Combined evaluation and prediction method and device for wind-solar power, equipment and medium

Abstract

The invention belongs to the technical field of renewable energy source prediction, and discloses a combined evaluation and prediction method and device for wind-solar power, equipment and a medium, and the method comprises the steps: carrying out the multi-model prediction through constructing a target spatial-temporal feature matrix; matching a target weather scene according to the current weather characteristics, and calculating an accuracy index of each model; according to the current weather feature matching degree and the historical score and accuracy index of each model in the target weather scene, calculating the dynamic weight of each preset model, and carrying out weighted summation on the wind and light power prediction values of each model to obtain a first fusion prediction value; inputting the target spatial-temporal feature matrix into a Stacking integrated model to obtain a second fusion prediction value; the two fusion prediction values are weighted and summed to obtain a target fusion prediction value, so that the weather scene self-adaptive calculation dynamic weight can be realized to carry out combined evaluation prediction, and the prediction precision is further improved; and meanwhile, the prediction accuracy can be improved based on a double fusion mode of dynamic weight weighting and adoption of a Stacking integrated model.

Description

Technical Field

[0001] This invention relates to the field of renewable energy prediction technology, specifically to a combined evaluation and prediction method and device for wind and solar power, electronic equipment, and storage medium. Background Technology

[0002] As the proportion of renewable energy in the energy mix continues to increase, wind and solar power, as an important component of clean energy, pose a severe challenge to grid stability and the economic operation of wind-solar hydrogen production systems due to the intermittent and uncertain nature of their output. Accurate wind and solar power forecasting is a key prerequisite for achieving efficient utilization of wind and solar energy, ensuring the safe and stable operation of the grid, and optimizing the scheduling of wind-solar hydrogen production systems.

[0003] Existing wind and solar power prediction methods are mostly based on single models, such as Long Short-Term Memory (LSTM), extreme gradient boosting (XGBoost), and physical models. For example, while a single Lightweight Gradient Boosting Machine (LightGBM) model can capture nonlinear relationships, it is prone to accumulating errors when dealing with long-term predictions. K-Nearest Neighbors (KNN) models excel at capturing local patterns, but lack the ability to grasp global trends. In summary, single models have weak generalization ability, exhibit unstable performance under different weather conditions, and have limited prediction accuracy.

[0004] Although some studies have attempted to integrate multiple models using weighted averaging or voting methods, existing integration methods mostly employ simple combinations, failing to fully exploit the complementarity between different models. Furthermore, problems such as noise accumulation and data distribution drift exist during long-term prediction, resulting in insufficient prediction accuracy. Summary of the Invention

[0005] To address the aforementioned shortcomings, the present invention aims to provide a combined evaluation and prediction method and device for wind and solar power, as well as an electronic device and storage medium, which can achieve scene-adaptive prediction and thus improve prediction accuracy.

[0006] The first aspect of this invention discloses a combined evaluation and prediction method for wind and solar power, comprising: Real-time acquisition of multi-source heterogeneous data, and construction of a target spatiotemporal feature matrix based on the multi-source heterogeneous data; The target spatiotemporal feature matrix is ​​input into each preset model to obtain the wind and solar power prediction values ​​of each preset model; Based on current weather characteristic data, the target weather scenario is matched and obtained; Calculate the first accuracy index for each of the preset models; Based on the current weather feature matching degree and the historical scores and first accuracy index of each preset model in the target weather scenario, the dynamic weight of each preset model is calculated. Based on the dynamic weights of each preset model, the wind and solar power prediction values ​​of each preset model are weighted and summed to obtain the first fused prediction value. The target spatiotemporal feature matrix is ​​input into the Stacking ensemble model to obtain a second fused prediction value; the Stacking ensemble model is constructed by integrating all the preset models. The first fusion prediction value and the second fusion prediction value are weighted and summed to obtain the target fusion prediction value.

[0007] In some embodiments, real-time acquisition of multi-source heterogeneous data and construction of a target spatiotemporal feature matrix based on the multi-source heterogeneous data include: Collect multi-source heterogeneous data, including measured data from wind and solar power stations, third-party meteorological data, satellite cloud image data, radar observation data, and third-party power prediction data; The multi-source heterogeneous data is preprocessed to obtain multi-source fused data; Based on the multi-source fusion data, a target spatiotemporal feature matrix is ​​constructed.

[0008] In some embodiments, calculating a first accuracy index for each of the preset models includes: Based on the prediction data within a specified recent period, various evaluation indicators for each preset model are calculated and standardized. Then, the comprehensive score of each preset model is obtained by weighted fusion and used as the first accuracy indicator.

[0009] In some embodiments, the dynamic weights of each preset model are calculated based on the current weather feature matching degree, the historical scores of each preset model in the target weather scenario, and a first accuracy index, including: The dynamic weights of the preset model k at time t are calculated using the following dynamic weight formula. And normalize the results: Where K is the total number of preset models. Based on the weights, ; It is the historical score of the preset model k at the current time t under the target weather scenario. For the first accuracy indicator, The degree of matching with current weather characteristics; , , These are the weighting coefficients for the historical score, the first accuracy index, and the current weather feature matching degree, respectively.

[0010] In some embodiments, after weighted summing of the first fusion prediction value and the second fusion prediction value to obtain the target fusion prediction value, the method further includes: Obtain the actual wind and solar power value, and calculate the second accuracy index based on the actual wind and solar power value and the target fused prediction value; The prediction error is calculated based on the actual wind and solar power value and the target fused prediction value, and the stability index is calculated based on the prediction error. The time taken for a single prediction is used as an efficiency indicator. Calculate the business value index based on the target fusion prediction value; The system monitors the second accuracy indicator, the stability indicator, the efficiency indicator, and the business value indicator in real time. When preset conditions are met, it triggers automatic optimization of each preset model, dynamic weight allocation strategy update, and weather mode adaptive adjustment.

[0011] A second aspect of this invention discloses a combined evaluation and prediction device for wind and solar power, comprising: A construction unit is used to collect multi-source heterogeneous data in real time and construct a target spatiotemporal feature matrix based on the multi-source heterogeneous data; The prediction unit is used to input the target spatiotemporal feature matrix into each preset model to obtain the wind and solar power prediction values ​​of each preset model; The matching unit is used to match the target weather scene based on the current weather feature data; A calculation unit is used to calculate the first accuracy index of each of the preset models respectively; The weighting unit is used to calculate the dynamic weight of each preset model based on the current weather feature matching degree and the historical scores and first accuracy index of each preset model in the target weather scenario. The first integration unit is used to perform a weighted summation of the wind and solar power prediction values ​​of each preset model according to the dynamic weights of each preset model to obtain a first fused prediction value. The second integration unit is used to input the target spatiotemporal feature matrix into the Stacking integration model to obtain a second fused prediction value; the Stacking integration model is constructed by integrating all the preset models. The fusion unit is used to perform a weighted summation of the first fused prediction value and the second fused prediction value to obtain the target fused prediction value.

[0012] In some embodiments, the building unit includes: The acquisition subunit is used to acquire multi-source heterogeneous data, including measured data from wind and solar power stations, third-party meteorological data, satellite cloud image data, radar observation data, and third-party power prediction data. The preprocessing subunit is used to preprocess the multi-source heterogeneous data to obtain multi-source fused data; A sub-unit is constructed to construct a target spatiotemporal feature matrix based on the multi-source fusion data.

[0013] In some embodiments, the apparatus further includes: The acquisition unit is used to acquire the actual wind and solar power value after the fusion unit performs a weighted summation of the first fusion prediction value and the second fusion prediction value to obtain the target fusion prediction value. An accuracy evaluation unit is used to calculate a second accuracy index based on the actual wind and solar power value and the target fused prediction value; The stability evaluation unit is used to calculate the prediction error based on the actual wind and solar power value and the target fused prediction value, and to calculate the stability index based on the prediction error. The efficiency evaluation unit is used to calculate the time taken for a single prediction as an efficiency indicator. The value evaluation unit is used to calculate business value indicators based on the target fusion prediction value; The monitoring unit is used to monitor the second accuracy indicator, the stability indicator, the efficiency indicator, and the business value indicator in real time. When preset conditions are met, it triggers the automatic optimization of each preset model, the dynamic weight allocation strategy update, and the weather mode adaptive adjustment.

[0014] A third aspect of the present invention discloses an electronic device, including a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the combined evaluation and prediction method of wind and solar power disclosed in the first aspect.

[0015] The fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the combined evaluation and prediction method for wind and solar power disclosed in the first aspect.

[0016] Compared with existing technologies, the advantages of this invention are as follows: It constructs a target spatiotemporal feature matrix for model prediction by real-time acquisition of multi-source heterogeneous data; it matches the target weather scenario with current weather features and calculates the first accuracy index of each preset model; it then calculates the dynamic weights of each preset model based on the current weather feature matching degree, the historical scores of each preset model under the target weather scenario, and the first accuracy index; and it weights and sums the wind and solar power prediction values ​​of each preset model to obtain a first fused prediction value. Simultaneously, it inputs the target spatiotemporal feature matrix into a Stacking ensemble model to obtain a second fused prediction value; finally, it weights and sums the two fused prediction values ​​to obtain the target fused prediction value. This allows for adaptive calculation of dynamic weights for combined evaluation and prediction of weather scenarios, thereby improving prediction accuracy. Furthermore, the dual fusion method based on dynamic weighting and the use of a Stacking ensemble model can improve prediction accuracy. Attached Figure Description

[0017] Figure 1 This is a flowchart of a combined evaluation and prediction method for wind and solar power disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a combined evaluation and prediction device for wind and solar power disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device disclosed in an embodiment of the present invention.

[0018] Explanation of reference numerals in the attached figures: 201. Construction unit; 202. Prediction unit; 203. Matching unit; 204. Calculation unit; 205. Weighting unit; 206. First integration unit; 207. Second integration unit; 208. Fusion unit; 301. Memory; 302. Processor. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0020] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 110, 120, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0021] It will be understood by those skilled in the art that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0022] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Throughout the description, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1This invention discloses a method for combined evaluation and prediction of wind and solar power. The method can be executed by electronic devices such as computers, laptops, and tablets, or by a combined evaluation and prediction device for wind and solar power embedded in an electronic device; this invention does not limit the specific device. In this embodiment, an electronic device is used as an example for illustration.

[0025] like Figure 1 As shown, the method includes the following steps 110-180: 110. Collect multi-source heterogeneous data in real time and construct the target spatiotemporal feature matrix based on the multi-source heterogeneous data.

[0026] Specific step 110 includes the following steps 1101 to 1103: 1101. Collect multi-source heterogeneous data, including measured data from wind and solar power stations, third-party meteorological data, satellite cloud image data, radar observation data, and third-party power prediction data.

[0027] The measured data from wind and solar power stations include wind speed, wind direction, solar irradiance, temperature, humidity, atmospheric pressure, and historical power generation. Third-party meteorological data includes numerical weather prediction (NWP) data. Third-party power prediction data includes short-term / ultra-short-term wind and solar power prediction results from multiple third-party supplier platforms.

[0028] In this embodiment of the invention, a standardized adapter can be used to access a third-party forecasting interface to obtain forecast results from a third-party meteorological forecasting platform for multi-source fusion. Specifically, a standardized RESTful API adapter enables bidirectional data interaction with the third-party forecasting interface, supporting power forecast push and operational feedback reception; a microservice architecture is deployed on a Kubernetes cluster to achieve elastic scaling; time-series data is stored in TimescaleDB, and configuration and metadata are stored in PostgreSQL; Apache Airflow triggers the data collection, forecasting, and evaluation process every 15 minutes; Grafana is used to display forecast curves, error distribution, weight evolution, and system health status. The forecast results are converted according to the hydrogen production load demand format and support a real-time feedback mechanism to optimize scheduling strategies. The forecast results from the third-party meteorological forecasting platform are quality-verified, for example, ensuring that the power does not exceed the installed capacity, to ensure its participation in subsequent multi-source fusion.

[0029] 1102. Preprocess the multi-source heterogeneous data to obtain multi-source fused data.

[0030] The preprocessing process includes imputing missing values, removing outliers, constructing features, and aligning time for multi-source heterogeneous data. The preprocessed data generates multi-source fusion data with a unified time granularity and spatiotemporal correlations. The unified time granularity can be set to 15 minutes.

[0031] Missing value imputation specifically employs multiple imputation methods to handle missing values, while outlier removal is specifically based on... The principle involves detecting and removing outliers; feature engineering specifically involves constructing time features, meteorological combination features, and lagged variables; among them, time features include hours, days of the week, and seasons, while meteorological combination features refer to comprehensive feature quantities used to characterize weather by combining multiple individual meteorological elements. Time alignment specifically involves using time series alignment algorithms to align multi-source heterogeneous data with different time resolutions into multi-source fused data with a unified time granularity.

[0032] 1103. Construct the target spatiotemporal feature matrix based on multi-source fusion data.

[0033] Specifically, a preset time window can be constructed, and a target spatiotemporal feature matrix can be built based on a portion of the multi-source fusion data within the preset time window. Optionally, the time step of the preset time window (horizon) can be set to 96, that is, the target spatiotemporal feature matrix is ​​constructed using the first 96 15-minute intervals (i.e., 24 hours) of multi-source fusion data.

[0034] 120. Input the target spatiotemporal feature matrix into each preset model to obtain the wind and solar power prediction values ​​of each preset model.

[0035] Each preset model includes at least a trained LightGBM model, a KNN model, and an FGL-LightGBM model. The FGL-LightGBM model is an improved LightGBM model based on Future-Guided Learning (FGL). The LightGBM model is used to model the global nonlinear meteorological-power mapping relationship. During training, the spatiotemporal feature matrices of all sample data are divided into training and test sets, with a sample size ratio of 7:3. Then, the model hyperparameters are configured, including: a fixed random seed of 42, a learning rate of 0.05, a maximum tree depth of 8, and 31 leaf nodes. The nonlinear relationship between meteorological variables and power generation is modeled using a gradient boosting tree structure, and the predicted value for a certain time point within the next 15 minutes to 2 hours is output.

[0036] The KNN model is used to capture power variation patterns under locally similar weather patterns. During training, the spatiotemporal feature matrix of the sample data is Z-score normalized, and a K value of 5 is selected. This model makes predictions based on the mean power of the local neighborhood of the current time and historical similar weather samples, and is suitable for ultra-short-term scenarios such as 0–1 hours, especially showing robustness during sudden weather changes.

[0037] The FGL_LightGBM model employs a Teacher-Student knowledge distillation architecture, using short-term prediction results to guide long-term predictions. Specifically, the FGL_LightGBM model includes a Teacher model and a Student model. The Teacher model is used to predict wind and solar power at a time step of horizon=1, i.e., the power over the next 15 minutes, and can be obtained by training the LightGBM model using the spatiotemporal feature matrix of the sample data. The Student model is used to predict wind and solar power at time steps of horizon=2–8, i.e., the power over the next 30 minutes to 2 hours. Its training objective is a weighted mixture of the hard loss of the true labels and the distillation loss of the soft labels output by the Teacher model. The loss function is defined as shown in equation (1): (1) Where y is the true label, α∈(0,1) is the distillation weight, used to balance the importance of the two losses, usually set to 0.1-0.5, Ps is the Student model prediction distribution, and Pt is the Teacher model prediction distribution. This represents the hard loss between y and Ps, and KL is the Kullback-Leibler divergence of the soft labels output by the Studen and Teacher models, i.e., the distillation loss. This represents the softmax function, where τ>0 is the distillation coefficient, and τ can be set to 1.0.

[0038] 130. Based on the current weather feature data, the target weather scene is matched.

[0039] In this embodiment of the invention, the database can store historical weather feature data for multiple weather scenarios, including sunny, cloudy, rainy, windy, and dusty weather. These weather scenarios are obtained by clustering historical weather data. By calculating the similarity between the current weather feature data and each historical weather feature data in the database, the weather scenario corresponding to the historical weather feature data with the highest similarity is determined as the target weather scenario.

[0040] 140. Calculate the first accuracy index for each preset model.

[0041] Specifically, based on recent forecast data within a specified timeframe, any evaluation metric for each preset model can be calculated as its primary accuracy metric. Alternatively, multiple evaluation metrics can be calculated for each preset model separately based on recent forecast data within a specified timeframe, standardized, and then weighted and fused to obtain the comprehensive score of each preset model as the primary accuracy metric. These multiple evaluation metrics can include Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Standard Deviation (SD), and Coefficient of Determination (R²), etc. The specified recent timeframe can be set to one month or one week, and the primary accuracy metric is used to characterize the recent forecast accuracy of the preset model.

[0042] 150. Calculate the dynamic weight of each preset model based on the current weather feature matching degree, the historical scores of each preset model in the target weather scenario, and the first accuracy index.

[0043] Among them, the current weather feature matching degree is the similarity between the current weather feature data and the historical weather feature data corresponding to the target weather scene.

[0044] Specifically, in some embodiments, the dynamic weights of the preset model k at time t can be calculated according to the following dynamic weight formula (2). And normalize the results: (2) Where K is the total number of preset models. Based on the weights, ; It is the historical score of the preset model k at the current time t under the target weather scenario. For the first accuracy indicator, The degree of matching with current weather characteristics; , , These are the weighting coefficients for historical scores, the primary accuracy index, and the current weather feature matching degree.

[0045] For example, the current weather feature matching degree Historical scores of each preset model First accuracy indicator The calculation formula is as follows: in, Indicates the current moment. Indicates a historical moment; This is current weather characteristic data. Historical weather feature data for the target weather scenario, The number of samples; The historical true observations of the preset model k, The historical predicted values ​​of the preset model k, This refers to the weather similarity score, which is the degree of matching between current weather features. The normalized value obtained after normalization is obtained using the normalization formula: Historical scores This value is obtained by multiplying the relative errors between the historical predicted values ​​and the historical actual observed values ​​of each sample by the preset model k, using weather similarity as an adjustment coefficient, and then summing the results. std() represents the standard deviation function, and in this example, the standard deviation is used as the primary accuracy indicator.

[0046] As an optional implementation method, the basic weight The calculation method includes: calculating the recent weight coefficient based on the time difference between the current time and historical times; and then weighting the historical scores of each preset model under the target weather scenario using the recent weight coefficient to obtain the base weight. This base weight serves as a set minimum weight threshold to ensure that the dynamic weight of each preset model is not lower than the minimum weight threshold, thus ensuring the participation of each preset model. The recent weight coefficient reflects the time decay effect; that is, the smaller the time difference, the larger the corresponding recent weight coefficient. The recent weight coefficient can be calculated using linear decay, as shown in the formula: ;in Indicates the length of the valid historical window, such as 7 days; It represents the time difference between the current moment and a historical moment.

[0047] 160. Based on the dynamic weights of each preset model, the wind and solar power prediction values ​​of each preset model are weighted and summed to obtain the first fused prediction value.

[0048] 170. Input the target spatiotemporal feature matrix into the Stacking ensemble model to obtain the second fusion prediction value; the Stacking ensemble model is constructed by integrating all the preset models.

[0049] In this invention, the Stacking ensemble model adopts a two-layer structure, namely a base learner and a meta-learner. The base learner includes various preset models. The meta-learner layer adopts a ridge regression model, and its input is a meta-feature matrix constructed from the prediction outputs of multiple base learners through K-fold cross-validation. The meta-learner is used to fuse the prediction outputs of each learner in the meta-feature matrix again and output a second fused prediction value, the contribution of each preset model, the weight allocation, and the 95% confidence interval.

[0050] Specifically, the optional training process of the Stacking ensemble model includes the following steps S1~S6: S1. Define each preset model as a base learner and use the ridge regression model as the meta learner.

[0051] S2. Divide the dataset into training and test sets, and use K-fold cross-validation to train the base learner on the training set and generate the training set meta-feature matrix.

[0052] During training, K-fold cross-validation is used to ensure no data leakage between the meta-features and the original labels, effectively overcoming the overfitting problem. The value of K can be set to 5.

[0053] S3. Train the meta-learner using the meta-feature matrix as input and the original labels of the training set as the target.

[0054] S4. Retrain the base learner using the entire training set, predict the test set, and generate the test set meta-feature matrix.

[0055] S5. Input the test set meta-feature matrix into the meta-learner to obtain the test set prediction results.

[0056] S6. Based on the prediction results of the test set, solve the objective function corresponding to the L2 regularization term to obtain the optimal weights, so as to complete the training of the meta-learner.

[0057] A ridge regression model is used to introduce an L2 regularization term when training the meta-learner. The optimal weights are obtained by solving the corresponding objective function to complete the training of the meta-learner. The L2 regularization coefficient α can be set to 1.0. By introducing an L2 regularization term when training the meta-learner using the ridge regression model, collinearity can be suppressed, overfitting can be prevented, and the generalization ability of the meta-learner can be improved.

[0058] 180. The first fusion prediction value and the second fusion prediction value are weighted and summed to obtain the target fusion prediction value.

[0059] The final prediction result is a weighted combination of the first fusion prediction value and the second fusion prediction value. Generally, the weight of the first fusion prediction value is set to 0.4 and the weight of the second fusion prediction value is set to 0.6 by default.

[0060] In other possible embodiments, a multi-dimensional indicator evaluation system covering technical performance and business benefits can be constructed to comprehensively quantify and continuously optimize the system's predictive capabilities. Specifically, a four-dimensional indicator evaluation system can be constructed, including the following four dimensions: accuracy, stability, efficiency, and business value. Based on this, after executing step 180, the following steps 181-185 can also be executed: 181. Obtain the actual wind and solar power values, and calculate the second accuracy index based on the actual wind and solar power values ​​and the target fusion prediction values.

[0061] In terms of accuracy, multiple evaluation indicators such as MAE, RMSE, and R² can be calculated based on the actual wind and solar power values ​​and the target fused prediction values. These indicators are then fused to calculate a second accuracy indicator, which measures the degree of approximation between the target fused prediction value and the actual wind and solar power values. The second accuracy indicator can use the same or different fusion calculation methods as the first accuracy indicator mentioned above.

[0062] 182. Calculate the prediction error based on the actual wind and solar power values ​​and the target fused prediction values, and calculate the stability index based on the prediction error.

[0063] In terms of stability, the rolling standard deviation or coefficient of variation of the prediction error within a preset sliding window is calculated as a stability index to evaluate the output volatility of the model under different meteorological conditions.

[0064] 183. Calculate the time taken for a single prediction as an efficiency indicator.

[0065] In terms of efficiency, the end-to-end single prediction latency, i.e. the total time from inputting the target spatiotemporal matrix into the preset model to outputting the target fused prediction value, is used as an efficiency indicator to measure whether the system efficiency meets the real-time scheduling requirements.

[0066] 184. Calculate business value indicators based on the target fusion prediction value.

[0067] In terms of business value, the target fusion forecast value is quantified to improve the economic efficiency of downstream systems by combining wind and solar hydrogen production application scenarios. Specifically, the energy saving rate per unit of hydrogen production or the decrease in the power curtailment rate of power plants is used as the business value indicator.

[0068] 185. Monitor the second accuracy indicator, stability indicator, efficiency indicator and business value indicator in real time. When the preset conditions are met, trigger the preset model automatic optimization, dynamic weight allocation strategy update and weather mode adaptive adjustment.

[0069] The aforementioned four-dimensional indicator evaluation system achieves closed-loop self-optimization through prediction result feedback, that is, it realizes automatic tuning of model parameters, dynamic weight allocation strategy updates, and adaptive adjustment of weather patterns. For example, the system automatically evaluates accuracy performance daily, and if the second accuracy indicator exceeds a specified threshold for three consecutive days, it triggers model retraining; monthly, it optimizes parameters, weather scene cluster centers, and feature engineering strategies based on historical data to ensure the system's adaptive capability.

[0070] In summary, by implementing the embodiments of the present invention, a scene-aware dynamic weighting mechanism is constructed, which integrates three indicators: historical performance, recent accuracy, and weather matching degree, to achieve adaptive weight allocation; and an overfitting-preventing Stacking ensemble framework is designed, which combines a weighted average and a meta-learner dual fusion strategy to fully explore the complementarity of KNN, LightGBM, and FGL-LightGBM models.

[0071] Furthermore, a Future-Guided Learning (FGL) mechanism is introduced to guide the training of a long-term Student model using a short-term high-precision Teacher model, which significantly suppresses error accumulation and improves long-term prediction robustness. At the same time, a multi-dimensional evaluation closed loop that includes business value is established to support point prediction, probability prediction and confidence assessment, providing highly reliable and interpretable power input prediction services for wind and solar hydrogen production systems.

[0072] In summary, this invention supports flexible expansion and cross-system interaction, enabling high-precision and robust short-term and ultra-short-term wind and solar power prediction. It meets the diverse needs of wind and solar hydrogen production systems for prediction results, effectively improves grid stability and the economic operation of hydrogen production systems, and significantly enhances the accuracy, robustness, and engineering practicality of short-term (6–24h) and ultra-short-term (0–6h) wind and solar power prediction.

[0073] like Figure 2 As shown in the figure, this embodiment of the invention also discloses a combined evaluation and prediction device for wind and solar power, including a construction unit 201, a prediction unit 202, a matching unit 203, a calculation unit 204, a weighting unit 205, a first integration unit 206, a second integration unit 207, and a fusion unit 208, wherein, Construction unit 201 is used to collect multi-source heterogeneous data in real time and construct a target spatiotemporal feature matrix based on the multi-source heterogeneous data; The prediction unit 202 is used to input the target spatiotemporal feature matrix into each preset model to obtain the wind and solar power prediction values ​​of each preset model; The matching unit 203 is used to match the target weather scene based on the current weather feature data; The calculation unit 204 is used to calculate the first accuracy index of each preset model respectively; The weighting unit 205 is used to calculate the dynamic weight of each preset model based on the current weather feature matching degree and the historical scores and first accuracy index of each preset model in the target weather scenario. The first integration unit 206 is used to perform a weighted summation of the wind and solar power prediction values ​​of each preset model according to the dynamic weights of each preset model to obtain the first fused prediction value. The second integration unit 207 is used to input the target spatiotemporal feature matrix into the Stacking integration model to obtain the second fused prediction value; the Stacking integration model is constructed by integrating all the preset models. The fusion unit 208 is used to perform a weighted summation of the first fusion prediction value and the second fusion prediction value to obtain the target fusion prediction value.

[0074] As an optional implementation, the calculation unit 204 is specifically used to calculate and standardize various evaluation indicators of each preset model based on the prediction data within a specified period of time, and then obtain the comprehensive score of each preset model as the first accuracy indicator through weighted fusion.

[0075] As an optional implementation, the building unit 201 includes: The acquisition subunit is used to collect multi-source heterogeneous data, including measured data from wind and solar power stations, third-party meteorological data, satellite cloud image data, radar observation data, and third-party power prediction data. The preprocessing subunit is used to preprocess multi-source heterogeneous data to obtain multi-source fused data; Construct sub-units to build the target spatiotemporal feature matrix based on multi-source fusion data.

[0076] As an optional implementation, the apparatus further includes: The acquisition unit is used to obtain the actual wind and solar power value after the fusion unit 208 performs a weighted summation of the first fusion prediction value and the second fusion prediction value to obtain the target fusion prediction value. The accuracy evaluation unit is used to calculate the second accuracy index based on the actual wind and solar power values ​​and the target fusion prediction values; The stability evaluation unit is used to calculate the prediction error based on the actual wind and solar power values ​​and the target fused prediction values, and to calculate the stability index based on the prediction error. The efficiency evaluation unit is used to calculate the time taken for a single prediction as an efficiency indicator. The value evaluation unit is used to calculate business value indicators based on the target fusion prediction value; The monitoring unit is used to monitor the second accuracy indicator, stability indicator, efficiency indicator and business value indicator in real time. When the preset conditions are met, it triggers the automatic optimization of each preset model, the update of the dynamic weight allocation strategy and the adaptive adjustment of the weather mode.

[0077] like Figure 3 As shown, this embodiment of the invention also discloses an electronic device, including a memory 301 storing executable program code and a processor 302 coupled to the memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute the combined evaluation and prediction method of wind and solar power described in the above embodiments.

[0078] like Figure 4 As shown in the illustration, this invention also discloses a computer device. This computer device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. The processor in this computer design provides computational and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores relevant data for the combined evaluation and prediction method of wind and solar power. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the combined evaluation and prediction method of wind and solar power described in the above embodiments.

[0079] This invention also discloses a computer-readable storage medium storing a computer program that causes a computer to execute the combined evaluation and prediction method for wind and solar power described in the above embodiments. The storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0080] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0081] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0082] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A combined evaluation and prediction method for wind and solar power, characterized in that, include: Real-time acquisition of multi-source heterogeneous data, and construction of a target spatiotemporal feature matrix based on the multi-source heterogeneous data; The target spatiotemporal feature matrix is ​​input into each preset model to obtain the wind and solar power prediction values ​​of each preset model; Based on current weather characteristic data, the target weather scenario is matched and obtained; Calculate the first accuracy index for each of the preset models; Based on the current weather feature matching degree and the historical scores and first accuracy index of each preset model in the target weather scenario, the dynamic weight of each preset model is calculated. Based on the dynamic weights of each preset model, the wind and solar power prediction values ​​of each preset model are weighted and summed to obtain the first fused prediction value. The target spatiotemporal feature matrix is ​​input into the Stacking ensemble model to obtain a second fused prediction value; the Stacking ensemble model is constructed by integrating all the preset models. The first fusion prediction value and the second fusion prediction value are weighted and summed to obtain the target fusion prediction value.

2. The combined evaluation and prediction method for wind and solar power according to claim 1, characterized in that, Real-time acquisition of multi-source heterogeneous data, and construction of a target spatiotemporal feature matrix based on the multi-source heterogeneous data, including: Collect multi-source heterogeneous data, including measured data from wind and solar power stations, third-party meteorological data, satellite cloud image data, radar observation data, and third-party power prediction data; The multi-source heterogeneous data is preprocessed to obtain multi-source fused data; Based on the multi-source fusion data, a target spatiotemporal feature matrix is ​​constructed.

3. The combined evaluation and prediction method for wind and solar power according to claim 1, characterized in that, Calculate the first accuracy index for each of the preset models, including: Based on the prediction data within a specified recent period, various evaluation indicators for each preset model are calculated and standardized. Then, the comprehensive score of each preset model is obtained by weighted fusion and used as the first accuracy indicator.

4. The combined evaluation and prediction method for wind and solar power according to claim 1, characterized in that, Based on the current weather feature matching degree and the historical scores and first accuracy index of each of the preset models in the target weather scenario, the dynamic weights of each of the preset models are calculated, including: The dynamic weights of the preset model k at time t are calculated using the following dynamic weight formula. And normalize the results: Where K is the total number of preset models. Based on weights, ; It is the historical score of the preset model k at the current time t under the target weather scenario. For the first accuracy indicator, The degree of matching with current weather characteristics; , , These are the weighting coefficients for the historical score, the first accuracy index, and the current weather feature matching degree, respectively.

5. The combined evaluation and prediction method for wind and solar power according to any one of claims 1 to 4, characterized in that, After weighted summing of the first fused prediction value and the second fused prediction value to obtain the target fused prediction value, the method further includes: Obtain the actual wind and solar power value, and calculate the second accuracy index based on the actual wind and solar power value and the target fused prediction value; The prediction error is calculated based on the actual wind and solar power value and the target fused prediction value, and the stability index is calculated based on the prediction error. The time taken for a single prediction is used as an efficiency indicator. Calculate the business value index based on the target fusion prediction value; The system monitors the second accuracy indicator, the stability indicator, the efficiency indicator, and the business value indicator in real time. When preset conditions are met, it triggers automatic optimization of each preset model, dynamic weight allocation strategy update, and weather mode adaptive adjustment.

6. A combined evaluation and prediction device for wind and solar power, characterized in that, include: A construction unit is used to collect multi-source heterogeneous data in real time and construct a target spatiotemporal feature matrix based on the multi-source heterogeneous data; The prediction unit is used to input the target spatiotemporal feature matrix into each preset model to obtain the wind and solar power prediction values ​​of each preset model; The matching unit is used to match the target weather scene based on the current weather feature data; A calculation unit is used to calculate the first accuracy index of each of the preset models respectively; The weighting unit is used to calculate the dynamic weight of each preset model based on the current weather feature matching degree and the historical scores and first accuracy index of each preset model in the target weather scenario. The first integration unit is used to perform a weighted summation of the wind and solar power prediction values ​​of each preset model according to the dynamic weights of each preset model to obtain a first fused prediction value. The second integration unit is used to input the target spatiotemporal feature matrix into the Stacking integration model to obtain a second fused prediction value; the Stacking integration model is constructed by integrating all the preset models. The fusion unit is used to perform a weighted summation of the first fused prediction value and the second fused prediction value to obtain the target fused prediction value.

7. The combined evaluation and prediction device for wind and solar power according to claim 6, characterized in that, The building unit includes: The acquisition subunit is used to acquire multi-source heterogeneous data, including measured data from wind and solar power stations, third-party meteorological data, satellite cloud image data, radar observation data, and third-party power prediction data. The preprocessing subunit is used to preprocess the multi-source heterogeneous data to obtain multi-source fused data; A sub-unit is constructed to construct a target spatiotemporal feature matrix based on the multi-source fusion data.

8. The combined evaluation and prediction device for wind and solar power according to claim 6 or 7, characterized in that, The device further includes: The acquisition unit is used to acquire the actual wind and solar power value after the fusion unit performs a weighted summation of the first fusion prediction value and the second fusion prediction value to obtain the target fusion prediction value. An accuracy evaluation unit is used to calculate a second accuracy index based on the actual wind and solar power value and the target fused prediction value; The stability evaluation unit is used to calculate the prediction error based on the actual wind and solar power value and the target fused prediction value, and to calculate the stability index based on the prediction error. The efficiency evaluation unit is used to calculate the time taken for a single prediction as an efficiency indicator. The value evaluation unit is used to calculate business value indicators based on the target fusion prediction value; The monitoring unit is used to monitor the second accuracy indicator, the stability indicator, the efficiency indicator, and the business value indicator in real time. When preset conditions are met, it triggers the automatic optimization of each preset model, the dynamic weight allocation strategy update, and the weather mode adaptive adjustment.

9. An electronic device, characterized in that, It includes a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the combined evaluation and prediction method for wind and solar power as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to perform the combined evaluation and prediction method for wind and solar power as described in any one of claims 1 to 5.

Images