Generating capacity prediction method of photovoltaic power generation system

Abstract

The invention provides a generating capacity prediction method for a photovoltaic power generation system, and relates to the technical field of photovoltaic power generation, and the method comprises the steps: executing the data collection of the photovoltaic power generation system, and building a feature data set; a functional model agent pool is established, prediction models in the pool are in one-to-one correspondence with the feature data sets, and the prediction models are weak prediction models; configuring a prediction revenue function, wherein the evaluation dimension comprises prediction accuracy, confidence distribution and association influence contribution degree; calling a corresponding prediction model to predict the generating capacity, and performing a prediction game by using a revenue function; continuous iteration is executed, the output weight is updated, and the cooperation strategy is adjusted; and when iteration stops at any moment, reading the corresponding output weight and cooperation strategy, and constructing a generating capacity prediction result. According to the invention, the technical problem that the prediction accuracy is poor because the prior art usually depends on a single prediction model to predict the photovoltaic generating capacity and various influence factors in the system cannot be fully captured is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and in particular to a method for predicting power generation of a photovoltaic power generation system. Background Art

[0002] Photovoltaic power generation is widely used as a green and environmentally friendly form of energy. Photovoltaic power generation systems provide a sustainable solution for energy supply by converting solar energy into electrical energy. However, the power generation of photovoltaic power generation systems is affected by many factors, such as weather, equipment status, seasonal changes, and geographical location. Therefore, accurately predicting the power generation of photovoltaic power generation systems is crucial for grid scheduling, energy management, and system maintenance.

[0003] However, existing technologies usually rely on a single prediction model to predict photovoltaic power generation. A single model cannot fully capture the various influencing factors in the system. Although ensemble learning methods attempt to improve prediction accuracy through multi-model fusion, in practical applications, existing multi-model methods do not fully utilize the complementary advantages between different models and often lack effective inter-model collaboration strategies. Therefore, these methods ignore the complementarity between models, resulting in unstable prediction results, especially in complex or extreme environments, where the prediction accuracy is poor. Summary of the Invention

[0004] This application provides a method for predicting the power generation of a photovoltaic power generation system, aiming to solve the technical problems that the existing technology usually relies on a single prediction model to predict photovoltaic power generation, which cannot fully capture various influencing factors in the system, and the existing multi-model method cannot fully utilize the complementary advantages between different models and lacks an effective inter-model collaboration strategy, resulting in poor prediction accuracy.

[0005] The present application discloses a method for predicting power generation of a photovoltaic power generation system, the method comprising: executing data collection of the photovoltaic power generation system to establish a feature data set, the feature data set comprising an environmental feature data set, an equipment feature data set, a time feature data set, and a spatial feature data set; establishing a functional model proxy pool, the prediction models in the functional model proxy pool corresponding one-to-one to the feature data sets, and the prediction models are weak prediction models; configuring a prediction benefit function, the evaluation dimensions of the prediction benefit function comprising prediction accuracy, confidence allocation, and correlation influence contribution; after inputting the feature data set into the functional model proxy pool, calling the corresponding prediction model to predict power generation, and using the benefit function to perform a prediction game; executing continuous iteration of the prediction game, updating the output weight of each prediction model, and adjusting the collaboration strategy; when the iteration stops at any time, reading the corresponding output weight and collaboration strategy to construct a power generation prediction result.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects: By comprehensively collecting various types of data from photovoltaic power generation systems and establishing a multi-dimensional feature data set, this process provides rich and diverse input data for power generation forecasting, which can fully reflect the various factors affecting power generation and improve the accuracy of the forecast; by creating a functional model agent pool, each forecasting model focuses on processing specific data features, so that each model can be optimized for different data features, which improves the flexibility and adaptability of the forecasting system. The use of weak forecasting models in the model pool allows the system to take advantage of multiple lightweight models, avoid the high computational cost of a single complex model, and reduce the resource usage of the system; the configuration of the benefit function ensures a multi-dimensional evaluation of the model performance by introducing multiple evaluation dimensions. This mechanism allows each model to be adjusted not only based on its accuracy, but also considering its contribution to the final forecast and its collaborative relationship, thereby improving the overall forecasting effect; through the forecasting game, multiple models can optimize their respective outputs through collaboration and competition mechanisms. During the game, the performance of the model is evaluated through the profit function, which not only can dynamically adjust the weight of each model, but also ensure the maximization of the synergy between models, thereby improving the overall performance of the prediction; through continuous iterative optimization, after each round of game, the weight of the model is automatically adjusted according to its performance, which enables the model weight to gradually converge to the optimal value, reducing the prediction error of the system, and by adjusting the collaboration strategy between models, the cooperation between multiple models can be made closer, avoiding excessive reliance on a single model, and ensuring that each model plays its due role in power generation forecasting; by reading the output weight and collaboration strategy when the iteration stops, a stable and optimal prediction result can be obtained, which means that the best model combination and collaboration strategy have been found in multiple iterations, which can provide more accurate and reliable power generation forecasts. This process effectively avoids unnecessary calculations and complexity in the prediction process, enabling the system to efficiently obtain the final power generation forecast results while ensuring a high prediction accuracy.

[0007] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A schematic flow chart of a method for predicting power generation of a photovoltaic power generation system is provided for an embodiment of the present application.

[0009] Figure 2 A schematic diagram of a prediction game process using a profit function in a method for predicting power generation of a photovoltaic power generation system is provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The embodiment of the present application provides a method for predicting the power generation of a photovoltaic power generation system, which solves the technical problems that the existing technology usually relies on a single prediction model to predict photovoltaic power generation, cannot fully capture various influencing factors in the system, and the existing multi-model method cannot fully utilize the complementary advantages between different models and lacks an effective inter-model collaboration strategy, resulting in poor prediction accuracy.

[0011] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically introduced in conjunction with the drawings in the specification.

[0012] like Figure 1 As shown, an embodiment of the present application provides a method for predicting power generation of a photovoltaic power generation system, the method comprising: Perform data collection of the photovoltaic power generation system and establish a feature data set, wherein the feature data set includes an environmental feature data set, an equipment feature data set, a time feature data set, and a space feature data set.

[0013] Various data about the photovoltaic power generation system are collected to establish a comprehensive feature dataset. The feature dataset includes multiple subsets. Specifically, the environmental feature dataset refers to various factors in the photovoltaic system's environment, such as weather conditions, temperature, humidity, solar radiation intensity, wind speed, and air quality. This data can be collected in real time through sensors or obtained from meteorological data sources. The equipment feature dataset refers to various hardware status information within the photovoltaic power generation system, such as the output power of the photovoltaic panels, the operating status of the inverter, power loss, and the health of the components. This data is collected through sensors, monitoring systems, and other equipment installed in the system, and involves equipment performance monitoring data. The temporal feature dataset refers to time-related factors, such as the system's operating time period, season, the distinction between weekdays and weekends, and the length of sunlight. Temporal features are particularly important for photovoltaic power generation system prediction because sunlight intensity varies over time, especially with the seasons. The spatial feature dataset involves the characteristics of the photovoltaic system's installation location, such as geographic coordinates (latitude and longitude), orientation, and tilt angle. These spatial features directly affect the amount of solar energy received, as the angle and orientation of the photovoltaic system determine the efficiency of solar radiation reception. By collecting these feature data and organizing them into structured feature data sets, rich input data can be provided for subsequent prediction models.

[0014] A functionalized model proxy pool is established, wherein the prediction models in the functionalized model proxy pool correspond one-to-one to the feature data sets, and the prediction models are weak prediction models.

[0015] Establish a functional model agent pool. A functional model agent pool refers to a centralized model management pool that contains multiple prediction models. These models can be based on different algorithms or frameworks, such as linear regression, decision trees, support vector machines, neural networks, etc., or other models suitable for time series prediction, regression analysis and other problems. These prediction models are weak prediction models, which means that they are not particularly powerful individually, but through integration methods, they can improve the accuracy of predictions. Each prediction model corresponds to a different part of the feature dataset. For example, some prediction models focus on environmental feature datasets, while other prediction models focus more on device features or time features. In the agent pool, a one-to-one correspondence is established between the prediction model and the feature dataset, which helps to optimize for different types of data.

[0016] A prediction benefit function is configured, wherein the evaluation dimensions of the prediction benefit function include prediction accuracy, confidence distribution, and correlation impact contribution.

[0017] Configure the prediction benefit function, which is used to evaluate the output of each prediction model so as to optimize the prediction model in the subsequent prediction game process. This prediction benefit function evaluates the performance of each model through multiple dimensions, including prediction accuracy, confidence allocation, and correlation impact contribution. Among them, prediction accuracy refers to the gap between the model prediction result and the actual result. Usually, prediction accuracy is measured by calculating the error (such as mean square error, absolute error). Smaller error values mean higher accuracy. Confidence allocation is the model's confidence in its prediction results, indicating that the model believes its prediction is correct under a certain prediction result. It is the degree of reliability. The higher the confidence, the more confident the model is in its own prediction. It can be represented by the probability distribution or confidence interval output by the model. For example, the regression model can output a predicted value and its confidence interval, which indicates the model's confidence in the prediction result. Prediction results with high confidence will be given higher weight in decision-making. The correlation influence contribution measures the contribution of each prediction model to the power generation prediction result, especially the importance of each feature in the prediction. The feature importance evaluation technology can be used to evaluate the impact of each feature on the prediction result. In the subsequent prediction game process, higher weights are given to features with greater influence.

[0018] After the feature data set is input into the functionalized model agent pool, the corresponding prediction model is called to perform power generation prediction, and the profit function is used to perform prediction game.

[0019] The complete feature dataset is input into the functional model agent pool. Each prediction model in the agent pool is trained for a specific feature dataset. Therefore, based on the input feature dataset, the model prediction calculation is performed to obtain the corresponding power generation forecast result. The prediction game is an optimization process in which each prediction model evaluates its performance in the prediction through a payoff function based on the comparison of its output results with the actual results. In the prediction game, the three evaluation dimensions of the payoff function (prediction accuracy, confidence distribution, and associated influence contribution) will determine how to adjust the output weight of each model. For example, if a model has a high prediction accuracy and a high confidence level, then its output will be given a higher weight. Conversely, if the model's prediction error is large, it will be given a smaller weight in the next round of the game.

[0020] Continuous iterations of the prediction game are performed, the output weights of each prediction model are updated, and the collaborative strategy is adjusted.

[0021] The continuous iteration of the prediction game is an iterative optimization process. In each iteration, the performance of each prediction model is evaluated using a prediction reward function based on the output of the current prediction model. The output weight of each model is updated based on the accuracy, confidence, and associated impact contribution of the prediction results. In other words, the influence of each model on the final power generation forecast is adjusted. Specifically, if a model provides a more accurate prediction in the current iteration (for example, a lower prediction error or a higher confidence level), its output weight is increased; if a model's prediction performance is poor (for example, a high error or a low confidence level), its output weight is reduced. This process ensures that more accurate models will play a greater role in the next round of predictions, while poorly performing models will gradually reduce their influence.

[0022] In each iteration, not only are the weights of individual models updated, but the collaboration strategies between models are also adjusted. This means dynamically adjusting the roles of different models in power generation forecasting based on their mutual influence, degree of collaboration, and relative importance. Adjusting the collaboration strategy optimizes information sharing and collaboration between different forecasting models. For example, if two models have a mutually beneficial collaborative relationship (i.e., their forecast results are highly complementary), their collaboration will increase. Conversely, if the collaboration between certain models is poor, their collaboration strategies will be weakened or reduced. Adjusting the collaboration strategy helps reduce redundant predictions and optimize the collaborative relationships between models, enabling the system to more accurately predict power generation after multiple iterations.

[0023] When the iteration stops at any time, the corresponding output weights and collaboration strategies are read to construct the power generation forecast results.

[0024] The iterative process ends when specific stopping conditions are met. Common stopping conditions include the maximum number of iterations, error convergence, and convergence of the profit function. The iterative process terminates when any of these conditions are met. After the iterations terminate, the final output weights and collaboration strategies of each prediction model are read. Each model's final weight reflects its performance during the iteration process: high-performing models receive higher weights, while low-performing models receive lower weights. In the multi-model collaboration framework, the final collaboration strategy determines how the models interact, such as how information is shared and how weighting is applied. Based on the output weights and collaboration strategy, a power generation forecast is constructed. Specifically, a weighted average is calculated based on the weights of each model, serving as the final forecast. If multiple prediction models are used, their outputs are combined and a weight adjustment fusion strategy is used to generate the final forecast. The collaboration strategy is then used to further adjust the forecasts of the different models to optimize overall performance. The final power generation forecast can be used to make practical power generation management decisions, such as resource scheduling, load forecasting, and equipment maintenance.

[0025] Furthermore, if Figure 2 As shown, the prediction game using the profit function includes: Establish initial weights for each prediction model; after inputting the feature data set into the prediction model, use the prediction model to predict power generation, and evaluate the profit value of each power generation prediction result through the profit function; use the profit value to adjust the output weight of each prediction model based on the initial weight, and adjust the collaboration strategy between models to complete a round of update iteration.

[0026] Establish an initial weight for each prediction model. This weight value reflects the degree of influence of the model on the prediction results in the initial stage. The selection of the initial weight can be based on multiple factors, such as the historical performance of the model, the model type, the coverage of the feature data, etc. For example, if a model has performed well in past prediction tasks, it can be assigned a higher initial weight.

[0027] The feature dataset is input into the prediction model. Each model processes its corresponding feature dataset and calculates the power generation forecast. The prediction result is usually a continuous value, representing the power generation under given conditions. The prediction results of each model are evaluated using a pre-defined benefit function, including prediction accuracy, confidence allocation, and the contribution of associated influence. After the prediction results of each model are evaluated by the benefit function, a benefit value is assigned to represent the model's effectiveness in the current forecast.

[0028] The initial weights of the prediction models are adjusted based on their payoffs, with better-performing models receiving higher weights and poorer-performing models receiving lower weights. Adjusting the collaborative strategy involves optimizing information sharing and collaboration mechanisms between different models so that they complement each other rather than duplicate work. For example, if the prediction results of two models are highly complementary (i.e., their errors differ), the collaborative relationship between them should be strengthened to increase information sharing. In each round of the game, the payoffs of each model are continuously evaluated, and the model weights and collaborative strategies are adjusted based on these payoffs. This process helps to continuously optimize prediction accuracy and maximize the overall system's prediction effect by adjusting the relative importance of the models and the collaborative mechanisms. The end of an iteration means that the weights and collaborative strategies of all models have been updated, and preparations are made for the next round of predictions.

[0029] Furthermore, adjusting the output weight of each prediction model based on the initial weight using the benefit value includes: The output weight is adjusted through the weight update formula as follows: ; in, Characterization At the first iteration, The output weight of the prediction model, Characterization At the first iteration, The output weight of the prediction model, The main driving factor of returns is For the The prediction model in The profit value of the iteration, Characterization in the The average return of all prediction models in iterations, is the collaborative feedback factor, Representation is different from predictive model The prediction model, Characterization in the The prediction model under the iteration and prediction models The cooperative mutual benefit coefficient is used to measure the predicted synergistic relationship between them. Characterization At the first iteration, The output weights of the prediction model.

[0030] Specifically, the weight update formula is as follows: ; in, Characterization in the The prediction model under the iteration and prediction models The collaborative mutual benefit coefficient is used to measure the predictive synergy between them, which is calculated by the following factors: output correlation (such as Pearson correlation coefficient), joint improvement rate (the improvement of the relative individual prediction when the two are used together), Bayesian contribution or Shapley approximation.

[0031] This formula is used to update the weight of the prediction model. Its core idea is to optimize the weight of each model through continuous iteration to make the overall prediction result more accurate. In each iteration, the weight of the model is determined by three factors, among which, is the initial weight item for each iteration; This is the profit-driven part of the model. It adjusts the weight based on the difference between the predicted profit of each model and the average profit of all models to ensure that the model's prediction results are closer to the overall trend. The collaboration part is based on the collaboration effect between models, through the collaboration factor Adjust the complementary effects between models, promote cooperation between models, and reduce the differences in weights between models.

[0032] In general, this formula optimizes the weights of the prediction model through two mechanisms: revenue-driven and collaborative adjustment. In each iteration, the model's weight is adjusted not only based on its own performance (revenue), but is also affected by the outputs of other models. By adjusting the weight differences and collaborative effects, the overall prediction effect is ultimately optimized.

[0033] Furthermore, the output weight adjustment by the weight update formula includes: A smoothing penalty mechanism is established, and a historical fluctuation analysis of the prediction model is performed. A trigger analysis of the smoothing penalty mechanism is performed based on the results of the historical fluctuation analysis, and a first weight evolution stability feedback is generated using the trigger analysis results. A memory update mechanism is established. If the historical prediction result of the prediction model is a stable result and the prediction of this iteration is abnormal, the memory update mechanism is triggered, and a second weight evolution stability feedback is generated using the triggered memory update mechanism. Output weight adjustment constraints are performed based on the first weight evolution stability feedback and the second weight evolution stability feedback.

[0034] A smoothing penalty mechanism is established. The smoothing penalty mechanism is used to control the changes in the weights of the prediction model in multiple rounds of iterations. If the weight changes of a model are too large, the smoothing penalty mechanism will limit this fluctuation, thereby avoiding instability.

[0035] Through historical fluctuation analysis, the difference between each model's historical predictions and actual results is calculated. The volatility of these differences reflects the stability and accuracy of the model. The volatility of the model output can be evaluated by calculating the error standard deviation or root mean square error of each iteration. Higher volatility means unstable model predictions, while lower volatility indicates more stable model performance. When historical fluctuations reach a preset threshold, a smoothing penalty mechanism is triggered. Specifically, models with large historical fluctuations are penalized to reduce their weight fluctuations in the prediction. This mechanism can be triggered by setting a fluctuation tolerance. For example, if the model's error standard deviation exceeds a certain range, a penalty is applied.

[0036] Based on historical fluctuation analysis and a triggered smoothing penalty mechanism, a first weight evolution stabilization feedback is generated. This feedback is used to guide the model to adjust weights in the next iteration, making the model's weight changes more stable and reducing the uncertainty caused by excessive fluctuations. Typically, this feedback is achieved by adjusting the update speed or intensity of the model weights, for example, by reducing the update rate to avoid large weight adjustments.

[0037] A memory update mechanism is established. The memory update mechanism is used to store and update the historical prediction model status. In particular, when the model's prediction results show prediction anomalies in a certain round of iterations, this mechanism can help the model recover to a more stable state. Prediction anomalies can be defined by large prediction errors, abnormal model outputs, or significant deviations from actual data. For example, when the model's historical prediction results are relatively stable, and the current prediction results differ greatly from the historical results, the memory update mechanism is triggered. This mechanism will refer to the model's past performance and correct the current prediction error.

[0038] The second weight evolution stability feedback is feedback generated based on the memory update mechanism, which is used to correct the current weight adjustment to make it more consistent with the stability of historical performance. The form of this feedback includes adjusting the current prediction result back to a level similar to the historical prediction, or resetting the weight by weighted averaging the historical stability results.

[0039] The first feedback, derived from the analysis of the smoothing penalty mechanism, focuses on the stability of the prediction model amidst historical fluctuations. This mechanism smoothes the model's output by limiting weight fluctuations. The second feedback, derived from the memory update mechanism, focuses on the model's recovery from unusual predictions, restoring its historical stability by adjusting the model weights. These two feedback mechanisms work together to adjust the model weights, ensuring that the model's output is more stable and unaffected by unusual predictions or excessive fluctuations. Ultimately, the smoothing penalty and memory update mechanisms generate the final output weights for each prediction model. These weights are used in subsequent iterations to ensure the model is more stable and accurate in the next round of predictions.

[0040] Furthermore, the adjustment of the collaboration strategy includes: Establish a benefit feedback feature, which includes individual benefit feedback and average benefit deviation; establish a prediction result association feature, which includes an output similarity index and a prediction coupling degree index; establish a scene label dominant feature, and adjust the collaboration strategy based on the benefit feedback feature, the prediction result association feature, and the scene label dominant feature.

[0041] Construct a profit feedback feature to evaluate the performance of each prediction model and optimize it. The profit feedback feature mainly includes individual profit feedback and average profit deviation. Among them, individual profit feedback refers to the profit of the prediction results of each prediction model in the current iteration. This profit value is calculated through the profit function. The performance of the model in this iteration is evaluated based on the model's prediction accuracy, confidence distribution, and contribution. Individual profit feedback reflects the contribution of a single model and reflects the prediction quality of the model. If a model's prediction is very accurate, its profit feedback will be higher; if the model prediction error is large, its profit feedback will be lower; the average profit deviation measures the deviation between the average profit of all current models and the average profit of the previous iteration. This indicator reflects the overall performance change of all models. If the profit of the current iteration is significantly higher than that of the previous iteration, it means that the overall performance of the model has improved, and vice versa. The average profit deviation helps to evaluate the improvement of the system in multiple rounds of iterations. If the collective performance of the model is not as expected, the collaboration of the model can be re-optimized by adjusting the strategy.

[0042] Establish prediction result correlation features to measure the relationship between different prediction models and further optimize the collaboration strategy. The prediction result correlation features include output similarity index and prediction coupling degree index. Among them, the output similarity index measures the similarity between the outputs of different prediction models. This index is achieved by calculating the similarity between the prediction results of different models, such as using cosine similarity, Pearson correlation coefficient and other methods. Models with higher output similarity represent that they process specific data features in a similar way, so their prediction results may have higher redundancy. In this case, the redundancy of these models is reduced by adjusting the collaboration strategy to improve system efficiency; the prediction coupling degree index measures the degree of collaboration and complementarity between multiple models. It measures the closeness of the relationship between the prediction results of different models and is quantified by the correlation or synergy between model outputs. If the degree of coupling between two models is high, it means that they provide valuable supplementary information for the same prediction task. In this case, the collaboration between the two models will be enhanced, increasing their influence in the final prediction.

[0043] Establishing the dominant feature of scenario labels. The dominant feature of scenario labels refers to assigning a scenario label to each prediction task based on the specific scenarios or environments of different prediction tasks. The scenario label reflects the specific requirements of the current task, such as weather conditions, time period, equipment status and other factors. These scenario labels help the system adjust the collaboration strategy according to the needs of different tasks to ensure that the model performs optimally in specific scenarios. For example, in extreme weather conditions, some models are better at processing environmental characteristics, while other models are more accurate in terms of equipment health status.

[0044] The collaboration strategy is adjusted based on the benefit feedback characteristics, prediction result association characteristics, and scenario label dominance characteristics. For example, when certain models perform well in a specific scenario, their weights can be increased to enhance their influence in the final prediction. If the output similarity between models is high and the prediction coupling is strong, the influence of these models can be weakened to prevent redundancy. Conversely, if certain models perform poorly in a specific scenario, the collaboration strategy can be adjusted to reduce the participation of these models or even temporarily disable them. Based on a comprehensive analysis of the above characteristics, the collaboration method of each model is dynamically adjusted, including changing the weight distribution between models, adjusting the information sharing mechanism, and optimizing the complementarity between models. This strategy adjustment helps the system maximize prediction performance in different scenarios and task requirements.

[0045] Furthermore, the predictive game using the profit function further includes: Create an extreme scenario recognition channel, input the feature data set into the extreme scenario recognition channel, perform anomaly recognition of extreme weather, abnormal equipment status, and spatial mutation events, and establish a scenario recognition label; use the scenario recognition label to perform micro-model hot loading; add the hot-loaded micro-model to the prediction game, and perform scenario enhancement management.

[0046] Create an extreme scenario recognition channel. This channel analyzes input data and detects anomalies. It receives various types of information from feature datasets (such as environmental characteristics, equipment status, spatial changes, etc.) and performs specific recognition tasks on this data. This channel uses machine learning models, rule engines, or deep learning networks to analyze data and detect possible extreme events in real time. For example, based on weather data, it can identify extreme weather conditions such as strong winds, heavy rain, and snowstorms; based on equipment data, it can identify equipment failures or abnormal operating conditions; and based on spatial data, it can identify emergencies caused by changes in geographic or spatial location, such as shading and equipment damage. Based on this analysis, corresponding scene recognition labels are generated. These labels represent the type of extreme event in the current data, such as heavy rain, equipment failure, and shading effects. Scene labels identify the current abnormal state and provide a basis for subsequent processing steps.

[0047] Micromodels are small, lightweight models optimized for specific scenarios. These models do not require global training, but are instead quickly deployed for specific abnormal scenarios. For example, in extreme weather conditions, a micromodel specifically tailored to weather impacts is loaded. Hot loading means dynamically loading new micromodels without restarting the entire system. This mechanism ensures that the system can quickly respond to new scenario changes in a real-time environment. When an abnormal scenario is identified through the extreme scenario recognition channel and a scenario identification label is generated, the corresponding micromodel is quickly loaded based on this label. For example, if a scenario label for a device failure is identified, a micromodel specifically designed to deal with the device failure is loaded to adjust the prediction strategy.

[0048] When a micromodel is hot-loaded and ready, it participates in the prediction game. This means that these micromodels are not used alone, but collaborate and compete with other models to improve the prediction capabilities of the entire system. Specifically, the output of the micromodel is integrated into the prediction system and optimized based on the profit function together with other existing prediction models to ensure accurate predictions in abnormal scenarios. Scenario enhancement management refers to adjusting the collaboration strategy between models after identifying specific abnormal scenarios, so that the micromodel and other main models can work together to maximize prediction performance. In this way, the collaboration mode of the models can be flexibly adjusted in the face of different scenarios and changes to ensure the accuracy of the prediction results.

[0049] Furthermore, the hot-loaded micro-model is added to the prediction game to perform scenario enhancement management, including: Configure an output intervention threshold; when executing scenario enhancement management, use the output intervention threshold to identify the intervention trigger of the micromodel; and use the intervention trigger identification result to perform predictive game management.

[0050] The output intervention threshold refers to the intervention mechanism that is triggered when the output result of the model exceeds or falls below a preset range during the prediction process. These thresholds can be defined based on the prediction error, the confidence level of the prediction result, or the deviation of the model output. A threshold that is too low may lead to excessive intervention, while a threshold that is too high may cause the system to respond slowly, so precise adjustment is required.

[0051] When performing scenario enhancement management, the current abnormal situation is first identified based on the scenario label, such as extreme weather, equipment failure, etc. Then, the prediction results of the current micromodel are compared with the output intervention threshold. If the output of the micromodel exceeds the preset intervention threshold, the intervention mechanism will be triggered to start adjusting or ignoring the prediction results of the model.

[0052] Based on the intervention trigger identification results, the model weights in the game are adjusted. If some models are marked as abnormal, their weights will be reduced or removed completely. By adjusting the game strategy and redistributing the weights, the weights of those models with reliable prediction results and stable performance are increased, thereby improving the accuracy of the prediction.

[0053] Furthermore, the continuous iteration of executing the prediction game further includes: Establish a low-profit threshold, and after calculating the profit value of the prediction model for each iteration, perform trigger identification of the low-profit threshold; if the low-profit threshold trigger of any prediction model within the preset iteration window meets the expected proportion, generate an initialization instruction; and perform parameter initialization management of the corresponding prediction model according to the initialization instruction.

[0054] The low-yield threshold refers to when the profit value of a forecast model falls below a set value, indicating that the model's forecast performance is poor and requires intervention. The low-yield threshold is set through historical data analysis or prior knowledge. In each iteration, the predicted profit of each model is calculated based on the profit function. The profit value includes indicators such as the model's prediction accuracy, confidence distribution, and contribution. The profit value of each model is tracked and determined to be below the set low-yield threshold. If the profit value of a model falls below the preset low-yield threshold, the low-yield threshold trigger identification is triggered, indicating that the model has failed to achieve the expected performance level in the current iteration cycle and requires further adjustment.

[0055] The preset iteration window is a fixed time range or number of iterations used to evaluate the performance of the model in several consecutive rounds of iterations. This window helps us understand the long-term performance of the model rather than the fluctuations of a single round of iterations. If the low-return threshold of any model is triggered more than a certain proportion within this iteration window, it is considered that the model has failed to adapt to the current data or task in the long term and needs to be reinitialized.

[0056] The expected ratio refers to the percentage of low-yield threshold triggers exceeding a predetermined threshold within a preset iteration window. For example, if the model's return value for two rounds within a five-round iteration window is lower than the low-yield threshold, the trigger ratio is 40%. If this ratio exceeds the set expected ratio (such as 50%), an initialization instruction is generated. The setting of the expected ratio can be flexibly adjusted based on the model's fault tolerance, task requirements, and business scenarios. When the low-yield threshold meets the expected ratio, an initialization instruction is automatically generated. This instruction is used to initiate the model's reinitialization process to ensure that the model can be restored to a better prediction state.

[0057] Parameter initialization refers to resetting the model's parameters when the model performs poorly to ensure that the model can adapt to the data and task requirements again. Parameter initialization includes resetting weights, etc. After initialization management, the reinitialized model is monitored to ensure its performance is improved and the parameters are adjusted in a timely manner.

[0058] Furthermore, if the low-return threshold trigger of any prediction model within the preset iteration window meets the expected ratio, an initialization instruction is generated, including: Activate the sliding window to perform residual accumulation analysis of the prediction model to generate a deviation trend prediction result; establish auxiliary authentication based on the deviation trend prediction result, and trigger compensation of the expected proportion through the auxiliary authentication.

[0059] A sliding window is a fixed-length time window used for local analysis of model output. Each time the window is moved, new data points are considered while the oldest data points are discarded. This helps dynamically evaluate model performance and, by analyzing the short-term changes in the model's error, known as residuals, promptly identify deviations and trends in model predictions. Residuals refer to the difference between the model's predicted value and the actual observed value, calculated as the difference between the model's predicted value and the actual observed value.

[0060] Within each sliding window, the residuals of each prediction model are calculated and cumulatively analyzed. The cumulative change in residuals can be used to assess the stability and accuracy of the model. If the model's residuals continue to increase or become unstable over multiple windows, the model's forecast trend shift is identified, providing a basis for subsequent adjustments. The shift trend prediction result predicts the future direction of the shift by analyzing the trend of residual changes within the sliding window. For example, if the residuals in several consecutive windows gradually increase, the performance of the prediction model will continue to decline. Otherwise, it indicates that the model is becoming more stable.

[0061] The generated deviation trend prediction results are used to establish auxiliary authentication. In the auxiliary authentication, the relationship between the deviation trend prediction results and the expected prediction results is compared to determine whether the model needs further adjustment. If the deviation trend prediction results indicate that a certain model continues to deviate, the auxiliary authentication mechanism is used to confirm whether the model needs compensation. Based on the auxiliary authentication results, if the deviation trend prediction results indicate that the model's prediction results deviate significantly from the actual needs, and this deviation meets the set expected ratio, compensation measures are triggered. According to the compensation strategy, the model's weights, parameters, or other factors are adjusted to correct the deviation. The purpose of compensation is to ensure that the system's prediction performance returns to the ideal state and reduce the error from the actual results.

[0062] Furthermore, after constructing the power generation forecast result, the following steps are included: Obtain power generation task requirements, perform matching analysis based on the power generation task requirements and the power generation forecast results, and establish supply and demand anomalies; use the supply and demand anomalies to issue abnormal warnings, and perform abnormal management through the abnormal warnings.

[0063] The power generation task demand is the amount of power generation that needs to be provided based on a specific time period, region or task requirements. This demand is usually determined by users, scheduling systems or external factors. The power generation forecast result is the future power generation obtained through the forecast model. The power generation task demand is matched and analyzed with the power generation forecast result. Specifically, the difference between the predicted power generation and the actual demand is compared to analyze whether the demand can be met. If the predicted power generation is lower than the demand and the gap exceeds the preset threshold, the supply shortage is identified; if the predicted power generation is higher than the demand and the gap exceeds the preset threshold, the surplus is identified. Supply and demand anomalies are established based on the difference between the forecast results and the task requirements, including insufficient supply and excess resources.

[0064] When supply and demand anomalies are identified, the anomaly warning mechanism is triggered. The anomaly warning includes the anomaly type, severity, and impact range. Exception management is implemented based on the anomaly warning. This involves taking a series of measures to address the supply and demand imbalance after receiving an anomaly warning. These measures include adjusting power generation plans, load distribution, optimizing resource scheduling, and coordinating external resources. This ensures that the system can promptly identify and respond to power supply and demand imbalances, thereby optimizing the allocation and scheduling of power resources and ensuring the stable operation of the power system.

[0065] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting power generation of a photovoltaic power generation system, characterized in that: The method comprises: Performing data collection of the photovoltaic power generation system to establish a feature data set, wherein the feature data set includes an environmental feature data set, a device feature data set, a time feature data set, and a spatial feature data set; Establishing a functionalized model proxy pool, wherein the prediction models in the functionalized model proxy pool correspond one-to-one to the feature data sets, and the prediction models are weak prediction models; Configuring a prediction benefit function, wherein the evaluation dimensions of the prediction benefit function include prediction accuracy, confidence distribution, and correlation impact contribution; After inputting the feature data set into the functionalized model agent pool, the corresponding prediction model is called to predict the power generation, and the profit function is used to perform the prediction game; Perform continuous iterations of the prediction game, update the output weights of each prediction model, and adjust the collaboration strategy; When the iteration stops at any time, the corresponding output weights and collaboration strategies are read to construct the power generation forecast results.

2. The method for predicting power generation of a photovoltaic power generation system according to claim 1, wherein: The method of using the profit function to conduct a prediction game includes: Establish the initial weights for each prediction model; After inputting the feature data set into the prediction model, the prediction model is used to predict the power generation, and the profit value of each power generation prediction result is evaluated by the profit function; The output weight of each prediction model is adjusted based on the initial weight using the benefit value, and the collaboration strategy between models is adjusted to complete a round of update iteration.

3. The method for predicting power generation of a photovoltaic power generation system according to claim 2, wherein: The adjusting the output weight of each prediction model based on the initial weight using the benefit value includes: The output weight is adjusted through the weight update formula as follows: ； in, Characterization At the first iteration, The output weight of the prediction model, Characterization At the first iteration, The output weight of the prediction model, The main driving factor of returns is For the The prediction model in The profit value of the iteration, Characterization in the The average return of all prediction models in iterations, is the collaborative feedback factor, Representation is different from predictive model The prediction model, Characterization in the The prediction model under the iteration and prediction models The cooperative mutual benefit coefficient is used to measure the predicted synergistic relationship between them. Characterization At the first iteration, The output weights of the prediction model.

4. The method for predicting power generation of a photovoltaic power generation system according to claim 3, wherein: The output weight adjustment is performed by using a weight update formula, including: Establish a smoothing penalty mechanism, perform historical fluctuation analysis of the forecast model, conduct trigger analysis of the smoothing penalty mechanism based on the historical fluctuation analysis results, and use the trigger analysis results to generate the first weight evolution stability feedback; Establishing a memory update mechanism. If the historical prediction results of the prediction model are stable and the prediction of this iteration is abnormal, the memory update mechanism is triggered, and the triggered memory update mechanism is used to generate a second weight evolution stable feedback; Output weight adjustment constraints are performed according to the first weight evolution stability feedback and the second weight evolution stability feedback.

5. The method for predicting power generation of a photovoltaic power generation system according to claim 1, wherein: The adjustment of the collaboration strategy includes: Establishing a benefit feedback feature, wherein the benefit feedback feature includes individual benefit feedback and average benefit deviation; Establishing prediction result correlation features, wherein the prediction result correlation features include an output similarity index and a prediction coupling degree index; Establish the dominant features of the scene labels, and adjust the collaboration strategy based on the benefit feedback features, the prediction result association features, and the dominant features of the scene labels.

6. The method for predicting power generation of a photovoltaic power generation system according to claim 1, wherein: The predictive game using the profit function further includes: Creating an extreme scene recognition channel, inputting the feature data set into the extreme scene recognition channel, performing anomaly recognition of extreme weather, abnormal equipment status, and spatial mutation events, and establishing a scene recognition label; performing micro-model hot loading using the scene recognition tag; Add hot-loaded micromodels to prediction games and perform scenario enhancement management.

7. The method for predicting power generation of a photovoltaic power generation system according to claim 6, wherein: The hot-loaded micro-model is added to the prediction game to perform scenario enhancement management, including: Configure output intervention thresholds; When performing scene enhancement management, using the output intervention threshold to perform intervention trigger identification of the micro-model; Leveraging intervention trigger identification results for predictive game management.

8. The method for predicting power generation of a photovoltaic power generation system according to claim 1, wherein: The continuous iteration of executing the prediction game also includes: Establish a low-return threshold and perform trigger identification of the low-return threshold after calculating the forecast model return value for each iteration; If the low-return threshold trigger of any prediction model within the preset iteration window meets the expected ratio, an initialization instruction is generated; Parameter initialization management of the corresponding prediction model is performed according to the initialization instruction.

9. The method for predicting power generation of a photovoltaic power generation system according to claim 8, wherein: If the low-return threshold trigger of any prediction model within the preset iteration window meets the expected ratio, an initialization instruction is generated, including: Activate the sliding window to perform residual accumulation analysis of the prediction model and generate the deviation trend prediction results; Auxiliary authentication is established based on the deviation trend prediction result, and trigger compensation of the expected ratio is performed through the auxiliary authentication.

10. The method for predicting power generation of a photovoltaic power generation system according to claim 1, wherein: After the power generation prediction result is constructed, the following steps are included: Obtaining power generation task requirements, performing matching analysis based on the power generation task requirements and the power generation forecast results, and establishing supply and demand anomalies; The supply and demand anomaly is used to report an abnormality warning, and abnormality management is performed through the abnormality warning.

Images