture the difference between the two feature maps. The process of difference embedding is not only to identify the differences, but more importantly to quantify the importance of these differences. The generated weight vector reflects the relative importance of each position, which is critical for the implementation of the subsequent compensation mechanism. From a data perspective, the role of the wind condition multi-dimensional time series branch weight vector and the predicted temperature time series branch weight vector is similar to a regulating factor, ensuring fair interaction and fusion even when there are differences in the original features.

[0082] The process can be expressed as follows: ;in, is the average pooling operation, and They are The corresponding first weight matrix and second weight matrix, and They are The corresponding first weight matrix and second weight matrix, is a nonlinear activation function, is the output activation function, is the wind condition multi-dimensional time series branch weight vector, is the predicted temperature timing branch weight vector.

[0083] In order to reduce the amount of data processing for aggregation interaction, before the aggregation interaction of the principal component aggregation coding feature map of the multi-dimensional time series characteristics of the wind condition and the principal component aggregation coding feature map of the predicted temperature time series characteristics based on the multi-dimensional time series branch weight vector of the wind condition and the predicted temperature time series branch weight vector, the positional mean vectors of the set of principal component coding vectors of the multi-dimensional time series characteristics of the wind condition and the set of principal component coding vectors of the predicted temperature time series direction field characteristics are calculated to obtain the principal component representation coding vector of the multi-dimensional time series characteristics of the wind condition and the principal component representation coding vector of the predicted temperature time series. It should be understood that this step is essentially another form of data compression. Unlike PCA, this step focuses on the average of the eigenvalues at the same position. The calculation of the mean vector lays the foundation for the subsequent aggregation interaction, because these vectors represent the core information of their respective eigenvectors.

[0084] The process can be expressed as follows: ;in, is the wind condition multi-dimensional time series branch weight vector, is the predicted temperature timing branch weight vector, It is subtracted by position point, To take the absolute value operation, is the wind condition-forecast temperature time series difference embedding compensation encoding weight vector, The first principal component encoding vector in the set of multi-dimensional time series characteristics of the wind condition is The principal component coding vector of the multidimensional time series characteristics of wind conditions, The first element in the set of principal component coding vectors of the predicted temperature time series direction field feature is The principal component encoding vector of the predicted temperature time series direction field features, is the number of vectors in the set of principal component coding vectors of wind condition multidimensional time series characteristics and the set of principal component coding vectors of predicted temperature time series direction field characteristics, is the encoding vector representing the principal component of the multidimensional time series of wind conditions, It is the encoding vector representing the principal component of the predicted temperature time series.

[0085] Finally, based on the difference embedding compensation coding weight vector, the wind condition multidimensional time series principal component representation coding vector and the predicted temperature time series principal component representation coding vector are aggregated and interacted to obtain the wind condition-temperature time series directional field compensation interaction feature vector. Here, the difference embedding compensation coding weight vector guides the adjustment of the interaction method based on the difference between the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series directional field feature vector. The aggregated interaction not only integrates the information from the two original features, but also ensures the sensitivity and adaptability of the interaction result to the differences between the original features by introducing a compensation mechanism.

[0086] The process can be expressed as follows: ;in, is the wind condition-forecast temperature time series difference embedding compensation encoding weight vector, It is the point product of position. is the encoding vector representing the principal component of the multidimensional time series of wind conditions, is the encoding vector representing the principal component of the predicted temperature time series, is point convolutional coding, is a nonlinear activation function, is the wind condition-temperature time series direction field compensation interaction eigenvector.

[0087] The wind condition-temperature time-series direction field compensation interaction feature vector is then passed through a decoder-based wind power source predictor to obtain a decoded value of the wind power feed-in power at the next moment. Specifically, the wind condition-temperature time-series direction field compensation interaction feature vector, obtained by principal component compensation interaction using the wind condition multi-dimensional time-series feature interaction representation vector and the predicted temperature time-series direction field feature vector, is decoded to obtain a decoded value of the wind power feed-in power at the next moment by utilizing the decoder-based wind power source predictor's ability to learn the complex mapping relationship between feature vectors and power values.

[0088] Accordingly, in step S1442, based on the wind condition-temperature time series direction field compensation interaction characteristic vector, the wind power feed-in power value at the next moment is obtained, including: passing the wind condition-temperature time series direction field compensation interaction characteristic vector through a decoder-based wind power predictor to obtain a wind power feed-in power decoding value as the wind power feed-in power value at the next moment.

[0089] In a preferred example, the wind condition-temperature time series direction field compensation interaction feature vector is passed through a decoder-based wind power source predictor to obtain a decoded value of wind power feed-in power at the next moment, including:

[0090] Determine the eigenvalue mean corresponding to the wind condition-temperature time series direction field compensation interaction eigenvector and the standard deviation of the eigenvalues ;

[0091] The first wind condition-temperature time series direction field compensation interaction fairness target vector is obtained by multiplying the point-to-point subtraction vector of the wind condition-temperature time series direction field compensation interaction eigenvector and the eigenvalue mean by the eigenvalue standard deviation: ;in, represents the wind condition-temperature time series direction field compensation interaction eigenvector, represents the mean eigenvalue of the wind condition-temperature time series direction field compensation interaction eigenvector corresponding to the mean eigenvalue, represents the standard deviation of the eigenvalue corresponding to the wind condition-temperature time series direction field compensation interaction eigenvector, Indicates point reduction, represents dot product, represents the first wind condition-temperature time series direction field compensation interaction fairness target vector;

[0092] The second wind condition-temperature time series direction field compensation interaction fairness target vector is obtained by multiplying the point-to-point subtraction vector of the wind condition-temperature time series direction field compensation interaction eigenvector and the eigenvalue standard deviation by the eigenvalue mean: ;in, represents the wind condition-temperature time series direction field compensation interaction eigenvector, represents the mean eigenvalue of the wind condition-temperature time series direction field compensation interaction eigenvector corresponding to the mean eigenvalue, represents the standard deviation of the eigenvalue corresponding to the wind condition-temperature time series direction field compensation interaction eigenvector, Indicates point reduction, represents dot product, represents the second wind condition-temperature time series direction field compensation interaction fairness target vector;

[0093] After multiplying the bit-by-bit reciprocal of the second wind condition-temperature time series direction field compensation interactive fairness target vector by the first wind condition-temperature time series direction field compensation interactive fairness target vector, take the bit-by-bit logarithm with base 2 to obtain the wind condition-temperature time series direction field compensation interactive information correction vector: ;in, represents the first wind condition-temperature time series direction field compensation interactive fairness target vector, represents the second wind condition-temperature time series direction field compensation interaction fairness target vector, represents the bit-by-bit inverse of the second wind condition-temperature time series direction field compensation interactive fairness target vector, represents dot product, represents the logarithmic function with base 2, represents the wind condition-temperature time series direction field compensation mutual information correction vector;

[0094] The mean of the eigenvalues Divide by the standard deviation of the eigenvalue The square root of the quotient is multiplied by the weight hyperparameter, and then combined with the wind condition-temperature time series direction field compensation interaction information correction vector point to obtain the optimized wind condition-temperature time series direction field compensation interaction feature vector: ;in, represents the mean eigenvalue of the wind condition-temperature time series direction field compensation interaction eigenvector corresponding to the mean eigenvalue, represents the standard deviation of the eigenvalue corresponding to the wind condition-temperature time series direction field compensation interaction eigenvector, represents the weight hyperparameter, Indicates point addition, represents the wind condition-temperature time series direction field compensation mutual information correction vector, A characteristic vector representing the optimized wind condition-temperature time series direction field compensation interaction;

[0095] The optimized wind condition-temperature time series direction field compensation interaction characteristic vector is passed through a decoder-based wind power source predictor to obtain a decoded value of wind power source feed-in power at the next moment.

[0096] Here, since the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series directional field feature vector respectively represent the multidimensional interactive fusion features of the local correlation of the wind speed-wind direction angle directional field and the local correlation features of the predicted temperature directional field, when performing feature principal component compensation interaction, the source time series directional field image semantic feature population attributes corresponding to different data will have interaction fairness differences based on the principal component compensation level, thereby affecting the feature distribution interaction inclusiveness of the wind condition-temperature time series directional field compensation interaction feature vector, and reducing the accuracy of the decoding results obtained by the decoder-based wind power source predictor.

[0097] Therefore, taking into account the attribute level fairness differences of the data population corresponding to the sequence fusion features of the wind condition-temperature time series directional field compensation interaction feature vector, in order to improve the interactive inclusiveness under the feature distribution diversity of the wind condition-temperature time series directional field compensation interaction feature vector, the cross probability value constraint based on the wind condition-temperature time series directional field compensation interaction feature vector is used as the interactive fairness target representation to correct the group feature information interactive propagation of the wind condition-temperature time series directional field compensation interaction feature vector, and the unified statistical feature response interaction based on the wind condition-temperature time series directional field compensation interaction feature vector is used as the feature distribution multi-level fairness target bias to achieve a robust distribution fairness unified representation of the wind condition-temperature time series directional field compensation interaction feature vector, forming a fair collaboration paradigm under the feature distribution framework of the wind condition-temperature time series directional field compensation interaction feature vector, thereby improving the accuracy of the next moment wind power feed-in power decoding value obtained by the decoder-based wind power predictor.

[0098] Based on the above embodiments, see Figure 6As shown, it is a structural diagram of an AI-based new energy power generation prediction system 100 in an embodiment of the present application. The AI-based new energy power generation prediction system 100 includes: a predicted radiation curve acquisition module 110, which is used to divide the photovoltaic power supply cycle and determine the predicted radiation curve of the next cycle based on the predicted meteorological data and the location of the photovoltaic power supply; a correction module 120, which is used to correct the predicted radiation curve to obtain a corrected predicted radiation curve; a predicted photovoltaic power supply feed-in power curve acquisition module 130, which is used to obtain a temperature prediction change curve, and obtain a predicted photovoltaic power supply feed-in power curve based on the temperature prediction change curve and the corrected predicted radiation curve; a predicted feed-in power calculation module 140, which is used to calculate the predicted feed-in power of the wind power supply at each moment to obtain a predicted wind power supply feed-in power curve; a strategy adjustment module 150, which is used to adjust the hydrogen energy storage strategy based on the predicted photovoltaic power supply feed-in power curve, the predicted wind power supply feed-in power curve and the real-time feed-in power.

[0099] In one example, the strategy adjustment module 150 includes: a timing parameter acquisition unit for acquiring the predicted wind condition timing parameters of the location of the wind power source, and acquiring the predicted temperature timing parameters of the location of the wind power source, wherein the predicted wind condition timing parameters include wind speed timing parameters and wind direction angle timing parameters; a feature extraction unit for performing direction field calculation and feature extraction on the wind speed timing parameters, the wind direction angle timing parameters and the predicted temperature timing parameters respectively to obtain wind speed timing direction field features, wind direction angle timing direction field features and predicted temperature timing direction field features. Characteristic; a multi-dimensional time series feature interactive fusion unit, used for performing a wind condition multi-dimensional time series feature interactive fusion on the wind speed time series directional field feature and the wind direction angle time series directional field feature to obtain a wind condition multi-dimensional time series feature interactive representation; a next moment wind power supply feed-in power value acquisition unit, used for performing a principal component guided feature compensation interaction on the wind condition multi-dimensional time series feature interactive representation and the predicted temperature time series directional field feature to obtain a wind condition-temperature time series directional field compensation interactive feature, and based on the wind condition-temperature time series directional field compensation interactive feature, obtain the wind power supply feed-in power value at the next moment.

[0100] In one example, the feature extraction unit is used to: respectively calculate the direction fields of the wind speed time series parameters, the wind direction angle time series parameters and the predicted temperature time series parameters to obtain a wind speed time series direction field diagram representation, a wind direction angle time series direction field diagram representation and a predicted temperature time series direction field diagram representation; and respectively pass the wind speed time series direction field diagram representation, the wind direction angle time series direction field diagram representation and the predicted temperature time series direction field diagram representation through a direction field feature extractor based on an expanded convolutional neural network to obtain a wind speed time series direction field feature vector as the wind speed time series direction field feature, a wind direction angle time series direction field feature vector as the wind direction angle time series direction field feature and a predicted temperature time series direction field feature vector as the predicted temperature time series direction field feature.

[0101] Here, those skilled in the art will appreciate that the specific functions and operations of each module in the above-mentioned AI-based new energy power generation prediction system 100 have been referred to above. Figures 1 to 5 The description of the AI-based new energy power generation power prediction method has been introduced in detail, and therefore, its repeated description will be omitted.

[0102] Those skilled in the art will appreciate that all or part of the steps in the above method can be performed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disk. Alternatively, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software functional modules. This application is not limited to any particular form of combination of hardware and software.

[0103] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology and should not be interpreted in an idealized or highly formal sense, unless expressly defined as such herein.

[0104] The above is an explanation of the present application and should not be considered as limiting thereof. Although several exemplary embodiments of the present application have been described, those skilled in the art will readily appreciate that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application.

Claims

1. A method for predicting renewable energy power generation based on AI, comprising: Divide the photovoltaic power supply cycle, and determine the predicted radiation curve of the next cycle based on the predicted meteorological data and the location of the photovoltaic power supply; correct the predicted radiation curve to obtain a corrected predicted radiation curve; obtain a temperature prediction change curve, and obtain a predicted photovoltaic power supply feed-in power curve based on the temperature prediction change curve and the corrected predicted radiation curve; calculate the predicted feed-in power of the wind power supply at each moment to obtain a predicted wind power supply feed-in power curve; adjust the hydrogen energy storage strategy based on the predicted photovoltaic power supply feed-in power curve, the predicted wind power supply feed-in power curve and the real-time feed-in power; characterized in that calculating the predicted feed-in power of the wind power supply at each moment to obtain the predicted wind power supply feed-in power curve includes: Obtaining predicted wind condition time series parameters at the location of the wind power source, and obtaining predicted temperature time series parameters at the location of the wind power source, wherein the predicted wind condition time series parameters include wind speed time series parameters and wind direction angle time series parameters; Calculating the direction fields of the wind speed time series parameters, the wind direction angle time series parameters, and the predicted temperature time series parameters respectively to obtain a wind speed time series direction field diagram representation, a wind direction angle time series direction field diagram representation, and a predicted temperature time series direction field diagram representation; passing the wind speed time series direction field diagram representation, the wind direction angle time series direction field diagram representation, and the predicted temperature time series direction field diagram representation respectively through a direction field feature extractor based on an expanded convolutional neural network to obtain a wind speed time series direction field feature vector as the wind speed time series direction field feature, a wind direction angle time series direction field feature vector as the wind direction angle time series direction field feature, and a predicted temperature time series direction field feature vector as the predicted temperature time series direction field feature; Performing interactive fusion of wind condition multi-dimensional time series features on the wind speed time series direction field features and the wind direction angle time series direction field features to obtain an interactive representation of wind condition multi-dimensional time series features; Performing feature principal component analysis on the wind condition multidimensional time series feature interaction representation vector and the predicted temperature time series direction field feature vector to obtain a set of wind condition multidimensional time series feature principal component coding vectors and a set of predicted temperature time series direction field feature principal component coding vectors; constructing the set of wind condition multidimensional time series feature principal component coding vectors and the set of predicted temperature time series direction field feature principal component coding vectors into a wind condition multidimensional time series feature principal component aggregation coding feature graph and a predicted temperature time series feature principal component aggregation coding feature graph; calculating a wind condition-predicted temperature time series difference embedding compensation coding weight vector between the wind condition multidimensional time series feature principal component aggregation coding feature graph and the predicted temperature time series feature principal component aggregation coding feature graph; based on the wind condition-predicted temperature time series difference embedding compensation coding weight vector, performing compensation aggregation on the set of wind condition multidimensional time series feature principal component coding vectors and the set of predicted temperature time series direction field feature principal component coding vectors to obtain a wind condition-temperature time series direction field compensation interaction feature vector; Based on the wind condition-temperature time series direction field compensation interaction characteristic vector, the wind power feed-in power value at the next moment is obtained.

2. The AI-based new energy power generation prediction method according to claim 1 is characterized in that: The wind speed time series direction field feature vector and the wind direction angle time series direction field feature vector are passed through a wind condition multidimensional time series feature interaction fusion device based on an X-linear attention bilinear pooling unit to obtain a wind condition multidimensional time series feature interaction representation vector as the wind condition multidimensional time series feature interaction representation.

3. The AI-based new energy power generation prediction method according to claim 2, characterized in that: Performing differential embedding processing on the principal component aggregation coding feature map of the wind condition multidimensional time series feature and the principal component aggregation coding feature map of the predicted temperature time series feature to obtain a wind condition multidimensional time series branch weight vector and a predicted temperature time series branch weight vector; The wind condition-predicted temperature time series difference embedding compensation coding weight vector is calculated between the wind condition multi-dimensional time series branch weight vector and the predicted temperature time series branch weight vector.

4. The AI-based new energy power generation prediction method according to claim 3 is characterized in that: Calculating the positional mean vector of the set of principal component coding vectors of the wind condition multidimensional time series characteristics and the set of principal component coding vectors of the predicted temperature time series directional field characteristics to obtain the wind condition multidimensional time series principal component representation coding vector and the predicted temperature time series principal component representation coding vector; Based on the wind condition-predicted temperature time series difference embedding compensation coding weight vector, the wind condition multidimensional time series principal component characterization coding vector and the predicted temperature time series principal component characterization coding vector are aggregated and interacted to obtain the wind condition-temperature time series directional field compensation interaction feature vector.

5. The AI-based new energy power generation prediction method according to claim 4 is characterized in that: The wind condition-temperature time series direction field compensation interaction feature vector is passed through a decoder-based wind power source predictor to obtain a wind power source feed-in power decoding value as the wind power source feed-in power value at the next moment.

6. An AI-based renewable energy power generation prediction system, using the AI-based renewable energy power generation prediction method according to any one of claims 1 to 5.

Images