Wind-solar power generation power probability analysis method, system and device and storage medium

Abstract

The invention discloses a wind-solar power generation power probability analysis method, system and device and a storage medium. The method comprises the following steps: preprocessing multi-source heterogeneous data of a target wind-solar power station to construct an input variable space; based on various input combinations in the input variable space, generating corresponding virtual power prediction parameters and probability distribution of power output of the virtual power prediction parameters, and forming an initial virtual power prediction parameter set; acquiring actual observation power data, and performing difference measurement on the prediction power corresponding to each group of virtual power prediction parameters and the actual observation power data; calculating the credibility weight of each group of virtual power prediction parameters based on the difference measurement result, and screening based on the credibility weight to obtain a high-weight virtual power extension parameter subset; and judging whether the current predicted power meets a preset critical condition based on the high-weight virtual power extension parameter subset, and outputting a corresponding probabilistic key index when the current predicted power meets the preset critical condition. And high-precision and interpretable probabilistic analysis of the wind-solar power generation power is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system dispatching management, and particularly relates to a wind-solar power probability analysis method, system, device and storage medium. BACKGROUND

[0002] In wind-solar power grid-connected operation and power system dispatching management, accurate prediction and reliability evaluation of wind-solar power are the core support for guaranteeing the safety and stability of the power grid and optimizing the allocation of energy resources. However, due to the inherent intermittency, volatility and strong uncertainty characteristics of wind-solar power, traditional deterministic prediction methods, such as numerical weather prediction models based on physical equations and single neural network regression models, can only output a single point estimate power value, and cannot effectively depict the power output probability distribution characteristics under the coupling action of complex factors such as sudden changes in meteorological conditions, equipment state degradation and dynamic adjustment of grid constraints, resulting in significant deviation between the prediction results and the actual operating state. Wind-solar power is easily affected by the nonlinear coupling of multiple factors such as weather, equipment, geography and power grid, but existing methods rely on a single type of data and cannot fully capture the potential power generation scenarios under the interaction of various factors, resulting in insufficient adaptability of the prediction model to complex operating environments. In actual operation, errors in weather forecasting, dynamic changes in equipment state over time and real-time adjustment of grid constraints will all cause the power output to exhibit probability distribution characteristics under the same input conditions. Traditional point estimation models cannot output the probability distribution function of power, cannot quantify the confidence interval and risk level of the prediction results, and cannot provide probabilistic support for dispatching decisions.

[0003] Wind-solar power systems must strictly meet boundary conditions such as power upper and lower limits and fluctuation margins during grid-connected operation. If the predicted power is close to or exceeds these critical values, risks need to be identified in advance and warnings need to be triggered. However, traditional methods lack the ability to track the dynamic evolution process of power within multiple time windows, making it difficult to dynamically determine whether the system is about to enter a high-risk operating state, resulting in a lag in risk response. When generating virtual power prediction parameters, the input variable space may contain millions of potential scenarios, but only a small number of combinations are highly matched with the actual operating state, and most are low-confidence interference terms. Traditional methods do not establish an effective parameter weight evaluation and screening mechanism, resulting in the need to process a large amount of redundant data in subsequent analysis, reducing prediction efficiency and accuracy, and severely restricting the practical application value of wind-solar power prediction results in power dispatching decisions, equipment operation and maintenance management, market trading strategies and system safety operation. Therefore, there is an urgent need for a probabilistic analysis method that can integrate multi-source heterogeneous data, quantify the probability distribution of power output, dynamically identify critical risks and output high-confidence prediction results to support high penetration and safe and reliable operation of wind-solar power in new power systems. SUMMARY

[0004] In view of the above existing problems, the present application is proposed. Therefore, the present application provides a wind and light power generation power probability analysis method, system, device and storage medium to solve the problems mentioned in the background art.

[0005] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application embodiment provides a wind and light power generation power probability analysis method, comprising: Obtaining multi-source heterogeneous data of a target wind and light power station, preprocessing the multi-source heterogeneous data to construct an input variable space; Based on multiple input combinations in the input variable space, generating corresponding virtual power prediction parameters and their power output probability distribution to form an initial virtual power prediction parameter set; Obtaining actual observed power data within a time window corresponding to the initial virtual power prediction parameter set, and measuring the difference between the predicted power corresponding to each group of virtual power prediction parameters and the actual observed power data; Based on the difference measurement result, calculating the credibility weight of each group of virtual power prediction parameters, and based on the credibility weight, filtering to obtain a high-weight virtual power expansion parameter subset; Based on the high-weight virtual power expansion parameter subset, determining whether the current predicted power meets the preset critical condition, and outputting the corresponding probabilistic key indicators when it meets.

[0006] As a preferred scheme of the wind and light power generation power probability analysis method of the present application, wherein: based on the high-weight virtual power expansion parameter subset, determining whether the current predicted power meets the preset critical condition, and outputting the corresponding probabilistic key indicators when it meets, comprising: If the current predicted power does not meet any critical condition, generate an expanded virtual power parameter set for the next prediction time window based on the high-weight parameter subset, and re-calculate and filter the weight; If the current predicted power meets any critical condition or reaches the end of the preset prediction period, terminate the iteration; Based on the high-weight virtual power expansion parameter subset at the time of terminating iteration, output the probabilistic key indicators.

[0007] As a preferred scheme of the wind and light power generation power probability analysis method of the present application, wherein: further comprising: The preset critical condition at least includes any one of the risk operating states that the power reaches the upper limit of the grid capacity, is lower than the lower limit of market transaction, the power fluctuation exceeds the system stability margin, or the energy storage is exhausted, frequent power limiting and insufficient backup; The probabilistic key indicators at least include a probability of reaching a critical power or a risk state, a confidence interval of predicted power, a probability of the system being in a high / low risk operating state, a probability of stable power operation in a future predetermined time, and an expected power.

[0008] As a preferred scheme of the wind-solar power generation power probability analysis method, the high-weight virtual power expansion parameter subset obtained based on the credibility weight screening includes: All virtual parameters in the initial virtual power prediction parameter set are sorted according to their corresponding credibility weight values to form an ordered parameter list; According to a preset screening strategy, high-weight parameters are selected from the ordered parameter list to construct a high-credibility virtual power expansion parameter subset, and the preset screening strategy includes any one or a combination of the following: Select the virtual parameters ranked in the top K positions by weight, wherein K is a preset positive integer; Select the first several virtual parameters whose cumulative weight proportion reaches a preset proportion threshold; Keep virtual parameters with a credibility weight not less than a preset weight threshold.

[0009] The beneficial effects of the preferred technical scheme are that the screening mechanism can effectively eliminate low-credibility parameter combinations with low weight and serious deviation from actual observation, avoid misleading of interference terms on critical judgment and probability output, and greatly reduce the calculation complexity and invalid information load, thereby improving the accuracy, efficiency and interpretability of the overall prediction process.

[0010] As a preferred scheme of the wind-solar power generation power probability analysis method, the calculation of the credibility weight of each group of virtual power prediction parameters based on the difference measurement result includes: For each group of virtual parameters in the initial virtual power prediction parameter set, a difference measurement value is calculated according to the predicted power corresponding to each group of virtual parameters and the actual observed power in the same prediction time window, the difference measurement value is converted into a positive credibility indicator, the error inverse of each group of virtual parameters is normalized to obtain the credibility weight of each group of virtual parameters.

[0011] As a preferred scheme of the wind-solar power generation power probability analysis method, the generation of corresponding virtual power prediction parameters and their power output probability distribution based on multiple input combinations in the input variable space to form an initial virtual power prediction parameter set includes: Generating N groups of input combinations based on the input variable space Each group of input combinations Corresponding to a wind-solar power generation operating scenario; For each group of input combinations generate corresponding virtual power prediction parameters ; integrate all N groups of virtual power prediction parameters into an initial virtual power prediction parameter set .

[0012] The beneficial effects of the preferred technical solution are that the construction of the initial virtual power prediction parameter set breaks through the limitations of traditional single prediction models, fully considers potential power generation behaviors under different hypothetical conditions, greatly enriches the diversity of prediction models, enables the model to more comprehensively reflect the complexity and uncertainty of the wind-solar power generation system, and provides a rich and reliable input basis for subsequent evaluation and screening.

[0013] As a preferred scheme of the wind-solar power generation power probability analysis method described in the application, the input variables include meteorological variables, equipment state variables, geographical and microclimate variables, and grid constraints and operating condition variables.

[0014] In a second aspect, the application provides a wind-solar power generation power probability analysis system, comprising: a data acquisition module for acquiring multi-source heterogeneous data of a target wind-solar power station, and preprocessing the multi-source heterogeneous data to construct an input variable space; a power prediction module for generating corresponding virtual power prediction parameters and their power output probability distribution based on multiple input combinations in the input variable space, forming an initial virtual power prediction parameter set; a difference analysis module for acquiring actual observed power data within a time window corresponding to the initial virtual power prediction parameter set, and performing difference measurement on the predicted power corresponding to each group of virtual power prediction parameters and the actual observed power data; a weight calculation module for calculating the credibility weight of each group of virtual power prediction parameters based on the difference measurement result, and filtering a high-weight virtual power expansion parameter subset based on the credibility weight; a probabilistic index output module for judging whether the current predicted power meets a preset critical condition based on the high-weight virtual power expansion parameter subset, and outputting corresponding probabilistic key indicators when the condition is met.

[0015] In a third aspect, the application provides an electronic device, comprising: a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, which implement the steps of the wind-solar power generation power probability analysis method when executed by the processor.

[0016] In a fourth aspect, the present application provides a computer readable storage medium storing computer executable instructions, which, when executed by a processor, implement the steps of the wind-solar power generation power probability analysis method.

[0017] Compared with the prior art, the present application has the following beneficial effects: the present application collects high-frequency and high-quality input data covering multiple dimensions such as weather, equipment, geography, power grid and historical operation, and goes through a pretreatment process, which can effectively ensure the consistency and availability of the input data in the time scale, physical meaning and numerical range, and lays a solid data foundation for subsequent accurate analysis and prediction, can fully tap the information contained in the multi-source data, and avoid analysis deviation caused by data quality problems.

[0018] By generating multiple sets of virtual power prediction parameters and their corresponding power output results, an initial virtual power prediction parameter set is constructed, which breaks through the limitations of traditional single prediction models, fully considers the potential power generation behavior under different hypothetical conditions, enriches the diversity of the prediction model, and makes the model more fully reflect the complexity and uncertainty of the wind-solar power generation system, providing a rich and reliable input basis for subsequent evaluation and screening.

[0019] Based on the difference index comparison between the real observation power data and the virtual power prediction parameters, the corresponding weight value of each set of virtual power prediction parameters is calculated, which accurately reflects the credibility and matching degree of the power generation mode represented by each set of parameters in the current actual operation environment. Then, by screening the high-weight virtual power expansion parameters, the low-credibility and actual-operation-deviated interference terms can be effectively eliminated, the pertinence and reliability of the prediction results are improved, unnecessary calculation amount is reduced, and the overall prediction efficiency is improved.

[0020] The present application can dynamically adapt to the changes of the operation state of the wind-solar power generation system, timely capture the critical state and risk situation that the system may appear, provide more accurate short-term power prediction, evaluate the long-term stability and reliability of the system, help the dispatch personnel more reasonably arrange the power generation plan and power grid operation mode, help to find the possible fault risk of the equipment in advance, develop more effective maintenance strategies, provide more accurate power prediction information for market participants in market transaction strategy development, reduce transaction risk, and timely discover and warn potential system risks in system safe operation, and ensure the stable and reliable operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic flowchart of a wind and solar power generation probability analysis method according to an embodiment of the present invention; Figure 2 This is a timing diagram of the prediction system for a wind and solar power generation probability analysis method according to an embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0023] Example 1, referring to Figures 1-2 As one embodiment of the present invention, this embodiment provides a method for probabilistic analysis of wind and solar power generation, such as... Figure 1 As shown, it includes: S100: Acquire multi-source heterogeneous data of the target wind and solar power station, and preprocess the multi-source heterogeneous data to construct the input variable space; S200: Based on multiple input combinations in the input variable space, generate corresponding virtual power prediction parameters and their power output probability distributions to form an initial set of virtual power prediction parameters; S300: Obtain the actual observed power data within the time window corresponding to the initial virtual power prediction parameter set, and measure the difference between the predicted power corresponding to each set of virtual power prediction parameters and the actual observed power data; S400: Calculate the confidence weight of each group of virtual power prediction parameters based on the difference measurement results, and select a subset of virtual power expansion parameters with high weights based on the confidence weights; S500: Based on a high-weight subset of virtual power extension parameters, determine whether the current predicted power meets the preset critical conditions, and output the corresponding probabilistic key indicators when they are met.

[0024] It should be noted that traditional methods often rely on single-type data and fail to effectively integrate multi-dimensional heterogeneous data. This makes it difficult for models to fully capture the nonlinear coupling effects between various factors, resulting in insufficient adaptability to complex operating environments. Furthermore, traditional methods only output a single point estimate of power, failing to characterize the probability distribution of power output and providing probabilistic support for scheduling decisions, thus unable to quantify risk levels. The lack of tracking capability for the dynamic evolution of power within multiple time windows leads to untimely risk response, affecting the safe and stable operation of the power grid. Traditional methods also lack effective parameter weight evaluation and screening mechanisms, resulting in redundant data in subsequent analysis and processing, reducing prediction efficiency and accuracy. This invention, by collecting high-frequency, high-quality input data covering multiple dimensions such as meteorology, equipment, geography, power grid, and historical operation, and through preprocessing, effectively ensures the consistency and usability of input data in terms of time scale, physical meaning, and numerical range. This lays a solid data foundation for subsequent accurate analysis and prediction, fully mining the information contained in multi-source data and avoiding analytical biases caused by data quality issues.

[0025] By generating multiple sets of virtual power prediction parameters and their corresponding power output results, an initial set of virtual power prediction parameters is constructed, which breaks through the limitations of traditional single prediction models. By fully considering the potential power generation behavior under different assumptions, the diversity of prediction models can be enriched, enabling the models to more comprehensively reflect the complexity and uncertainty of wind and solar power generation systems, and providing a rich and reliable input basis for subsequent evaluation and screening.

[0026] Based on a comparison of the differences between actual observed power data and virtual power prediction parameters, a corresponding weight value is calculated for each group of virtual power prediction parameters. This weight accurately reflects the credibility and matching degree of the power generation mode represented by each group of parameters under the current actual operating environment. Subsequently, by selecting virtual power extension parameters with high weights, low-credibility and interference items that deviate from actual operation can be effectively eliminated, improving the relevance and reliability of the prediction results, reducing unnecessary computation, and improving the overall prediction efficiency.

[0027] This invention can dynamically adapt to changes in the operating status of wind and solar power generation systems, promptly capture potential critical states and risks, provide more accurate short-term power forecasts, and assess the long-term stability and reliability of the system. It helps dispatchers to more rationally arrange power generation plans and grid operation modes. In equipment operation and maintenance management, it helps to identify potential equipment failure risks in advance and formulate more effective maintenance strategies. In market trading strategy formulation, it can provide market participants with more accurate power forecast information and reduce trading risks. In terms of system safety operation, it can promptly detect and warn of potential system risks, ensuring the stable and reliable operation of the power system.

[0028] ReferenceFigure 2 It should be noted that in step S100 of the present invention, a multi-source heterogeneous data acquisition system deployed in a wind farm or photovoltaic power station can be used to simultaneously acquire high-frequency, high-quality input data covering multiple dimensions such as meteorology, equipment, geography, power grid and historical operation, including but not limited to: meteorological condition data, equipment status data, terrain and microclimate data, and power grid and operating condition data.

[0029] Specifically, the multi-source heterogeneous data sources cover the core external conditions and internal state variables affecting wind and solar power output, including: Meteorological condition data primarily reflects the direct impact of the natural environment on power generation potential, including but not limited to measured or forecasted wind speed, solar irradiance, wind direction, ambient temperature, and relative humidity. Meteorological condition data generally originates from weather stations, satellite remote sensing, numerical weather prediction model outputs, or field-installed sensor arrays.

[0030] Equipment status data reflects the operating status and performance level of power generation equipment, including but not limited to real-time active power output of wind turbines or photovoltaic modules, DC / AC side voltage and current, conversion efficiency, equipment health status indicators, historical fault records, maintenance operation logs, start-up and shutdown status, and equipment operating time. Equipment status data is typically collected in real time by SCADA systems, PLC controllers, or smart sensors.

[0031] Geographic and microclimate data are used to describe the topographic features of the power station area and their moderating effect on local meteorological conditions, including but not limited to topographic elevation, slope angle, surface orientation, surface reflectance, vegetation index, local shadow distribution, and airflow disturbances caused by altitude differences. Geographic and microclimate data are usually derived from digital elevation models, geographic information systems, remote sensing image interpretation, or field surveys, and are used to characterize microscale effects such as uneven radiation and wind speed distortion caused by geographical factors.

[0032] Grid and operational data are used to reflect the constraints and guidance of the power system on generation behavior, including but not limited to real-time grid load levels, system frequency, node voltage amplitude and phase angle, power curtailment instructions issued by the higher-level dispatch center, energy storage system charging and discharging status and capacity, market electricity price signals, and day-ahead / intra-day generation dispatch plans. Grid and operational data are provided by grid dispatch automation systems, market trading systems, or local energy management systems.

[0033] In an optional embodiment, the preprocessing of multi-source heterogeneous data in step S100 of the present invention includes: performing preprocessing procedures such as time alignment, intelligent missing value imputation, anomaly detection and removal, and standardization / normalization on the collected multi-source heterogeneous data to ensure the consistency and usability of the input data in terms of time scale, physical meaning, and numerical range.

[0034] In this embodiment of the invention, the input variables in step S100 include: meteorological variables, equipment status variables, geographical and microclimate variables, and power grid constraints and operating condition variables.

[0035] In an optional embodiment, a probabilistic modeling method is used to simulate the diverse operating states and power output behaviors that a wind and solar power generation system may exhibit under different combinations of input conditions. This probabilistic modeling method avoids relying on a single deterministic mapping relationship. Instead, it constructs a prediction model with probabilistic distribution characteristics, enumerating and generating a large number of potential input scenarios and their corresponding virtual power responses. This forms a multi-scale, multi-scenario, and multi-probability initial set of virtual power prediction parameters, providing comprehensive input support for subsequent prediction evaluation and parameter selection.

[0036] Specifically, key influencing factors are extracted from preprocessed multi-source data to form a set of input variables X for modeling. Typical components of this set include meteorological variables, equipment status variables, geographical and microclimate variables, and power grid constraints and operating condition variables. Each variable may exist in multiple states or have a continuous distribution within its reasonable value range; therefore, it is necessary to construct an input variable space that covers various typical and extreme cases.

[0037] Based on meteorological conditions, typical input variables include wind speed and solar irradiance. For example, a wind turbine with a rated wind speed of 12 m / s may have a power output close to zero at a wind speed of 3 m / s (cut-in wind speed), while reaching its rated power (e.g., 2 MW) at a wind speed of around 12 m / s. When the wind speed exceeds 25 m / s (cut-out wind speed), the turbine may shut down for protection. Similarly, for photovoltaic systems, the power output is zero when the irradiance is 0 W / m² (e.g., at night). Under standard test conditions, when the irradiance is approximately 1000 W / m², the photovoltaic modules may output their nominal power (e.g., 400 W / module). However, when the irradiance is only 200 W / m² (e.g., in cloudy weather), the actual output may only be about 20% of the nominal value (approximately 80 W / module). Therefore, wind speed and irradiance are not only continuous variables, but also have multiple typical states within their range, such as wind speed (usually 0–30 m / s) and irradiance (0–1200 W / m²), such as low wind speed (0–5 m / s), medium wind speed (5–12 m / s), high wind speed (12–20 m / s), and extreme states, such as strong winds, weak light, and blizzards. These need to be fully considered in the input variable space.

[0038] Based on the equipment status dimension, key input variables include current power output, equipment efficiency, and health index or fault status indicators, such as 0 indicating healthy, 1 indicating slight derating, and 2 indicating outage due to fault. For example, a wind turbine in a wind farm may have an actual power output of 1.2MW and an efficiency of 85% at the current moment, indicating good health (health index of 0). However, at another moment, due to blade icing or bearing wear, the turbine's efficiency may drop to 70%, the health index may become 1, and the power output may decrease to 0.9MW. These variables not only affect the current power output but also reflect the equipment's operational reliability and potential derating risk over a future period, making them dynamic factors that cannot be ignored in the modeling process.

[0039] Based on geographical and microclimate dimensions, input variables can include topographic slope, surface orientation, surface reflectance, and vegetation index. For example, for a photovoltaic power station located on a south-facing slope (approximately 30° slope, facing due south), its photovoltaic modules can receive longer and higher intensity direct radiation at noon. Compared to flat land or a north-facing slope (0° slope, facing north), its average daily power generation per unit area may be 10% to 20% higher. Furthermore, if the area has high vegetation cover (NDVI close to 0.8), it may mean more local shading, further affecting the actual uniformity of light received by the modules. Although these geographical features are relatively static, their coupling effects with meteorological conditions, such as slope and wind speed acceleration, and orientation and irradiance angle, can significantly alter power generation behavior. Therefore, they should also be included in the input variable space to characterize typical and extreme power generation scenarios under different geographical environments.

[0040] From the perspective of grid constraints and operating conditions, key input variables include grid power curtailment orders, system frequency, node voltage levels, and market dispatch plans or energy storage status. For example, if a wind farm receives a grid dispatch order during a certain forecast period, requiring its output to not exceed 1.5MW (curtailment power), even if the current wind speed is as high as 15m / s (corresponding to a potential power of 2.5MW), the actual output will still be limited to within 1.5MW. Similarly, if the current system frequency is low (e.g., below 49.8Hz), the grid may require renewable energy power plants to proactively reduce their output to maintain system balance. These variables reflect the hard constraints and dynamic adjustment requirements of the external power system on power generation behavior, and are key external conditions affecting power output boundaries and operating modes.

[0041] By extracting and combining key influencing factors such as meteorological variables (e.g., wind speed 3–25 m / s, irradiance 0–1200 W / m²), equipment status variables (e.g., power 0–2 MW, efficiency 70%–95%, health index 0–2), geographical variables (e.g., slope 0°–45°, orientation 0°–360°), and grid constraint variables (e.g., power curtailment 0–3 MW, dispatch plan instructions), an input variable space is constructed that covers typical operating states (e.g., full power generation at rated wind speed, optimal output at standard irradiance) and extreme operating scenarios (e.g., strong wind shedding, low light curtailment, equipment failure, emergency grid curtailment). Each specific input combination in this space represents a possible operating scenario, providing a comprehensive and detailed scenario coverage foundation for subsequent probabilistic modeling and virtual parameter generation, ensuring that the model can fully learn and simulate the actual response characteristics and potential output boundaries of the wind and solar power generation system under different conditions. For example, wind speed 11m / s, irradiance 800W / m², power 1.8MW, efficiency 88%, health index 0, slope 30°, and power limitation 2MW.

[0042] In the input variable space, a series of representative input scenarios are enumerated by setting different combinations of variables (which can be combinations of actual observations or expected combinations generated based on historical statistics or prediction models). These scenarios include both normal operating conditions (such as stable power generation at rated wind speed) and edge or extreme situations (such as low wind speed / weak radiation, equipment derating, grid curtailment constraints, etc.), thereby ensuring the diversity and coverage of the virtual parameter set.

[0043] For each specific combination of input variable space Using probabilistic modeling methods, the virtual power prediction parameters and their corresponding power output results of the wind and solar power generation system under this scenario are inferred or simulated. The virtual power prediction parameters include not only the final power value, but also other key parameters related to power generation behavior, such as the expected value, variance, probability density distribution of the predicted power, and the contribution of key factors affecting power output.

[0044] Furthermore, the specific modeling steps include: For each input combination The probabilistic irradiance-power mapping model simulates the power generation process under a given scenario based on built-in mapping rules or probability distribution parameters obtained through training, and outputs a probability distribution function of power.

[0045] The probabilistic irradiance-power mapping model simulates the probability distribution of power output based on two main mechanisms: built-in mapping rules and probability distribution parameters trained through a data-driven approach. These two mechanisms can be used individually or in combination, but their core objective is the same: to establish a probabilistic correlation model between input variables and output power, enabling the model to reasonably infer and simulate various possible power output values ​​and their corresponding probabilities given input conditions.

[0046] In the first mechanism, the built-in mapping rules are typically based on a deep understanding and simplified abstraction of the physical processes of wind and solar power generation, combined with probabilistic modeling methods to construct a nonlinear mapping logic between inputs and outputs. For example, the model may include built-in IV characteristic curves of photovoltaic modules, power-irradiance-temperature response relationships, wind turbine power curves, and correction modules that consider environmental losses such as shading, dust, temperature correction, and equipment operating conditions such as efficiency and health indices. These physical rules provide the model with a basic power generation logic framework. On this basis, the model further introduces probabilistic distribution descriptions. For example, it is assumed that under given irradiance and temperature conditions, the actual output of the modules will fluctuate randomly around the theoretical value, such as due to rapid cloud movement, measurement errors, or minor performance drift of the equipment. The built-in probability distributions, such as normal distribution, truncated distribution, or mixed distribution, are used to characterize the range of this fluctuation and its probability of occurrence, thereby simulating the uncertainty of power output.

[0047] In the second mechanism, the model employs data-driven probabilistic learning methods, utilizing probabilistic machine learning algorithms such as Gaussian process regression, Bayesian neural networks, probabilistic random forests, variational inference models, or conditional kernel density estimation based on historical operating data to jointly model the relationship between input variables and output power. During this process, the model stores the conditional probability distribution of the output power given a specific set of input conditions as a set of probability distribution parameters within the model, such as the mean and variance of the Gaussian distribution, the weights and distribution parameters of each component in the mixture model, and the bandwidth and kernel function of the kernel density estimation.

[0048] When the model receives a specific input combination during runtime, such as irradiance of 800 W / m², ambient temperature of 25°C, good equipment health, and a power limitation of 2 MW, it automatically triggers a simulation of the power generation process under that input scenario using built-in mapping rules or trained probability distribution parameters. Based on the probability distribution logic learned or defined within the model, it generates the possible range of power output values ​​and their corresponding probability densities under that input condition, ultimately outputting a probability density function describing the power distribution under that scenario. The probability density function intuitively reflects the range where power is most likely to concentrate, the probability of low or high power output, and the overall power output fluctuation characteristics and uncertainty level under that input combination.

[0049] For example, given the input combination: irradiance = 800 W / m², temperature = 25°C, equipment in normal condition, and power limitation = 2 MW, the model might output a probability density function centered at 1.8 MW with an approximately normal shape. The peak value corresponds to the highest probability, indicating that this power value is most likely to occur. Simultaneously, the probability density function curve extends to both sides, reflecting that the power may also fall between 1.6 MW and 2.0 MW, but the probability of these values ​​gradually decreases. In some more complex model settings, if there is significant uncertainty in the input conditions, such as large weather forecast errors or unknown degradation of equipment status, this probability density function may exhibit a multimodal or skewed distribution, more realistically reflecting the multiple operating modes and risk scenarios that the actual system may face.

[0050] It should be noted that the probabilistic irradiance-power mapping model realizes the probabilistic simulation and distribution characterization of the power generation process under each input scenario. Its output probability density function not only includes the expected value of power, but also fully retains all possible values ​​of power output under that scenario and their corresponding probability information. This provides a rich, detailed and physically meaningful statistical basis for subsequent construction of virtual power prediction parameter sets, calculation of difference indicators, evaluation of parameter credibility, and probabilistic risk assessment and decision-making.

[0051] Combine each input Its corresponding analog output, such as power value Predicted distribution parameters, key influencing factors, etc., are packaged into a virtual power prediction parameter unit, denoted as V_i={x_i,P_i,D_i}, where It can represent additional distribution information or metadata in this context.

[0052] By traversing or sampling a large number of combinations in the input variable space, a set containing N sets of virtual parameters is finally generated. This set constitutes the initial set of virtual power prediction parameters constructed in this step. This set covers a variety of possible scenarios, from normal operation to extreme boundary conditions, with each set of parameters representing a potential power generation state. This provides a rich and representative input basis for subsequent prediction assessment, weight calculation, and parameter selection.

[0053] In this embodiment of the invention, step S200, based on multiple input combinations in the input variable space, generates corresponding virtual power prediction parameters and their probability distributions for power output, forming an initial set of virtual power prediction parameters, including: Generate N sets of input combinations based on the input variable space. Each input combination This corresponds to a wind and solar power generation operation scenario; For each input combination Generate corresponding virtual power prediction parameters ; Integrate all N sets of virtual power prediction parameters into an initial set of virtual power prediction parameters. .

[0054] Specifically, in step 300 of this embodiment of the invention, after obtaining the actual observed power data for the same prediction time window, the actual observed power data is compared with the difference index of each set of virtual power prediction parameters generated in step S100. The measurement index used may include mean square error, mean absolute error, cosine similarity, probability distribution distance, etc., to accurately characterize the degree of agreement between each set of virtual prediction results and the actual situation.

[0055] In this embodiment of the invention, step S400, which calculates the confidence weight of each group of virtual power prediction parameters based on the difference measurement results, includes: For each set of virtual parameters in the initial set of virtual power prediction parameters, a difference metric is calculated based on the predicted power corresponding to each set of virtual parameters and the actual observed power within the same prediction time window. The difference metric is then converted into a positive confidence index. The inverse error of each set of virtual parameters is normalized to obtain the confidence weight of each set of virtual parameters.

[0056] In this embodiment of the invention, step S400, which involves filtering based on confidence weights to obtain a subset of virtual power expansion parameters with high weights, includes: All virtual parameters in the initial virtual power prediction parameter set are sorted according to their corresponding confidence weight values ​​to form an ordered parameter list; According to a preset screening strategy, high-weight parameters are selected from an ordered parameter list to construct a subset of highly reliable virtual power expansion parameters. The preset screening strategy includes any one or more combinations of the following: Select the top K virtual parameters by weight, where K is a preset positive integer; Select the first few virtual parameters whose cumulative weight percentage reaches the preset threshold. Retain all virtual parameters whose credibility weight is not less than the preset weight threshold.

[0057] In an optional embodiment, based on the comparison results of the difference index, a corresponding weight value is calculated for each group of virtual power prediction parameters. This weight reflects the credibility and matching degree of the power generation mode represented by the group of parameters under the current actual operating environment—the higher the weight, the closer the prediction result of the group of virtual power prediction parameters is to the actual power, and the more representative and reliable the parameter combination it represents. According to the weight calculation results, the Top-K selection method is used to select one or more groups of high-weight virtual power extension parameters from all virtual power prediction parameter combinations. The retained parameters mainly reflect the key factors and system behavior patterns that are most likely to affect the power generation output under the current conditions, providing a focused and reliable input basis for subsequent critical state judgment and iterative prediction. At the same time, it effectively eliminates interference items with low credibility and deviation from actual operation, improving the overall prediction accuracy and efficiency.

[0058] Furthermore, within the probabilistic analysis framework for wind and solar power generation, step S200, after acquiring the actual observed power data within a certain prediction time window, such as 15 minutes, 30 minutes, or 60 minutes from the current moment, systematically compares the predicted power corresponding to each set of virtual power prediction parameters generated in step S100 with the predicted power of the virtual power prediction parameter sets. This quantitatively assesses the degree of consistency between the power generation mode represented by each set of virtual parameters and the actual operating state, and calculates a credibility weight for each set of virtual parameters accordingly. Finally, a screening mechanism retains parameter combinations with high weights and high representativeness, while eliminating interference items with low matching degree and deviation from reality. This constructs a subset of highly reliable virtual parameters that focuses on key influencing factors and has high prediction consistency, providing an accurate and efficient input basis for subsequent critical state judgment and iterative prediction processes.

[0059] In an optional embodiment, the power observation of the actual system output within a specific prediction time window is used. As a benchmark reference, this data typically comes from real-time acquisition by the power plant's SCADA system, power measurement devices, or grid dispatching platform, and has high reliability and time synchronization. Correspondingly, step S100 has already constructed a set containing N sets of virtual power prediction parameters. Each set of parameters This includes a set of input conditions (such as weather, equipment, and power grid conditions) and their corresponding predicted power output. Here It can be a point estimate or it can include probability distribution information, but in this step, we mainly focus on its predicted power value or central trend for direct comparison with the true value.

[0060] Furthermore, in order to quantitatively characterize the degree of difference between each set of virtual predicted parameters and actual observations, difference metrics with clear physical meaning and statistical properties are introduced. Commonly used metrics include, but are not limited to, the following categories: Mean square error: Calculates the squared mean of the difference between predicted power and actual power. It is sensitive to large deviations and is suitable for scenarios that emphasize the impact of extreme errors.

[0061] Mean Absolute Error: Calculates the average absolute value of the difference between the prediction and the actual value. It is more robust to outliers and reflects the overall level of deviation.

[0062] Cosine similarity: Treating predicted power and actual power as vectors, it calculates the degree of similarity between the two in their directions. It is suitable for situations where the consistency of prediction trends is of concern.

[0063] Probability distribution distance index: When the virtual prediction parameters are accompanied by a probability distribution, K-divergence and other metrics can be used to measure the statistical difference between the predicted distribution and the actual observed distribution. It is suitable for uncertainty modeling and distribution matching scenarios.

[0064] For example, consider predicting and evaluating the power output of a small wind power station over a certain time period. It is known that the actual observed power output follows a certain probability distribution, and a corresponding probability distribution of the virtual predicted power output is obtained through a virtual prediction method.

[0065] Actual observed power distribution: Assuming the actual observed power There are only two possible values ​​and their probability distributions are as follows: when power kilowatt-hour, probability When power kilowatt-hour, probability ; Virtual predicted power distribution: Power obtained through virtual prediction There are also two possible values ​​and their probability distributions as follows: when power kilowatt-hour, probability When power kilowatt-hour, probability .

[0066] The Wasserstein distance calculation steps apply to two discrete probability distributions. and The formula for calculating the Wasserstein distance is: in, This represents the probability distribution vector of the actual observed values. It is the number of possible values ​​that the actual observed value can take. The actual observed value is taken as the first Value The probability of and satisfying .

[0067] In the example of wind power generation prediction, This means that the probability of having a power of 100 kilowatts is 0.6, and the probability of having a power of 200 kilowatts is 0.4. This represents the probability distribution vector of the virtual predicted values. Similarly, It is the number of possible values ​​for the virtual predicted value. The virtual predicted value is taken as the first Value The probability of and satisfying , In the example, That is, the probability of having a power of 120 kilowatts is 0.5, and the probability of having a power of 180 kilowatts is 0.5. It is the first possible actual observation value Each possible value For example, in wind power generation scenarios, kilowatt, kilowatt. Is the virtual predicted value the first possible Each possible value As in the example kilowatt, Kilowatt. In the example, , , , , , , , , The calculation process is as follows: ; Calculated Wasserstein distance In kilowatts, the Wasserstein distance reflects the degree of difference between the virtual predicted power distribution and the actual observed power distribution. The smaller the Wasserstein distance, the smaller the difference between the two distributions, meaning the probability distribution of the virtual predicted parameters is closer to the actual observed distribution, and the better the virtual prediction method performs in uncertainty modeling and distribution matching. Conversely, the larger the Wasserstein distance, the greater the difference between the two, and the accuracy of the virtual prediction method needs to be improved.

[0068] For each set of virtual parameters The system will calculate its difference from the actual observation based on one or more of the above-mentioned difference measurement methods. Difference measure Difference measure This objectively reflects the degree of deviation between the power generation mode represented by the set of virtual parameters and the actual operating state. The smaller the difference, the closer the virtual parameter combination is to reality, and the more likely the underlying input condition combination is to be the key factor leading to the current actual power output.

[0069] Based on the difference measurement results, further analysis was performed on each set of virtual parameters. Calculate a credibility weight value The credibility weight value is used to quantify the representativeness, reliability, and predictive consistency of the power generation scenario represented by this set of parameters under the current actual operating environment.

[0070] The logical steps for calculating the weights include: Difference measure This is mapped to a positive, monotonically decreasing weight value. For example, the normalized inverse of the error can be used to convert the difference value into a weight value.

[0071] Indicates the first A measure of the difference between individual samples or prediction results. In different application scenarios... The specific calculation methods differ; for example, in a simple single-sample point comparison scenario, if the concept of mean squared error is used to approximate the difference, for the ... For each sample, the predicted value is... The actual value is ,but When there are multiple samples, it is common practice to first calculate the similarity difference values ​​for each sample, and then calculate a certain statistic of these differences as a measure of the difference in the sample set. The larger the value, the more significant the first... The greater the deviation between the predicted value and the true value of a sample, the worse the matching effect.

[0072] Indicates according to The converted result is the first The weight values ​​for each sample or prediction result. The goal of this embodiment is to assign higher weights to samples with smaller differences and lower weights to samples with larger differences. And it decreases monotonically, when When decreasing, It will increase; when When it increases, It will decrease.

[0073] This represents the total number of samples. It is used in calculating weight values. During the normalization process, it is necessary to consider the case of all samples. This represents the number of samples used in the calculation. For example, if there are 100 data samples used to evaluate the performance of the prediction model, then... Calculation formula and derivation logic for weight values The calculation formula is: The derivation logic is as follows: molecule This embodies the idea of ​​the reciprocal of the error. Because The larger the value, the greater the error. The smaller it is; conversely, The smaller the value, the smaller the error. The larger the denominator, the more accurate the calculation. This establishes a preliminary measure that is inversely proportional to the magnitude of the error. This involves summing the reciprocals of the errors of all samples. Normalization is achieved by dividing the numerator by the denominator. The result is... The value will be between 0 and 1, and the sum of the weights of all samples is 1, satisfying the basic property of weights, while also ensuring that the weight value changes with the difference measure. It increases and then decreases monotonically.

[0074] Furthermore, to ensure that all weights are within a uniform comparison range, the original weights need to be normalized so that the sum of the weights of all virtual parameters is 1, thereby forming a set of comparable and interpretable weight distributions.

[0075] If multiple difference metrics are used, such as considering both MSE and cosine similarity, the results of multiple metrics can be integrated into a single comprehensive weight through methods such as weighted averaging, principal component fusion, or ranking consistency aggregation, thereby improving the robustness and comprehensiveness of weight calculation.

[0076] Furthermore, each set of virtual parameters Each is assigned a weight value between 0 and 1. The higher this value, the closer the power generation mode represented by this set of parameters matches the actual operating state. The more likely the corresponding combination of input conditions is to be the key driving factor leading to the current power output, thus possessing higher reliability and representativeness. After obtaining the weight values ​​of all virtual parameters, a Top-K selection strategy is used to filter the parameter set, retaining the set or groups of virtual parameters with the highest weights that best represent the current actual operating state. The specific filtering steps are as follows: Sort all N sets of virtual parameters according to their weight values ​​from high to low to form an ordered list. Based on actual needs or system settings, select the top K virtual parameter combinations by weight, where K is a preset positive integer, such as Top 5, Top 10, or dynamically determined based on the number of parameters with a cumulative weight ratio exceeding 80%, or set a weight threshold to retain all parameter combinations with weights higher than that threshold.

[0077] Construction of a high-confidence parameter subset: combining the selected high-weight parameters into... This constitutes a new, streamlined set of virtual parameters that reflects the combination of input conditions and system behavior patterns most likely to result in actual power output within the current prediction time window.

[0078] It should be noted that this screening mechanism can effectively eliminate low-confidence parameter combinations with low weights and serious deviations from actual observations, avoiding interference terms from misleading subsequent analysis. At the same time, it can significantly reduce computational complexity and invalid information load, and improve the accuracy, efficiency and interpretability of the overall prediction process.

[0079] In this embodiment of the invention, step S500 determines whether the current predicted power meets the preset critical condition based on a high-weight subset of virtual power extension parameters, and outputs the corresponding probabilistic key indicators when the condition is met, including: If the current predicted power does not meet any critical condition, an extended virtual power parameter set for the next prediction time window is generated based on the high-weight parameter subset, and the weights are recalculated and filtered. If the current predicted power meets any critical condition or reaches the preset prediction period end, the iteration terminates. Based on the high-weight virtual power extended parameter subset at the time of termination iteration, output probabilistic key indicators.

[0080] In this embodiment of the invention, step S500 further includes: The preset critical conditions include at least any of the following risky operating states: power reaching the upper limit of grid-connected capacity, power falling below the lower limit of market transactions, power fluctuations exceeding the system stability margin or energy storage depletion, frequent power curtailment, and insufficient reserves. The key probabilistic indicators include at least the probability of reaching critical power or risk state, the confidence interval of predicted power, the probability of the system being in a high / low risk operating state, the probability of stable power operation within a predetermined time period, and the expected power.

[0081] In an optional embodiment, if the judgment result indicates that the currently retained virtual power prediction result has not yet reached any critical condition, then a probabilistic modeling method is used to further time-recursively or conditionally expand the prediction of the retained virtual power expansion parameters to obtain the expanded virtual power data and the corresponding updated set of virtual power expansion parameters in the next prediction time window. The weights of the expanded virtual power data are recalculated, high-confidence parameter combinations are selected and retained, and it is judged again whether the critical state has been reached.

[0082] If the retained virtual power prediction results meet the preset critical conditions, such as the power reaching the limit, the system risk exceeding the tolerance boundary, the prediction period ending, or the probability distribution falling into the unacceptable range, then the prediction time point corresponding to that moment will be taken as the critical failure time, or a series of probabilistic indicators under the current state will be directly output to realize the probabilistic analysis, reliability assessment, and risk warning of wind and solar power output, providing scientific, quantitative, and interpretable technical support for power dispatching decisions, equipment operation and maintenance management, market trading strategies, and system safe operation.

[0083] For example, within each prediction time window, such as a 15-minute, 30-minute, or 60-minute prediction period from the current moment, it is determined whether the system has triggered any predefined critical condition based on the currently retained high-weight virtual power prediction results. Critical conditions are key boundaries for the system's operational safety, stability, and economy, and typically include, but are not limited to, the following typical scenarios: If the predicted power value reaches or exceeds the upper limit of the system's grid-connected capacity, such as inverter capacity, transformer limits, and line transmission capacity, it may lead to wind and solar power curtailment or equipment overload. If the real-time or predicted power is lower than the lower limit of market transactions required by the grid dispatch, such as minimum output commitments or ancillary service thresholds, it may trigger performance evaluations or transaction defaults. If the time series fluctuation of the predicted power, such as standard deviation, peak-to-valley difference, and ramp rate, exceeds the system's preset stability margin or grid acceptance capacity threshold, it may threaten system frequency stability or trigger protection actions. If the system frequently triggers power curtailment commands within the predicted time window, the energy storage system's state of charge (SOC) is close to depletion, reserve capacity is insufficient, grid congestion intensifies, or there is a risk of cascading equipment failures, it indicates that the system is operating in a high-risk state.

[0084] For each type of critical condition, the system will base its prediction on the current virtual power and its extended parameters. The system checks whether any critical trigger condition is met one by one through logical judgment, threshold comparison, or state machine model. If none of the critical conditions are triggered, it indicates that the system's operating state within the current prediction window is still within an acceptable range, and the system enters the next stage of extended prediction and iterative update process; conversely, if any critical condition is met, the current moment is determined to be the critical failure time point, and the final probabilistic index output process is triggered.

[0085] If no critical conditions are detected within the current prediction window, the system will further use probabilistic modeling methods based on the selected high-weight virtual power extension parameters to perform time-based or conditional extension predictions on the current parameter set to obtain the next prediction time window, such as the extended virtual power data and the corresponding updated set of virtual power extension parameters for the next 15 / 30 / 60 minutes. The processing steps for this extension prediction are as follows: Based on current weather forecast data, equipment status evolution trends, and dynamic adjustments to power grid constraints, the input variables in the original virtual parameter set are reasonably extrapolated and updated over time to generate input condition combinations suitable for the next prediction window. For each expanded input condition combination, its corresponding power output probability distribution characteristics are retained or re-estimated to ensure that the prediction results reflect both deterministic trends and uncertainties. The updated input conditions and their corresponding predicted power are encapsulated into new virtual parameter units, forming the expanded virtual parameter set for the next time window. This serves as the input basis for the next iteration cycle. Through this extension mechanism, the system achieves dynamic continuous prediction from the current moment to multiple future time windows, thereby capturing the dynamic characteristics of power generation evolution over time and the accumulation of potential risks.

[0086] In obtaining the extended virtual parameter set Then, the system will execute the same processing logic as the previous round for each set of virtual parameters in the new set, including: Based on the latest actual observed power, calculate the difference measure between each set of extended parameters and the actual / reference power; Calculate the confidence weight of each group of parameters based on the difference results; Top-K or threshold filtering methods are used to retain a subset of virtual parameters with high weight and high representativeness; Re-execute the critical state judgment logic to check whether any risk or constraint boundary is triggered, until any of the following termination conditions are met: Critical conditions are triggered within a certain prediction window, such as power exceeding limits, fluctuation margin exceeding limits, and unacceptable system risk. When the forecast period ends, such as when the preset maximum forecast duration is reached, such as 24 hours or 72 hours; The probability distribution falls into an unacceptable range, such as when the confidence interval is too wide or the expected power deviates too much from the scheduling target.

[0087] Once the iteration process terminates due to a critical trigger or the end of the prediction period, the system will output a set of probabilistic key performance and risk assessment indicators based on the last round's high-weight virtual parameter set. These indicators comprehensively characterize the system's operational status and risk level at that moment or within the prediction period. The indicators typically include, but are not limited to: Probability of reaching critical power or risk state: Based on the distribution characteristics of the reserved parameter set, calculate the probability that the system will encounter various critical conditions in a specific period of time in the future; Confidence interval for predicted power: Provides the expected value of future power output and its upper and lower bounds at the confidence level, reflecting the range of reliability of the prediction; Probability of the system operating in a high / low risk state: Based on the distribution of risk indicators, quantify the likelihood of the system operating in different risk level ranges; Probability of stable power operation within a certain period of time in the future: assess the overall probability that the system will maintain stable output within the prediction period; Expected power and distribution characteristics: The expected future power, typical distribution patterns and key statistics derived from the reserved parameter set provide a quantitative reference for scheduling decisions.

[0088] Example 2: The above example is an illustrative scheme of a wind and solar power generation probability analysis method. It should be noted that the technical solution of this wind and solar power generation probability analysis system belongs to the same concept as the technical solution of the above-described wind and solar power generation probability analysis method. Details not described in detail in this example can be found in the description of the technical solution of the above-described wind and solar power generation probability analysis method.

[0089] This embodiment provides a wind and solar power generation probability analysis system, comprising: The data acquisition module is used to acquire multi-source heterogeneous data from the target wind and solar power station and preprocess the multi-source heterogeneous data to construct the input variable space. The power prediction module is used to generate corresponding virtual power prediction parameters and their power output probability distribution based on multiple input combinations in the input variable space, forming an initial set of virtual power prediction parameters. The difference analysis module is used to obtain the actual observed power data within the time window corresponding to the initial virtual power prediction parameter set, and to measure the difference between the predicted power corresponding to each set of virtual power prediction parameters and the actual observed power data. The weight calculation module is used to calculate the confidence weight of each group of virtual power prediction parameters based on the difference measurement results, and to filter out the subset of virtual power extended parameters with high weights based on the confidence weights. The probabilistic index output module is used to determine whether the current predicted power meets the preset critical conditions based on a high-weight virtual power extended parameter subset, and outputs the corresponding probabilistic key indexes when they are met.

[0090] This embodiment also provides an electronic device applicable to the wind and solar power generation probability analysis method, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the wind and solar power generation probability analysis method proposed in the above embodiments.

[0091] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the wind and solar power generation probability analysis method proposed in the above embodiments.

[0092] The storage medium proposed in this embodiment and the method for probabilistic analysis of wind and solar power generation proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0093] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for probabilistic analysis of wind and solar power generation, characterized in that, include: Acquire multi-source heterogeneous data of the target wind and solar power station, and preprocess the multi-source heterogeneous data to construct an input variable space; Based on multiple input combinations in the input variable space, corresponding virtual power prediction parameters and their power output probability distributions are generated, forming an initial set of virtual power prediction parameters. Obtain the actual observed power data within the time window corresponding to the initial virtual power prediction parameter set, and measure the difference between the predicted power corresponding to each set of virtual power prediction parameters and the actual observed power data; Based on the difference measurement results, the confidence weight of each group of virtual power prediction parameters is calculated, and a subset of virtual power extended parameters with high weights is obtained based on the confidence weights. Based on the high-weighted subset of virtual power extension parameters, it is determined whether the current predicted power meets the preset critical conditions, and when it does, the corresponding probabilistic key indicators are output.

2. The wind and solar power generation probability analysis method as described in claim 1, characterized in that, The step of determining whether the current predicted power meets the preset critical condition based on the high-weight virtual power extension parameter subset, and outputting the corresponding probabilistic key indicators when the condition is met, includes: If the current predicted power does not meet any critical condition, an extended virtual power parameter set for the next prediction time window is generated based on the high-weight parameter subset, and the weights are recalculated and filtered. If the current predicted power meets any critical condition or reaches the preset prediction period end, the iteration terminates. Based on the high-weight virtual power extended parameter subset at the time of termination iteration, output probabilistic key indicators.

3. The wind and solar power generation probability analysis method as described in claim 2, characterized in that, Also includes: The preset critical conditions include at least any of the following risky operating states: power reaching the upper limit of grid-connected capacity, power falling below the lower limit of market transactions, power fluctuations exceeding the system stability margin or energy storage depletion, frequent power curtailment, and insufficient reserves. The probabilistic key indicators include at least the probability of reaching critical power or risk state, the confidence interval of predicted power, the probability of the system being in a high / low risk operating state, the probability of stable power operation within a predetermined time period, and the expected power.

4. The wind and solar power generation probability analysis method as described in claim 3, characterized in that, The subset of virtual power expansion parameters with high weights obtained based on confidence weighting includes: All virtual parameters in the initial virtual power prediction parameter set are sorted according to their corresponding confidence weight values ​​to form an ordered parameter list; According to a preset screening strategy, high-weight parameters are selected from the ordered parameter list to construct a subset of high-confidence virtual power expansion parameters. The preset screening strategy includes any one or more combinations of the following: Select the top K virtual parameters by weight, where K is a preset positive integer; Select the first few virtual parameters whose cumulative weight percentage reaches the preset threshold. Retain all virtual parameters whose credibility weight is not less than the preset weight threshold.

5. The wind and solar power generation probability analysis method as described in claim 4, characterized in that, The confidence weights for each group of virtual power prediction parameters are calculated based on the difference measurement results, including: For each set of virtual parameters in the initial set of virtual power prediction parameters, a difference metric is calculated based on the predicted power corresponding to each set of virtual parameters and the actual observed power within the same prediction time window. The difference metric is then converted into a positive confidence index. The inverse error of each set of virtual parameters is normalized to obtain the confidence weight of each set of virtual parameters.

6. The wind and solar power generation probability analysis method as described in claim 5, characterized in that, Based on multiple input combinations in the input variable space, corresponding virtual power prediction parameters and their power output probability distributions are generated, forming an initial set of virtual power prediction parameters, including: Generate N sets of input combinations based on the input variable space. Each input combination This corresponds to a wind and solar power generation operation scenario; For each input combination Generate corresponding virtual power prediction parameters ; Integrate all N sets of virtual power prediction parameters into an initial set of virtual power prediction parameters. .

7. The wind and solar power generation probability analysis method as described in claim 6, characterized in that, The input variables include: meteorological variables, equipment status variables, geographical and microclimate variables, and power grid constraints and operating condition variables.

8. A wind and solar power generation probability analysis system, applied to the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous data of the target wind and solar power station and preprocess the multi-source heterogeneous data to construct an input variable space; The power prediction module is used to generate corresponding virtual power prediction parameters and their power output probability distribution based on multiple input combinations in the input variable space, forming an initial set of virtual power prediction parameters. The difference analysis module is used to obtain the actual observed power data within the time window corresponding to the initial virtual power prediction parameter set, and to measure the difference between the predicted power corresponding to each set of virtual power prediction parameters and the actual observed power data. The weight calculation module is used to calculate the confidence weight of each group of virtual power prediction parameters based on the difference measurement results, and to filter out a subset of virtual power extended parameters with high weights based on the confidence weights. The probabilistic index output module is used to determine whether the current predicted power meets the preset critical conditions based on the high-weight virtual power extended parameter subset, and output the corresponding probabilistic key index when the conditions are met.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the wind and solar power generation probability analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the wind and solar power generation probability analysis method according to any one of claims 1 to 7.

Images