A method, apparatus, equipment, and storage medium for determining a power risk prediction model.

Abstract

This application discloses a method, apparatus, device, and storage medium for determining a power risk prediction model. The method includes: determining a radiation monitoring model for a target power market area based on multi-source spatiotemporal data and topographic sequence data; adjusting the state variables and model parameters of the radiation monitoring model based on a multi-scale variational assimilation algorithm and the multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset; determining an initial power risk prediction model for the target power market area based on an improved long short-term memory network algorithm and the multi-source spatiotemporal data; determining the initial power risk prediction result for the target power market area based on the initial power risk prediction model and the radiation spatiotemporal dataset; and optimizing the initial power risk prediction model based on the radiation spatiotemporal dataset and the initial power risk prediction result to obtain the target power risk prediction model. The above technical solution helps improve the accuracy of the model in predicting power risks.

Description

Technical Field

[0001] This application relates to the field of power risk prediction technology, and more particularly to the field of power market-related risk prediction technology, specifically to a method, apparatus, equipment, and storage medium for determining a power risk prediction model. Background Technology

[0002] Solar radiation monitoring is one of the core technologies in the renewable energy sector, especially in photovoltaic power generation and electricity market trading.

[0003] The spatiotemporal variability of solar radiation directly affects photovoltaic power generation forecasts, thereby amplifying electricity market risks such as price fluctuations and supply-demand imbalances. Inaccurate risk forecasts increase transaction costs, prolong decision-making cycles, and amplify potential risks in the energy market, such as market losses due to underestimation of power generation.

[0004] Currently, existing technologies are limited by the depth of data integration and the adaptability of models, making it impossible to achieve efficient and accurate risk prediction optimization, and thus unable to fully meet the needs of refined risk management under complex spatiotemporal environments and volatile market conditions. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, and storage medium for determining a power risk prediction model, so as to improve the accuracy of the model in predicting power risks and thus meet the needs of refined risk management in a complex energy market environment.

[0006] According to one aspect of this application, a method for determining a power risk prediction model is provided, the method comprising:

[0007] Based on multi-source spatiotemporal data and topographic sequence data of the target electricity market area, a radiation monitoring model for the target electricity market area is determined; wherein, the multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market electricity allocation data;

[0008] Based on the multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to the multi-source spatiotemporal data to obtain the radiation spatiotemporal dataset;

[0009] Based on the improved long short-term memory network algorithm, an initial power risk prediction model for the target power market region is determined according to the multi-source spatiotemporal data.

[0010] Based on the initial power risk prediction model, and according to the radiation spatiotemporal dataset, the initial power risk prediction results for the target power market area are determined;

[0011] Based on multi-scale fusion components and graph neural networks, the initial power risk prediction model is optimized according to the radiation spatiotemporal dataset and the initial power risk prediction results to obtain the target power risk prediction model.

[0012] According to another aspect of this application, an apparatus for determining a power risk prediction model is provided, the apparatus comprising:

[0013] The monitoring model determination module is used to determine the radiation monitoring model of the target electricity market area based on multi-source spatiotemporal data and topographic sequence data of the target electricity market area; wherein, the multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data and market electricity allocation data;

[0014] The dataset determination module is used to adjust the state variables and model parameters of the radiation monitoring model based on the multi-scale variational assimilation algorithm and the multi-source spatiotemporal data to obtain the radiation spatiotemporal dataset.

[0015] The prediction module is used to determine the initial power risk prediction model for the target power market area based on the improved long short-term memory network algorithm and the multi-source spatiotemporal data.

[0016] The prediction result determination module is used to determine the initial power risk prediction result of the target power market area based on the initial power risk prediction model and the radiation spatiotemporal dataset.

[0017] The target model determination module is used to optimize the initial power risk prediction model based on the radiation spatiotemporal dataset and the initial power risk prediction results, using a multi-scale fusion component and a graph neural network, to obtain the target power risk prediction model.

[0018] According to another aspect of this application, an electronic device is provided, the electronic device comprising:

[0019] One or more processors;

[0020] Memory, used to store one or more programs;

[0021] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the power risk prediction model determination methods provided in the embodiments of this application.

[0022] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements any of the power risk prediction model determination methods provided in the embodiments of this application.

[0023] According to another aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the power risk prediction model determination methods provided in the embodiments of this application.

[0024] This application determines a radiation monitoring model for a target electricity market area based on multi-source spatiotemporal data and topographic sequence data. The multi-source spatiotemporal data includes ground-based radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market electricity allocation data. Based on a multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to the multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset. Based on an improved long short-term memory network algorithm, an initial electricity risk prediction model for the target electricity market area is determined according to the multi-source spatiotemporal data. Based on the initial electricity risk prediction model, the initial electricity risk prediction result for the target electricity market area is determined according to the radiation spatiotemporal dataset. Based on a multi-scale fusion component and a graph neural network, the initial electricity risk prediction model is optimized according to the radiation spatiotemporal dataset and the initial electricity risk prediction result to obtain the target electricity risk prediction model. The above technical solution determines the required power risk prediction model by combining radiation monitoring models, multi-scale variational assimilation algorithms, and improved long short-term memory networks, and uses this model to predict power market-related risks. It takes into account the impact of the spatiotemporal variability of solar radiation on market risks, and balances prediction accuracy and risk quantification efficiency. This enables the model to efficiently and accurately determine the prediction results of power market-related risks, meeting the needs of refined risk management in complex energy market environments. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for determining a power risk prediction model according to Embodiment 1 of this application;

[0026] Figure 2 This is a flowchart of a method for determining a power risk prediction model according to Embodiment 2 of this application;

[0027] Figure 3 This is a schematic diagram of the structure of a power risk prediction model determination device provided in Embodiment 3 of this application;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the power risk prediction model determination method of Embodiment 4 of this application. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] Furthermore, it should be noted that the collection, storage, use, processing, transmission, provision, and disclosure of multi-source spatiotemporal data and terrain sequence data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0032] Example 1

[0033] Figure 1 This is a flowchart of a method for determining a power risk prediction model according to Embodiment 1 of this application. This embodiment is applicable to situations where a power risk prediction model based on solar radiation is trained to predict power market-related risks. It can be executed by a power risk prediction model determination device, which can be implemented in hardware and / or software and can be configured in a computer device, such as a server. Figure 1 As shown, the method includes:

[0034] S110. Based on the multi-source spatiotemporal data and topographic sequence data of the target electricity market area, determine the radiation monitoring model of the target electricity market area; wherein, the multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data and market electricity allocation data.

[0035] In this embodiment, the target electricity market area refers to a specific geographical area or electricity grid area that is of interest or defined in electricity market analysis and management, used for electricity market risk assessment, resource allocation, monitoring, and optimization. This area can be part of a physical power grid or a market unit divided according to demand and analysis objectives. Multi-source spatiotemporal data refers to data with temporal and spatial attributes collected from multiple sources (such as ground monitoring, satellite remote sensing, and market electricity allocation). Topographic sequence data refers to spatiotemporal data from the Global Digital Elevation Model (GEM), which is a digital representation of global surface height information and includes data in both temporal and spatial dimensions; in short, it includes elevation (i.e., height) information of the Earth's surface, and this information changes with time and geographical location. Ground radiation monitoring data refers to high-precision radiation sequences (such as spatiotemporal sequences of total irradiance and scattering components) from ground monitoring observations, as well as auxiliary parameters such as sunshine sequences and radiation intensity provided by satellite remote sensing products; these data collectively depict the spatiotemporal distribution characteristics of regional radiation resources. Satellite remote sensing spatiotemporal sequence data refers to remote sensing data with temporal and spatial attributes acquired by satellites, typically including meteorological variables such as temperature, humidity, and air pressure, as well as radiation data. Electricity market allocation data refers to the collection of various data used in the operation, management, analysis, and decision-making processes of the electricity market. This data covers various aspects such as electricity supply and demand, market transactions, resource allocation, price fluctuations, and load forecasting, and can help the electricity market optimize and manage risks. Radiation monitoring models are monitoring models that combine multi-source data (ground monitoring data, satellite remote sensing data, etc.) with topographic sequence data to create a model capable of predicting radiation levels within a region, particularly for the needs of electricity market areas.

[0036] Optionally, the ground radiation monitoring data includes at least hourly radiation indices; the satellite remote sensing spatiotemporal sequence data includes at least meteorological variable data; correspondingly, multidimensional data fusion is performed on the ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market power allocation data of the target power market area to obtain radiation observation data of the target power market area; based on the topographic sequence data, hourly radiation indices, and meteorological variable data of the target power market area, a high spatiotemporal resolution regional radiation static parameter field of the target power market area is determined; using the regional radiation static parameter field as boundary conditions, a radiation monitoring model for the target power market area is determined based on the radiation observation data.

[0037] In this embodiment, the hourly radiation index refers to the radiation value per hour, typically measured by ground meteorological stations or satellite remote sensing equipment, covering shortwave radiation, longwave radiation, etc. Meteorological variable data refers to key data affecting radiation and climate models, such as temperature, humidity, wind speed, and air pressure. Radiation observation data refers to the combined and analyzed dataset of data from different sources and of different types (such as ground radiation monitoring data, satellite remote sensing data, etc.). The regional static radiation parameter field refers to the spatial distribution model of radiation parameters (such as radiation intensity, radiation distribution, etc.) at different times and geographical locations within a specific geographical area; this parameter field typically does not change over time and is a static description of the radiation conditions in that area. Boundary conditions refer to the constraints on the model or equations within a specific spatial or temporal range; for example, when constructing a radiation monitoring model, the regional static radiation parameter field is used as boundary conditions. These parameters help define the external input conditions of the model, enabling the model to reflect the actual situation in the real world.

[0038] S120. Based on the multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset.

[0039] In this embodiment, the multi-scale variational assimilation algorithm refers to processing and fusing data at different time and spatial scales (such as short-term and long-term data, local and large-scale data). This approach allows for better parameter optimization and state adjustment of the radiation monitoring model, improving prediction accuracy. State variables typically refer to the quantitative parameters within the model used to describe radiation conditions; these variables can be radiation intensity, cloud cover, ground reflectivity, etc. By adjusting these state variables, the model can better reflect the actual radiation situation. Model parameters are numerical constants used to describe and control the radiation monitoring model; these parameters are usually set based on physical laws or observational data. By adjusting these parameters, the model can more accurately simulate the distribution and changes in radiation. The radiation spatiotemporal dataset refers to the dataset processed through multi-source data fusion and variational assimilation algorithms, possessing both temporal and spatial characteristics.

[0040] Optionally, multi-source spatiotemporal data can be used as physical constraint data, and the initial field and key parameters of the radiation monitoring model can be iteratively adjusted through a multi-scale variational assimilation algorithm. By adjusting the initial field and key parameters, the state variables and model parameters of the radiation monitoring model can be adjusted, and the radiation spatiotemporal state dataset generated iteratively in the adjustment of state variables and parameters can be obtained to form the radiation spatiotemporal dataset generated by the radiation monitoring model.

[0041] In this embodiment, physical constraint data refers to data obtained according to physical laws or actual conditions that need to be followed during model establishment. This data provides constraints on the model output, ensuring the authenticity and scientific validity of the simulation results. In radiation monitoring models, physical constraint data typically includes meteorological data, topographic data, and other physical information related to radiation propagation. The initial field refers to the preliminary state or data distribution upon which the radiation monitoring model relies at the start of calculations; for example, the initial radiation intensity distribution, ground reflection radiation, etc. These initial data are the starting point for model calculations, and as data is fused and iteratively optimized, the model gradually approaches the true radiation distribution. Key parameters refer to numerical constants or coefficients that play an important role in the radiation monitoring model; these parameters directly affect the output results of the radiation monitoring model, such as the radiative transfer coefficient and atmospheric transmittance. By adjusting these key parameters, the model can more accurately simulate the radiation distribution. The radiation spatiotemporal state dataset refers to a dataset containing temporal and spatial features obtained through optimization using multi-source spatiotemporal data and variational assimilation algorithms; this dataset provides a more accurate description of the radiation state by adjusting the initial field and key parameters.

[0042] For example, multi-source spatiotemporal data serves as a physical constraint, guiding the model's state variables and parameters to continuously approximate the real situation. By iteratively adjusting the initial field and key parameters of the radiation monitoring model through a multi-scale variational assimilation algorithm, a set of spatiotemporally continuous, high-precision radiation spatiotemporal datasets with consistent internal physical processes is finally generated, providing sufficient and reliable data support for subsequent optimization to obtain a multi-scale spatiotemporal sequence prediction model.

[0043] S130. Based on the improved long short-term memory network algorithm, the initial power risk prediction model for the target power market area is determined according to multi-source spatiotemporal data.

[0044] In this embodiment, the improved Long Short-Term Memory (LSTM) network algorithm is an improved version of the standard LSM network algorithm, specifically designed for the particular scenario of electricity market-related risk prediction based on solar radiation. The initial electricity risk prediction model is an electricity risk model established under preliminary algorithm and data processing; it is the starting version of risk prediction and may be based on simple assumptions and preliminary parameter settings; as more data is introduced and the model is further optimized, this initial model will gradually become more refined.

[0045] Optionally, the improved long short-term memory network algorithm updates the sequence weights based on an adaptive attention mechanism, uses the minimum market volatility threshold as a constraint, takes maximizing prediction accuracy and minimizing risk bias as the objective function, and uses the difference between the product of the spatiotemporal coverage percentage and the market weight and the uncertainty penalty as the fitness function.

[0046] Adaptive attention is a technique that dynamically adjusts the focus based on the characteristics of the data. In time series data processing tasks, attention helps the model focus on the most relevant parts of the prediction result while ignoring irrelevant parts. Adaptiveness refers to adjusting the attention allocation strategy according to different input data, allowing the model to adapt more flexibly to changes in the data. The minimum market volatility threshold refers to the minimum level of volatility in the electricity market. If market volatility is below this threshold, the model may consider the market to be in a stable state. This threshold provides a criterion for the prediction model, helping it adjust its prediction strategy when volatility is low to avoid unnecessary overreaction. Maximizing prediction accuracy means the model needs to predict the future state of the electricity market as accurately as possible. Minimizing risk bias means the model should minimize the gap between the predicted and actual results, thereby reducing the risk in decision-making. The objective function is the value to be maximized or minimized during the optimization process. Spatiotemporal coverage percentage refers to the proportion of data that the model can cover in both time and space dimensions. Market weight refers to the relative importance of different markets (e.g., different regions or different electricity markets) in the prediction. The product of spatiotemporal coverage percentage and market weight is a quantitative indicator measuring the model's predictive performance in different regions and time periods. Uncertainty penalty is a constraint introduced during the optimization process to penalize predictions with high uncertainty in the model. Prediction uncertainty can stem from factors such as data noise and limitations of the model itself. The purpose of this penalty is to encourage the model to produce more stable and reliable predictions, reducing erroneous predictions caused by excessive uncertainty. The fitness function measures whether a solution is suitable during the optimization process. In this model, the fitness function is optimized by comparing the difference between the product of the spatiotemporal coverage percentage and the market weights and the uncertainty penalty. The goal of the fitness function is to balance prediction accuracy and model stability, ensuring that optimal market predictions are made while satisfying volatility constraints.

[0047] Understandably, by combining an improved Long Short-Term Memory (LSTM) network algorithm to construct the initial spatiotemporal sequence model, the algorithm's sequence optimization advantages are fully utilized. Furthermore, the addition of an adaptive update mechanism enables the algorithm to handle long-term variations under market changes, improving its robustness and accuracy. Simultaneously, by incorporating multi-source spatiotemporal data through multi-source fusion, the algorithm balances maximizing accuracy with minimizing risk bias, achieving a balance between accuracy and efficiency in predicting electricity market risks. This improves the prediction accuracy, reliability, and effectiveness of the initial spatiotemporal sequence model, thereby enhancing the sustainability of energy market management.

[0048] S140. Based on the initial power risk prediction model, determine the initial power risk prediction results for the target power market area according to the radiation spatiotemporal dataset.

[0049] In this embodiment, the initial power risk prediction result refers to the preliminary assessment result calculated based on existing data and algorithms. These results may include assessments of potential risks in the power market, such as insufficient power supply, price fluctuations, and changes in demand, as well as a risk prediction error sequence. The risk prediction error sequence refers to the sequence of differences between the predicted values ​​and the actual observed values ​​calculated in the initial power risk prediction model. This error sequence helps to analyze the accuracy of the model at different time and space points.

[0050] For example, the radiation spatiotemporal dataset is input into the initial power risk prediction model to obtain the initial power risk prediction results for the target power market area.

[0051] S150. Based on multi-scale fusion components and graph neural networks, the initial power risk prediction model is optimized according to the radiation spatiotemporal dataset and the initial power risk prediction results to obtain the target power risk prediction model.

[0052] In this embodiment, the multi-scale fusion component is a technique that improves the performance of a prediction model by combining information from different time and spatial scales. In power risk prediction, the power market is affected by many factors (such as climate change, seasonal demand fluctuations, etc.), which manifest differently at different time scales (hours, days, months) and spatial scales (regions, countries). The multi-scale fusion component can effectively integrate these different levels of influence information, thereby improving the model's predictive ability. Graph neural networks refer to a neural network model that represents data and learns through a graph structure. In power market prediction, graph neural networks can help model the interrelationships between different market regions in the power system, such as power flow and market price linkages between different regions. Through graph neural networks, the model can capture complex dependencies between regions, improving the overall prediction effect. The target power risk prediction model refers to the final model obtained after optimization by the multi-scale fusion component and graph neural network based on the initial model. This model can provide a more accurate and detailed power market risk assessment, considering more factors and more complex market interactions.

[0053] Understandably, the target power risk prediction model is the final model optimized by introducing multi-scale fusion components and graph neural network technology based on the initial power risk prediction results. It can comprehensively process information from different time and space dimensions, reveal the complex relationships in the power market, thereby improving the accuracy and reliability of power risk prediction and providing more scientific data support for power market management and decision-making.

[0054] This application embodiment determines a radiation monitoring model for a target electricity market area based on multi-source spatiotemporal data and topographic sequence data. The multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market electricity allocation data. Based on a multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to the multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset. Based on an improved long short-term memory network algorithm, an initial electricity risk prediction model for the target electricity market area is determined according to the multi-source spatiotemporal data. Based on the initial electricity risk prediction model, the initial electricity risk prediction result for the target electricity market area is determined according to the radiation spatiotemporal dataset. Based on a multi-scale fusion component and a graph neural network, the initial electricity risk prediction model is optimized according to the radiation spatiotemporal dataset and the initial electricity risk prediction result to obtain the target electricity risk prediction model. The above technical solution determines the required power risk prediction model by combining radiation monitoring models, multi-scale variational assimilation algorithms, and improved long short-term memory networks, and uses this model to predict power market-related risks. It takes into account the impact of the spatiotemporal variability of solar radiation on market risks, and balances prediction accuracy and risk quantification efficiency. This enables the model to efficiently and accurately determine the prediction results of power market-related risks, meeting the needs of refined risk management in complex energy market environments.

[0055] Example 2

[0056] Figure 2 This is a flowchart of a method for determining a power risk prediction model according to Embodiment 2 of this application. Based on the technical solutions of the above embodiments, this embodiment refines the process of "optimizing the initial power risk prediction model based on a multi-scale fusion component and a graph neural network, according to the radiation spatiotemporal dataset and the initial power risk prediction results, to obtain a target power risk prediction model" into "determining the regional average risk coverage rate of the target power market area based on the initial power risk prediction results; and extracting features from the radiation spatiotemporal dataset when the regional average risk coverage rate meets the feature extraction conditions, obtaining a static spatiotemporal feature sequence and a dynamic multi-scale feature sequence; determining the target correction component of the initial power risk prediction results based on the multi-scale fusion component and the graph neural network, according to the static spatiotemporal feature sequence and the dynamic multi-scale feature sequence; and optimizing the initial power risk prediction model based on the target correction component and the initial power risk prediction results to obtain the target power risk prediction model." It should be noted that for parts not detailed in this embodiment, please refer to the relevant descriptions in other embodiments. Figure 2 As shown, the method includes:

[0057] S210. Based on the multi-source spatiotemporal data and topographic sequence data of the target electricity market area, determine the radiation monitoring model of the target electricity market area; wherein, the multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data and market electricity allocation data.

[0058] S220. Based on the multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset.

[0059] S230. Based on the improved long short-term memory network algorithm, an initial power risk prediction model for the target power market area is determined according to multi-source spatiotemporal data.

[0060] S240. Based on the initial power risk prediction model, determine the initial power risk prediction results for the target power market area according to the radiation spatiotemporal dataset.

[0061] S250. Based on the initial power risk prediction results, determine the regional average risk coverage of the target power market area, and extract features from the radiation spatiotemporal dataset when the regional average risk coverage meets the feature extraction conditions, to obtain static spatiotemporal feature sequences and dynamic multi-scale feature sequences.

[0062] In this embodiment, the regional average risk coverage rate is obtained by weighted averaging of all risk prediction results for the target electricity market region, resulting in a figure reflecting the overall risk level of the region. Feature extraction conditions are pre-set based on actual conditions or empirical values; these conditions may include a regional average risk coverage rate below a preset coverage threshold. Static spatiotemporal feature sequences are sequences extracted from the radiation spatiotemporal dataset that describe spatial and temporal characteristics. These features are relatively static and do not change over time; they can include factors such as the geographical location of the electricity market, infrastructure layout, and historical climate data, helping to describe the long-term risk characteristics of the electricity market. Dynamic multi-scale feature sequences refer to feature sequences extracted from the radiation spatiotemporal dataset that dynamically change with time and space; these features can reflect the volatility and instability of the electricity market at different time scales (e.g., hours, days, months) and spatial scales (e.g., different regions); through these dynamic feature sequences, the model can capture short-term risk changes in the electricity market and the specific manifestations of market fluctuations.

[0063] For example, the regional average risk coverage rate can be determined by the following formula:

[0064]

[0065] Where C refers to the regional average risk coverage rate. This refers to the area of ​​the risk zone predicted by the model. This refers to the total area of ​​the risk zone.

[0066] For example, the average risk coverage rate of the region is determined based on the initial risk prediction results. If the average risk coverage rate of the region is lower than the preset coverage rate threshold, feature extraction is performed on the radiation spatiotemporal dataset to obtain static spatiotemporal feature sequences and dynamic multi-scale feature sequences.

[0067] S260. Based on multi-scale fusion components and graph neural networks, the target correction component of the initial power risk prediction result is determined according to the static spatiotemporal feature sequence and the dynamic multi-scale feature sequence.

[0068] In this embodiment, the target correction component is the result of optimization and correction based on the initial power risk prediction results by introducing static spatiotemporal feature sequences and dynamic multi-scale feature sequences; it provides more accurate adjustment parameters for the power risk prediction model, so that the final prediction can better reflect the complex dynamics and potential risks of the power market.

[0069] Optionally, the static spatiotemporal feature sequence and the risk prediction error sequence are imported into the multi-scale fusion component to obtain the static correction component; the dynamic multi-scale feature sequence is input into the graph neural network to obtain the dynamic correction component; and the static correction component and the dynamic correction component are weighted and summed to obtain the target correction component of the initial power risk prediction result.

[0070] In this embodiment, the static correction component is an adjustment factor extracted from the static spatiotemporal feature sequence through a multi-scale fusion component. It represents the correction of the initial power risk prediction result based on static factors (such as geographical location, infrastructure layout, etc.). These static features do not change over time, so their correction effect mainly reflects the long-term stability and fundamental conditions of the power market. The dynamic correction component is an adjustment factor extracted from the dynamic multi-scale feature sequence through a graph neural network. It represents the correction of the initial power risk prediction result based on the changes in the power market at different time and spatial scales. These dynamic features can capture the short-term volatility and instability of the power market. Therefore, the dynamic correction component helps to adjust and optimize the risk prediction model to cope with more complex market changes.

[0071] For example, the static spatiotemporal feature sequence and historical risk error sequence are input into a multi-scale fusion component. The global search capability of the fusion component is used to capture the joint nonlinear effects of risk and variables, fluctuations and radiation, and output a static correction component. The dynamic multi-scale feature sequence (especially the radiation fluctuation and fluctuation standard deviation sequence) is input into a graph neural network, and an adaptive attention mechanism is introduced. Through weight adjustment and perturbation strategies, the risk mutation and spatiotemporal dependence are learned, and a dynamic correction component is output. The static correction component and the dynamic correction component are weighted and summed to obtain the target correction component of the initial power risk prediction result.

[0072] It should be noted that parameters for the multi-scale fusion component can be set in advance, such as the fusion depth (20-50 layers) and learning rate (0.05-0.3). The fusion component can also be pre-trained with static spatiotemporal feature sequences, and key factors can be screened through importance analysis to optimize the global mapping. Similarly, a graph neural network can be pre-built, such as a graph neural network with 128 hidden units, using the Adam (Adaptive Moment Estimation) optimizer (learning rate 0.01), and training the graph neural network with spatiotemporal window data to focus on capturing high-frequency fluctuation patterns.

[0073] S270. Based on the target correction component and the initial power risk prediction results, optimize the initial power risk prediction model to obtain the target power risk prediction model.

[0074] Optionally, the target correction component is superimposed on the initial power risk prediction result to obtain the power risk prediction result to be optimized; the current regional risk coverage of the power risk prediction result to be optimized is compared with the actual regional risk coverage of the target power market area to obtain the risk prediction deviation of the power risk prediction result to be optimized; based on the risk prediction deviation, the initial power risk prediction model is optimized and the initial power risk prediction result is redefined until the risk prediction deviation meets the model generation conditions to obtain the target power risk prediction model.

[0075] In this embodiment, the power risk prediction result to be optimized refers to the power risk prediction output that needs further adjustment and optimization after the target correction component is superimposed on the initial power risk prediction result. This prediction result includes a preliminary assessment after correction, but may still have prediction bias, requiring further optimization steps to improve its accuracy and reliability. The current regional risk coverage rate refers to the risk coverage range predicted for the current power market region in the power risk prediction result to be optimized. This indicator reflects the model's comprehensiveness and coverage of power risks within the current region, helping to identify which regions' risks have not been fully predicted. The actual regional risk coverage rate refers to the range of actual power risks within the target power market region. This indicator measures the distribution of actual market risks and compares it with the prediction results, thereby helping to assess the model's accuracy. Risk prediction bias refers to the difference between the current regional risk coverage rate of the power risk prediction result to be optimized and the actual regional risk coverage rate of the target power market region. This bias reflects the error between the model's prediction results and actual risks, indicating that the model's predictions in some regions are inaccurate or incomplete. The model generation conditions are preset by humans based on actual conditions or experience values, and this application embodiment does not specifically limit them; for example, the model generation conditions may be that the risk prediction deviation is less than a preset deviation threshold.

[0076] In one alternative implementation, if the risk prediction deviation indicates an overestimation problem in a high-risk area, the variable penalty function is dynamically corrected based on the gradient descent method; if the risk prediction deviation indicates a blind spot problem in a low-risk area, the upper limit of the sequence weights in the prediction model is adaptively adjusted according to the standard deviation of volatility.

[0077] In this embodiment, the overestimation problem in high-risk areas refers to the situation in power risk prediction where the model incorrectly overestimates the risk level of certain areas, exceeding the actual risk level. This typically occurs when the model's predicted values ​​for certain areas are higher than the actual risk values, potentially leading to excessive vigilance or intervention in these areas. The gradient descent method for dynamically correcting the penalty function refers to using gradient descent to correct relevant variables in the power risk prediction model when a risk prediction deviation indicates an overestimation problem in high-risk areas. Gradient descent is an optimization method that iteratively minimizes errors; in this context, it is used to adjust the penalty function in the model to reduce overestimation of high-risk areas, ensuring that the model's output is more consistent with the actual situation. The low-risk area blind spot problem refers to the situation in risk prediction where the model may fail to adequately cover the risks of certain specific areas due to their lower risk. This situation may result in some low-risk areas not being fully assessed, thus affecting the accuracy of the risk prediction results and forming a "blind spot." Adaptive adjustment of volatility standard deviation refers to the dynamic adjustment of sequence weights in the prediction model based on the volatility standard deviation of the power market risk sequence. The volatility standard deviation reflects the degree of market volatility. Based on this indicator, the model can adaptively adjust the upper limit of the sequence weights, thereby better coping with market volatility and reducing the blind spot problem in the low-risk area. This adaptive adjustment helps the model maintain efficient and accurate prediction capabilities when facing different market volatility.

[0078] This application embodiment determines a radiation monitoring model for a target electricity market area based on multi-source spatiotemporal data and topographic sequence data. The multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market electricity allocation data. Based on a multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to the multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset. Based on an improved long short-term memory network algorithm, an initial electricity risk prediction model for the target electricity market area is determined according to the multi-source spatiotemporal data. Based on the initial electricity risk prediction model, and according to the radiation spatiotemporal dataset, the target... The initial power risk prediction results for the power market area are obtained. Based on these initial results, the regional average risk coverage rate of the target power market area is determined. If the regional average risk coverage rate meets the feature extraction conditions, features are extracted from the radiation spatiotemporal dataset to obtain static spatiotemporal feature sequences and dynamic multi-scale feature sequences. Based on a multi-scale fusion component and a graph neural network, the target correction component of the initial power risk prediction results is determined according to the static spatiotemporal feature sequences and the dynamic multi-scale feature sequences. Based on the target correction component and the initial power risk prediction results, the initial power risk prediction model is optimized to obtain the target power risk prediction model. This technical solution, by combining a radiation monitoring model, a multi-scale variational assimilation algorithm, and an improved long short-term memory network, determines the required power risk prediction model and uses it for power market-related risk prediction. It considers the impact of the spatiotemporal variability of solar radiation on market risk and balances prediction accuracy and risk quantification efficiency, enabling the model to efficiently and accurately determine the power market-related risk prediction results and meet the needs of refined risk management in a complex energy market environment.

[0079] Example 3

[0080] Figure 3 This is a schematic diagram of a power risk prediction model determination device according to Embodiment 3 of this application. It is applicable to situations where a power risk prediction model based on solar radiation is trained to predict power market-related risks. This power risk prediction model determination device can be implemented in hardware and / or software and can be configured in a computer device, such as a server. Figure 3 As shown, the device includes:

[0081] The monitoring model determination module 310 is used to determine the radiation monitoring model of the target power market area based on multi-source spatiotemporal data and topographic sequence data of the target power market area; wherein, the multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data and market power allocation data;

[0082] The dataset determination module 320 is used to adjust the state variables and model parameters of the radiation monitoring model based on multi-scale variational assimilation algorithm and multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset.

[0083] The prediction module determines module 330, which is used to determine the initial power risk prediction model for the target power market area based on the improved long short-term memory network algorithm and multi-source spatiotemporal data.

[0084] The prediction result determination module 340 is used to determine the initial power risk prediction result of the target power market area based on the initial power risk prediction model and the radiation spatiotemporal dataset.

[0085] The target model determination module 350 is used to optimize the initial power risk prediction model based on the radiation spatiotemporal dataset and the initial power risk prediction results, using a multi-scale fusion component and a graph neural network, to obtain the target power risk prediction model.

[0086] This application embodiment determines a radiation monitoring model for a target electricity market area based on multi-source spatiotemporal data and topographic sequence data. The multi-source spatiotemporal data includes ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market electricity allocation data. Based on a multi-scale variational assimilation algorithm, the state variables and model parameters of the radiation monitoring model are adjusted according to the multi-source spatiotemporal data to obtain a radiation spatiotemporal dataset. Based on an improved long short-term memory network algorithm, an initial electricity risk prediction model for the target electricity market area is determined according to the multi-source spatiotemporal data. Based on the initial electricity risk prediction model, the initial electricity risk prediction result for the target electricity market area is determined according to the radiation spatiotemporal dataset. Based on a multi-scale fusion component and a graph neural network, the initial electricity risk prediction model is optimized according to the radiation spatiotemporal dataset and the initial electricity risk prediction result to obtain the target electricity risk prediction model. The above technical solution determines the required power risk prediction model by combining radiation monitoring models, multi-scale variational assimilation algorithms, and improved long short-term memory networks, and uses this model to predict power market-related risks. It takes into account the impact of the spatiotemporal variability of solar radiation on market risks, and balances prediction accuracy and risk quantification efficiency. This enables the model to efficiently and accurately determine the prediction results of power market-related risks, meeting the needs of refined risk management in complex energy market environments.

[0087] Optionally, the initial power risk prediction results include at least a risk prediction error sequence; correspondingly, the target model determination module 350 includes:

[0088] The feature extraction unit is used to determine the regional average risk coverage of the target power market area based on the initial power risk prediction results, and to extract features from the radiation spatiotemporal dataset when the regional average risk coverage meets the feature extraction conditions, so as to obtain static spatiotemporal feature sequences and dynamic multi-scale feature sequences.

[0089] The component determination unit is used to determine the target correction component of the initial power risk prediction result based on the static spatiotemporal feature sequence and the dynamic multi-scale feature sequence, using a multi-scale fusion component and a graph neural network.

[0090] The model optimization unit is used to optimize the initial power risk prediction model based on the target correction component and the initial power risk prediction results, so as to obtain the target power risk prediction model.

[0091] Optional, model optimization unit, specifically used for:

[0092] The target correction component is superimposed on the initial power risk prediction result to obtain the power risk prediction result to be optimized.

[0093] The current regional risk coverage rate of the power risk prediction results to be optimized is compared with the actual regional risk coverage rate of the target power market area to obtain the risk prediction deviation of the power risk prediction results to be optimized.

[0094] Based on the risk prediction deviation, the initial power risk prediction model is optimized, and the initial power risk prediction results are redefined until the risk prediction deviation meets the model generation conditions, thus obtaining the target power risk prediction model.

[0095] Optional, component determination unit, specifically used for:

[0096] The static spatiotemporal feature sequence and the risk prediction error sequence are imported into the multi-scale fusion component to obtain the static correction component;

[0097] The dynamic multi-scale feature sequence is input into the graph neural network to obtain the dynamic correction component;

[0098] The target correction component of the initial power risk prediction result is obtained by weighted summation of the static correction component and the dynamic correction component.

[0099] Optionally, the ground radiation monitoring data includes at least the hourly radiation index; the satellite remote sensing spatiotemporal sequence data includes at least meteorological variable data; correspondingly, the monitoring model determination module 310 is specifically used for:

[0100] Multidimensional data fusion was performed on ground radiation monitoring data, satellite remote sensing spatiotemporal sequence data, and market power allocation data of the target electricity market area to obtain radiation observation data of the target electricity market area.

[0101] Based on topographic sequence data, hourly radiation index and meteorological variable data of the target electricity market area, determine the regional radiation static parameter field with high spatiotemporal resolution of the target electricity market area;

[0102] Using the regional static radiation parameter field as boundary conditions, a radiation monitoring model for the target electricity market area is determined based on radiation observation data.

[0103] Optionally, the improved long short-term memory network algorithm updates the sequence weights based on an adaptive attention mechanism, uses the minimum market volatility threshold as a constraint, takes maximizing prediction accuracy and minimizing risk bias as the objective function, and uses the difference between the product of the spatiotemporal coverage percentage and the market weight and the uncertainty penalty as the fitness function.

[0104] The power risk prediction model determination device provided in this application embodiment can execute the power risk prediction model determination method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing each power risk prediction model determination method.

[0105] According to embodiments of this application, this application also provides an electronic device, a readable storage medium, and a computer program product.

[0106] Example 4

[0107] Figure 4 This is a schematic diagram of the structure of an electronic device 410 that implements the power risk prediction model determination method of the embodiments of this application. 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 (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0108] like Figure 4As shown, the electronic device 410 includes at least one processor 411 and a memory, such as a read-only memory (ROM) 412 or a random access memory (RAM) 413, communicatively connected to the at least one processor 411. The memory stores computer programs executable by the at least one processor. The processor 411 can perform various appropriate actions and processes based on the computer program stored in the ROM 412 or loaded from storage unit 418 into the RAM 413. The RAM 413 may also store various programs and data required for the operation of the electronic device 410. The processor 411, ROM 412, and RAM 413 are interconnected via a bus 414. An input / output (I / O) interface 415 is also connected to the bus 414.

[0109] Multiple components in electronic device 410 are connected to I / O interface 415, including: input unit 416, such as keyboard, mouse, etc.; output unit 417, such as various types of displays, speakers, etc.; storage unit 418, such as disk, optical disk, etc.; and communication unit 419, such as network card, modem, wireless transceiver, etc. Communication unit 419 allows electronic device 410 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0110] Processor 411 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 411 performs the various methods and processes described above, such as the power risk prediction model determination method.

[0111] In some embodiments, the power risk prediction model determination method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 418. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 410 via ROM 412 and / or communication unit 419. When the computer program is loaded into RAM 413 and executed by processor 411, one or more steps of the power risk prediction model determination method described above may be performed. Alternatively, in other embodiments, processor 411 may be configured as the power risk prediction model determination method by any other suitable means (e.g., by means of firmware).

[0112] 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.

[0113] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable power risk prediction model determination device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0114] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium 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. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. 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 fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0115] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. 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).

[0116] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users 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., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0117] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0118] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0119] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for determining an electric power risk prediction model, characterized in that, The method comprises the following steps: determining a radiation monitoring model of a target power market area according to multi-source spatio-temporal data and terrain sequence data of the target power market area; wherein the multi-source spatio-temporal data comprises ground radiation monitoring data, satellite remote sensing spatio-temporal sequence data and market power configuration data; adjusting state variables and model parameters of the radiation monitoring model according to the multi-source spatio-temporal data based on a multi-scale variational assimilation algorithm to obtain a radiation spatio-temporal data set; determining an initial power risk prediction model of the target power market area according to the multi-source spatio-temporal data based on an improved long short-term memory network algorithm; determining an initial power risk prediction result of the target power market area according to the radiation spatio-temporal data set based on the initial power risk prediction model; wherein the initial power risk prediction result at least comprises a risk prediction error sequence; determining a regional average risk coverage rate of the target power market area according to the initial power risk prediction result, and performing feature extraction on the radiation spatio-temporal data set in the case that the regional average risk coverage rate meets a feature extraction condition to obtain a static spatio-temporal feature sequence and a dynamic multi-scale feature sequence; inputting the static spatio-temporal feature sequence and the risk prediction error sequence into a multi-scale fusion component to obtain a static correction component; inputting the dynamic multi-scale feature sequence into a graph neural network to obtain a dynamic correction component; performing weighted summation on the static correction component and the dynamic correction component to obtain a target correction component of the initial power risk prediction result; optimizing the initial power risk prediction model according to the target correction component and the initial power risk prediction result to obtain a target power risk prediction model.

2. The method of claim 1, wherein, The method of optimizing the initial power risk prediction model according to the target correction component and the initial power risk prediction result to obtain a target power risk prediction model comprises the following steps: superimposing the target correction component on the initial power risk prediction result to obtain a to-be-optimized power risk prediction result; comparing a current regional risk coverage rate of the to-be-optimized power risk prediction result with an actual regional risk coverage rate of the target power market area to obtain a risk prediction deviation of the to-be-optimized power risk prediction result; optimizing the initial power risk prediction model according to the risk prediction deviation and re-determining the initial power risk prediction result until the risk prediction deviation meets a model generation condition to obtain a target power risk prediction model.

3. The method of claim 1, wherein, The ground radiation monitoring data at least comprises a per-hour radiation index; the satellite remote sensing spatio-temporal sequence data at least comprises meteorological variable data; accordingly, the method of determining a radiation monitoring model of a target power market area according to multi-source spatio-temporal data and terrain sequence data of the target power market area comprises the following steps: performing multi-dimensional data fusion on ground radiation monitoring data, satellite remote sensing spatio-temporal sequence data and market power configuration data of the target power market area to obtain radiation observation data of the target power market area; determine a high-spatial-temporal resolution regional radiation static parameter field of the target power market region according to terrain sequence data, hourly radiation index and meteorological variable data of the target power market region; determine a radiation monitoring model of the target power market region according to the radiation observation data with the regional radiation static parameter field as a boundary condition.

4. The method of claim 1, wherein, The improved long short-term memory network algorithm updates sequence weights based on an adaptive attention mechanism, takes a minimum market fluctuation threshold as a constraint condition, takes maximizing prediction accuracy and minimizing risk deviation as objective functions, and takes a difference between a product value of a spatial-temporal coverage percentage and a market weight and an uncertainty penalty term as a fitness function.

5. An electric power risk prediction model determination apparatus characterized by comprising: The method comprises the following steps: a monitoring model determination module is configured to determine a radiation monitoring model of a target power market region according to multi-source spatial-temporal data and terrain sequence data of the target power market region; wherein the multi-source spatial-temporal data comprises ground radiation monitoring data, satellite remote sensing spatial-temporal sequence data and market power configuration data; a data set determination module is configured to adjust state variables and model parameters of the radiation monitoring model according to the multi-source spatial-temporal data to obtain a radiation spatial-temporal data set based on a multi-scale variational assimilation algorithm; a prediction module determination module is configured to determine an initial power risk prediction model of the target power market region according to the multi-source spatial-temporal data based on an improved long short-term memory network algorithm; a prediction result determination module is configured to determine an initial power risk prediction result of the target power market region according to the radiation spatial-temporal data set based on the initial power risk prediction model; wherein the initial power risk prediction result at least comprises a risk prediction error sequence; a target model determination module is configured to optimize the initial power risk prediction model according to the radiation spatial-temporal data set and the initial power risk prediction result to obtain a target power risk prediction model based on a multi-scale fusion component and a graph neural network; wherein the target model determination module comprises: a feature extraction unit is configured to determine a regional average risk coverage rate of the target power market region according to the initial power risk prediction result, and perform feature extraction on the radiation spatial-temporal data set to obtain a static spatial-temporal feature sequence and a dynamic multi-scale feature sequence in a case where the regional average risk coverage rate meets a feature extraction condition; a component determination unit is configured to determine a target correction component of the initial power risk prediction result according to the static spatial-temporal feature sequence and the dynamic multi-scale feature sequence based on a multi-scale fusion component and a graph neural network; a model optimization unit is configured to optimize the initial power risk prediction model according to the target correction component and the initial power risk prediction result to obtain a target power risk prediction model; wherein the component determination unit is specifically configured to: input the static spatial-temporal feature sequence and the risk prediction error sequence into a multi-scale fusion component to obtain a static correction component; input the dynamic multi-scale feature sequence into a graph neural network to obtain a dynamic correction component; weighting and summing the static correction component and the dynamic correction component to obtain a target correction component of an initial power risk prediction result.

6. An electronic device, comprising: comprising: one or more processors; memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the power risk prediction model determination method according to any one of claims 1-4.

7. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the power risk prediction model determination method according to any one of claims 1-4.

8. A computer program product comprising a computer program which, when executed by a processor, implements the power risk prediction model determination method according to any one of claims 1-4.

Images