Multi-source data fusion wind and light generating capacity intelligent prediction method and system

Abstract

The invention relates to the technical field of wind and light power generation, and particularly discloses a multi-source data fusion wind and light power generation capacity intelligent prediction method and system, and the method achieves the high-precision intelligent prediction of the new energy power generation capacity through three core steps. Power generation performance, environment and equipment state multi-source data are collected and fused in real time, intelligent fusion adjustment is carried out on the basis of preprocessing, and the accuracy and integrity of input data are ensured; the method comprises the following steps: constructing space-time correlation characteristics by utilizing multi-source data, modeling layout of wind power plant and photovoltaic power station equipment into a graph structure, and effectively capturing space interaction among the equipment; after the features are input into a prediction model, the model can intelligently adjust parameters based on error indexes, and the prediction precision is continuously optimized; closed-loop feedback is formed by comparing the predicted value with the actual generating capacity, model parameter adjustment is triggered again, and it is ensured that the predicted result is reliable for a long time.

Description

Technical Field

[0001] This invention relates to the field of wind and solar power generation technology, specifically to a method and system for intelligent prediction of wind and solar power generation based on multi-source data fusion. Background Technology

[0002] This invention relates to a multi-source data fusion-based intelligent prediction method and system for wind and solar power generation. It aggregates data from various sources, including numerical weather prediction data, historical meteorological observation data, geographic information data, historical power output data from wind farms / solar power plants, and real-time operational status data. The system employs techniques such as data cleaning, spatiotemporal alignment, and feature engineering to fuse and improve the quality of this heterogeneous data. Based on the fused high-quality dataset, a hybrid intelligent prediction model integrating physical mechanisms and data-driven approaches is constructed and trained. This model can capture the complex nonlinear spatiotemporal variations of wind and solar energy. The trained model receives real-time fused data input and predicts short-term or ultra-short-term wind and solar power generation. After outputting the prediction results, the system performs error analysis, visualizes the results, and dynamically updates and optimizes the model parameters based on feedback information, forming a continuously improving closed-loop prediction process.

[0003] For example, Chinese invention patent application CN119834203A discloses a method for predicting the power generation of a wind-solar hybrid power generation system. The method includes: collecting data to obtain historical data; preprocessing the historical data to obtain preprocessed data; discretizing the preprocessed data to obtain discrete data; constructing an HMMC model; training the HMMC model using a historical dataset constructed from the historical data; collecting real-time data; calculating the heat index based on the real-time data; and using the HMMC model to predict power based on the heat index.

[0004] For example, Chinese invention patent application CN118213970A discloses a method and system for predicting distributed photovoltaic power generation. The method includes: collecting historical power generation data of a target area and preprocessing the historical power generation data; the historical power generation data includes historical photovoltaic power generation and corresponding power generation time; dividing the daily time into several time periods, and statistically analyzing the power generation of the historical power generation data in each time period to obtain a photovoltaic power generation set; obtaining historical weather forecast information of the target area; the historical weather forecast information includes solar radiation intensity, temperature, cloudy / rainy weather, wind speed, and seasonal changes; processing the historical weather forecast information, and marking the historical weather forecast information according to the daily time periods of the photovoltaic power generation set to obtain a power generation association set.

[0005] The above-mentioned technology has at least the following technical problems: In the data fusion process, raw data is first collected directly from multiple heterogeneous data sources; then preliminary data cleaning and transformation are performed; then a specific fusion algorithm is applied to merge the data; finally, the fusion result is output. This process has significant defects. The fusion algorithm itself often lacks adaptability and has high computational complexity. In addition, the adjustment and optimization links in the process are weak or unsystematic, which ultimately leads to low efficiency of the entire data fusion process. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent prediction of wind and solar power generation based on multi-source data fusion, which can effectively solve the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a method for intelligent prediction of wind and solar power generation through multi-source data fusion, comprising: Step 1: Real-time collection of power generation performance parameters, environmental data, and equipment operating status data monitored by the regional new energy control center; preprocessing of the power generation performance parameters, environmental data, and equipment operating status data; fusion of the preprocessed power generation performance parameters, environmental data, and equipment operating status data; collection and analysis of data fusion process parameters; determination of whether to adjust the data fusion process; thereby obtaining an accurate multi-source data set; Step 2: Using the multi-source data set… Spatiotemporal correlation features are constructed and input into the power generation prediction model. The prediction model outputs the predicted power generation value. The model error index is collected and analyzed to determine whether the parameters of the prediction model need to be adjusted, thereby intelligently optimizing the accuracy of the power generation prediction model. The spatiotemporal correlation features represent the equipment layout of wind farms and photovoltaic power stations as a graph structure, where nodes represent wind turbines or photovoltaic arrays and edges represent the spatial correlation between equipment. Step 3: The actual total power generation is collected and compared with the predicted power generation value output by the prediction model. Based on the comparison results, it is determined whether the parameters of the prediction model need to be readjusted, thereby optimizing the results of intelligent prediction of wind and solar power generation in the prediction model.

[0008] The second aspect of the present invention provides an intelligent prediction system for wind and solar power generation based on multi-source data fusion, comprising: a multi-source data fusion module, a prediction model module, an intelligent optimization module, and a database.

[0009] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) This invention provides a method and system for intelligent prediction of wind and solar power generation through multi-source data fusion. This process achieves high-precision intelligent prediction of new energy power generation through three core steps. First, multi-source data on power generation performance, environment, and equipment status are collected and fused in real time. Intelligent fusion adjustment is performed on the basis of preprocessing to ensure the accuracy and completeness of the input data. Second, spatiotemporal correlation features are constructed using multi-source data to model the layout of wind farms and photovoltaic power stations as a graph structure, effectively capturing the spatial interaction between equipment. After these features are input into the prediction model, the model can intelligently adjust parameters based on error indicators to continuously optimize the prediction accuracy. Finally, by comparing the predicted value with the actual power generation, a closed-loop feedback is formed, which triggers the adjustment of model parameters again to ensure the long-term reliability of wind and solar power generation prediction results.

[0010] (2) This invention effectively eliminates noise, outliers, and missing values ​​by preprocessing power generation performance, environmental data, and equipment status, ensuring the reliability and consistency of the data. By fusing these three types of heterogeneous data, data silos are broken down, and a more comprehensive multi-dimensional view reflecting the actual operating status of the new energy power generation system is constructed. Real-time monitoring and dynamic adjustment of the fusion process, through continuous and intelligent quality optimization and adaptive adjustment of the key basic data input to the wind and solar prediction model, ensures that the data foundation upon which the prediction model relies is always the most accurate, up-to-date, and most closely aligned with the current actual operating status.

[0011] (3) This invention utilizes the obtained high-quality fused data to construct spatiotemporal correlation features, which can more effectively capture the complex spatiotemporal dependencies affecting wind and solar power generation, significantly enhancing the model's ability to understand power generation change patterns. After inputting the constructed features into the prediction model, the model error index is collected and analyzed in real time, and the model parameters are intelligently adjusted according to the analysis results, realizing dynamic closed-loop optimization of the prediction model. This greatly improves the accuracy of wind and solar power generation prediction and the robustness of the model, enabling it to adapt to different weather conditions, seasonal changes, and equipment status fluctuations.

[0012] (4) This invention enables continuous improvement of the model. The predicted power generation output by the prediction model is compared with the actual measured total power generation in real time or periodically to directly evaluate the model's performance in the real world. Based on the comparison results, it is determined whether the model parameters need to be readjusted, forming a closed-loop mechanism of prediction, verification, adjustment, and re-prediction. This mechanism ensures that the prediction model does not stagnate but can continuously learn new operating data and pattern changes, constantly narrowing the gap between the predicted wind and solar power generation values ​​and the actual wind and solar power generation values. Attached Figure Description

[0013] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0015] Figure 2 This is a schematic diagram of the system module connections of the present invention.

[0016] Figure 3 This is a flowchart illustrating the adjustment process of the data fusion process in this invention.

[0017] Figure 4 This is a flowchart of the secondary adjustment process of the data fusion process of the present invention.

[0018] Figure 5 This is a flowchart of the adjustment process for the prediction model of the present invention.

[0019] Figure 6 This is a flowchart of the secondary adjustment process of the prediction model of the present invention.

[0020] Figure 7 This is a flowchart illustrating the relationship between the power generation difference and the actual power generation according to the present invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0022] Reference Figure 1As shown, the first aspect of this invention provides a method for intelligent prediction of wind and solar power generation through multi-source data fusion, comprising: Step 1, real-time collection of power generation performance parameters, environmental data, and equipment operating status data monitored by a regional new energy control center; preprocessing of the power generation performance parameters, environmental data, and equipment operating status data respectively; fusing the preprocessed power generation performance parameters, environmental data, and equipment operating status data; collecting and analyzing parameters of the data fusion process; determining whether to adjust the data fusion process; thereby obtaining an accurate multi-source data set; Step 2, constructing spatiotemporal correlation features through the multi-source data set, and... The data is input into the power generation prediction model, which outputs the predicted power generation value. The model error index is collected and analyzed to determine whether the parameters of the prediction model need to be adjusted, thereby intelligently optimizing the accuracy of the power generation prediction model. The spatiotemporal correlation feature represents the modeling of the equipment layout of wind farms and photovoltaic power stations as a graph structure, where nodes represent wind turbines or photovoltaic arrays and edges represent the spatial correlation between equipment. Step 3: Collect the actual total power generation and compare it with the predicted power generation value output by the prediction model. Based on the comparison results, determine whether to readjust the prediction model parameters to optimize the intelligent prediction results of wind and solar power generation in the prediction model.

[0023] Figure 3 This is a flowchart illustrating the adjustment process of the data fusion process in this invention. The reliability of the fusion result is determined by calculating an anomaly index. If the anomaly index is not greater than the anomaly threshold, the fusion result is stored and further processing is performed. If the anomaly index is greater than the anomaly threshold, the fusion is cancelled, the original data is saved, and the fusion process is attempted to be adjusted. After adjustment, the process is checked again. If successful, the normal process continues; if it still fails, a decision must be made regarding whether to perform a second adjustment. Figure 4 This is a flowchart of the secondary adjustment process of the data fusion process in this invention. The process determines whether the data fusion process has undergone secondary adjustment. The core step is to check whether the abnormal index of the adjusted data fusion process does not exceed the abnormal threshold. If the abnormal index does not exceed the abnormal threshold, the multi-source data is stored in the normal region and spatiotemporal correlation features are constructed. If the abnormal index exceeds the abnormal threshold, the preprocessed data is first stored in the abnormal region, then the fusion process is adjusted, a new abnormal index is calculated, and the process is judged again. If the abnormal index of the readjusted data fusion process does not exceed the abnormal threshold, normal storage and feature construction are performed again. If the abnormal index of the readjusted data fusion process exceeds the abnormal threshold, an early warning is triggered.

[0024] Specifically, the process involves determining whether to adjust the data fusion process. The specific determination process is as follows: collect and analyze the parameters of the data fusion process to obtain the data fusion process anomaly index.

[0025] Power generation performance parameters reflect the real-time power generation capacity and efficiency of power generation equipment.

[0026] Environmental data reflects external environmental factors that affect power generation performance and the operation of power generation equipment.

[0027] Equipment operating status data reflects the operating status of power generation equipment.

[0028] The abnormality index of the data fusion process is compared with the abnormality threshold of the data fusion process. If the abnormality index of the data fusion process is less than or equal to the abnormality threshold of the data fusion process, the multi-source data set is stored in the normal area of ​​the database, and the spatiotemporal correlation features of the multi-source data set are constructed.

[0029] The multi-source data set is a comprehensive dataset formed by preprocessing, merging and optimizing the power generation performance parameters, environmental data and equipment operation status data monitored by the regional new energy centralized control center.

[0030] If the data fusion process anomaly index exceeds the data fusion process anomaly threshold, the current fusion is cancelled, and the preprocessed power generation performance parameters, environmental data, and equipment operating status data are stored in the anomaly area of ​​the database. Then, based on the data fusion process anomaly index, the observation noise covariance matrix in the data fusion algorithm is increased. The preprocessed power generation performance parameters, environmental data, and equipment operating status data stored in the database anomaly area are fused, and the data fusion process parameters are re-collected and analyzed to obtain the adjusted data fusion process anomaly index. Finally, it is determined whether to perform a second adjustment to the data fusion process.

[0031] The normal zone serves as the standard data warehouse for system operation, while the abnormal zone acts as an isolation buffer and repair workspace to ensure this standard remains uncontaminated. Their clearly defined roles work together to build a more robust, reliable, and maintainable data fusion and management system.

[0032] The data fusion process anomaly threshold is used to characterize the maximum upper limit of the data fusion process anomaly index. It is determined by analyzing the anomaly index of multiple batches of normal data fusion processes generated by the regional new energy central control center during its historical normal operation, combined with the experience of domain experts and the system's reliability requirements, and is stored in the database.

[0033] Based on the increase of the observation noise covariance matrix in the data fusion algorithm by the anomaly index in the data fusion process, the anomaly index in the data fusion process is divided into different level intervals in the database. Each level interval corresponds to a different increase value of the observation noise covariance matrix in the data fusion algorithm. The observation noise covariance matrix in the current data fusion algorithm is added to the increase value of the observation noise covariance matrix in the data fusion algorithm to obtain the observation noise covariance matrix in the next data fusion algorithm.

[0034] Increasing the observation noise covariance matrix in the data fusion algorithm by increasing the anomaly index during the data fusion process can effectively reduce the interference of outlier data on the fusion results and improve the robustness of the algorithm. In intelligent prediction of wind and solar power generation, multi-source observation data from meteorological stations, historical power, numerical weather prediction, and even data from adjacent stations may introduce outliers due to factors such as sensor failure, communication interference, or local extreme weather. Adjusting the noise covariance matrix makes the fusion algorithm more inclined to rely on the predicted values ​​of the system model rather than potentially outlier observation data, thereby suppressing the negative impact of outliers on the fusion results. By dynamically adjusting the noise covariance matrix, the algorithm can adaptively cope with the data quality fluctuations caused by the inherent intermittency and volatility of wind and solar data sources, improving the stability of the fusion results. This adjustment can indirectly reduce the anomaly index of the data fusion process because the contribution of outlier data is weakened, reducing inconsistencies in the fusion process. This may cause the second-calculated anomaly index of the data fusion process to fall below the anomaly threshold, avoiding unnecessary repeated adjustments. For wind and solar forecasting systems, this means that even if some data sources are temporarily unreliable, the system can maintain relatively stable and accurate forecast outputs, reducing forecast jumps or deviations caused by data anomalies.

[0035] Power generation performance parameters include: active power, reactive power, voltage, current, frequency, conversion efficiency, capacity factor, and performance ratio.

[0036] Environmental data includes: light intensity, wind speed, wind direction, ambient temperature, humidity, air pressure, and local wind speed turbulence.

[0037] Equipment operating status data includes: vibration spectrum, bearing temperature, gearbox oil quality, blade pitch angle, insulation resistance, winding temperature, circuit breaker status, and grounding resistance, etc.

[0038] Data preprocessing includes: cleaning real-time collected wind and solar power operating parameters and equipment status data, removing outliers and filling in missing data, then standardizing the data to unify the dimensions and sampling frequency of multi-source data, and analyzing equipment performance trends through time-series decomposition analysis.

[0039] To unify the units and sampling frequencies of multi-source data, the process involves standardizing all data with different units and ranges and aligning all time series data to the same timestamp grid with the same sampling interval.

[0040] By collecting and analyzing power generation performance parameters, environmental data, and equipment operating status data in real time, a multi-source dataset is merged to calculate an anomaly index for the data fusion process. This allows for a comprehensive and objective assessment of the quality and reliability of the data fusion. When the anomaly index exceeds the anomaly threshold, the system automatically cancels the fusion and stores the original data in the anomaly region. Simultaneously, the fusion algorithm is optimized by increasing the observation noise covariance matrix, effectively reducing the interference of abnormal data on the fusion results and improving the accuracy and robustness of the data fusion. Secondary adjustments and re-fusion further ensure the effective utilization of data and avoid data waste. The entire process not only achieves precise monitoring of the operating status of new energy power generation equipment but also provides a high-quality data foundation for subsequent fault diagnosis and performance optimization, thereby improving the operational efficiency and reliability of the new energy control center.

[0041] Specifically, the process of determining whether to perform secondary adjustments to the data fusion process is as follows: if the abnormality index of the adjusted data fusion process is less than or equal to the abnormality threshold of the data fusion process, then the multi-source data set is stored in the normal area of ​​the database, and spatiotemporal correlation features are constructed through the multi-source data set.

[0042] If the adjusted data fusion process anomaly index is greater than the data fusion process anomaly threshold, the preprocessed power generation performance parameters, environmental data, and equipment operating status data are stored again in the anomaly area of ​​the database. Based on the adjusted data fusion process anomaly index, the observation noise covariance matrix in the data fusion algorithm is increased and the buffer size inside the data fusion algorithm is increased. The power generation performance parameters, environmental data, and equipment operating status data in the anomaly area of ​​the database are then re-fused to obtain the readjusted data fusion process anomaly index.

[0043] If the abnormal index of the readjusted data fusion process is less than or equal to the abnormal threshold of the data fusion process, the multi-source data set will be stored in the normal area of ​​the database, and spatiotemporal correlation features will be constructed through the multi-source data set.

[0044] Constructing spatiotemporal correlation features from multi-source datasets refers to integrating heterogeneous data from different sources, aligning and fusing these data within a unified geospatial framework and temporal benchmark, and using specific algorithms or models to mine and extract key indicators or patterns that characterize the state and dynamic changes of a target object or phenomenon at a specific geographical location and time point. Its core lies in revealing the implicit spatiotemporal dependencies, interactions, and evolutionary trends among the data, thereby forming a comprehensive feature expression that transcends the perspective of a single data source and contains spatiotemporal contextual information, providing a deeper and more comprehensive foundation for subsequent analysis, prediction, or decision-making.

[0045] If the abnormal index of the readjusted data fusion process exceeds the abnormal threshold of the data fusion process, an early warning will be triggered.

[0046] The system triggers an alarm, providing both sound and light alerts to remind staff.

[0047] In the data fusion adjustment process, increasing the observation noise covariance matrix has a more significant impact than increasing the buffer size. The noise covariance matrix directly controls the fusion algorithm's trust in the observation data; increasing it reduces the sensitivity to outliers, thus more directly suppressing the anomaly index in the data fusion process. In contrast, increasing the buffer size primarily affects the fusion result indirectly by smoothing historical data, and its effect is relatively weaker. Therefore, if a rapid adjustment of the anomaly index in the data fusion process is needed, the noise covariance matrix should be adjusted first; if the anomalies are time-dependent, buffer optimization can be used as a supplement.

[0048] Based on the adjusted anomaly index of the data fusion process, the observation noise covariance matrix in the data fusion algorithm is increased, and the buffer size within the data fusion algorithm is increased. The anomaly index of the data fusion process is divided into different level intervals in the database. Each level interval corresponds to a different increase in the observation noise covariance matrix in the data fusion algorithm. The increased value of the observation noise covariance matrix in the current data fusion algorithm is added to the increased value of the observation noise covariance matrix in the next data fusion algorithm. Similarly, the anomaly index of the data fusion process is divided into different level intervals in the database. Each level interval corresponds to a different increase in the buffer size within the data fusion algorithm. The increased value of the increased buffer size in the current data fusion algorithm is added to the increased value of the buffer size in the next data fusion algorithm.

[0049] An adjustment strategy based on increasing the observation noise covariance matrix and buffer size after adjusting the anomaly index of the data fusion process can effectively improve the fault tolerance and stability of the data fusion algorithm for anomalous data. Increasing the observation noise covariance matrix can reduce the algorithm's sensitivity to the current adjusted anomaly index of the data fusion process, preventing the fusion result from deviating from the true state due to individual outliers; while increasing the buffer size can smooth out anomalous fluctuations by introducing more historical or neighboring data, improving the robustness of the fusion result. The combined effect of this dual adjustment can significantly suppress further increases in the anomaly index of the data fusion process, and even reduce it below the anomaly threshold of the data fusion process through re-fusion, thereby avoiding unnecessary warnings and ensuring the reliability of the fusion result. If adjusted properly, the adjusted anomaly index of the data fusion process may converge rapidly due to the enhanced adaptability of the algorithm, eventually meeting the conditions for normal fusion.

[0050] By dynamically adjusting the observation noise covariance matrix and buffer size of the data fusion algorithm, the system can adaptively optimize the fusion process, effectively reducing interference from abnormal data and improving the accuracy and stability of the fusion results. When the abnormality index of the adjusted data fusion process exceeds the abnormality threshold, the system adopts a tiered processing strategy: first, a secondary fusion optimization is performed; if the threshold is still not met, a sound-activated early warning is triggered to ensure that abnormal data is processed in a timely manner and to prevent erroneous data from entering the normal database. This multi-level adjustment mechanism not only enhances the robustness of data fusion but also improves the data reliability of the new energy control center, providing more accurate data support for the status monitoring, fault diagnosis, and performance optimization of power generation equipment. Simultaneously, the early warning mechanism alerts maintenance personnel to intervene, further ensuring the safe and stable operation of the system.

[0051] Furthermore, the parameters of the data fusion process are collected and analyzed. The specific analysis process includes the noise covariance, coverage, and error covariance of the data fusion process.

[0052] The noise covariance of the data fusion process can be calculated by statistically analyzing the output data of sensors (temperature sensors, radar, cameras, IMUs, etc.) under steady-state conditions. That is, multiple sets of data are collected when there is no target change, and their variance matrix is ​​calculated to characterize the noise characteristics. The coverage of the data fusion process needs to be evaluated in combination with spatial and information dimensions and quantified by the overlap rate of the sensor detection range. The error covariance of the data fusion process is usually calculated based on the residual between the fusion result and the true value, and the covariance matrix of the residual is statistically obtained through multiple experiments.

[0053] The noise covariance matrix is ​​used to quantify the statistical characteristics of sensor measurement noise. It is calculated as follows: under steady-state conditions, multiple sets of data output from the sensors are collected, their mean values ​​are calculated, and then the noise covariance matrix is ​​obtained using the covariance formula. The diagonal elements of this matrix represent the noise variance of each sensor, while the off-diagonal elements reflect the correlation between the noise levels of different sensors.

[0054] Coverage measures the spatial or informational redundancy of data fusion, which can be quantified by the overlap rate of sensor detection ranges. In terms of information dimension, it assesses the feature complementarity when different sensors observe the same target, such as the proportion of jointly observed features to the total features.

[0055] The error covariance matrix reflects the accuracy of the fusion result, and its calculation is based on the residual between the fused estimate and the true value. The specific steps are as follows: conduct multiple independent experiments, record the error between the fused estimate and the true value each time, and then calculate the error covariance matrix using the formula. If the true value cannot be directly obtained, cross-validation or confidence-weighted methods can be used to indirectly estimate the error, ensuring the basis for optimizing the fusion algorithm.

[0056] Conducting multiple independent experiments means repeatedly executing the fusion algorithm multiple times, each execution based on newly generated, statistically independent input data. Each execution produces a fusion result and a corresponding error vector. After collecting a sufficient number of these independent error vectors, statistical methods can be used to estimate the error covariance matrix of the fusion result. This matrix quantifies the uncertainty distribution of the fusion estimate around the true value and the correlation between the errors of each state component, serving as the core basis for evaluating and optimizing the accuracy of the fusion algorithm.

[0057] The calculation of the anomaly index of the data fusion process is based on the proportional relationship between the noise covariance of the data fusion process and the boundary noise covariance, the proportional relationship between the boundary coverage and the coverage of the data fusion process, and the proportional relationship between the error covariance of the data fusion process and the boundary error covariance. Specifically, the contrast of the boundary noise covariance, boundary coverage, and boundary error covariance is used as a benchmark. The proportional relationships between the noise covariance, coverage, and error covariance of the data fusion process and the corresponding boundary values ​​are calculated respectively. Different measurement ratios are assigned according to the degree of influence of each proportional relationship on the anomaly index of the data fusion process, and the results are summarized to finally obtain the anomaly index of the data fusion process.

[0058] The data fusion process anomaly index is used to quantify the overall deviation between the data fusion result and the expected normal state.

[0059] ; SJH is the anomaly index of the data fusion process, SRZ is the noise covariance of the data fusion process, SRF is the coverage of the data fusion process, SRW is the error covariance of the data fusion process, SRZ_L is the preset boundary noise covariance in the database, SRF_L is the preset boundary coverage in the database, SRW_L is the preset boundary error covariance in the database, A1 is the measurement ratio corresponding to the preset noise covariance in the database, A2 is the measurement ratio corresponding to the preset coverage in the database, and A3 is the measurement ratio corresponding to the preset error covariance in the database.

[0060] Define the noise covariance to characterize the upper limit of the noise covariance in the data fusion process; define the coverage to characterize the lower limit of the coverage in the data fusion process; define the error covariance to characterize the upper limit of the error covariance in the data fusion process.

[0061] The quality and reliability of the data fusion process are jointly determined by three closely related parameters: noise covariance, coverage, and error covariance. Noise covariance measures the degree of noise interference in the fusion process; coverage characterizes the completeness of the data source's coverage of the target area; and error covariance directly reflects the accuracy deviation of the fusion result. These three parameters are interconnected: insufficient coverage reduces the system's ability to distinguish between signal and noise, indirectly amplifying the impact of noise and directly increasing the error covariance; high noise covariance directly contaminates the fusion result, significantly increasing the error covariance; ultimately, the error covariance, as a core indicator of fusion accuracy, is a direct result of the combined effects of insufficient coverage and noise interference, forming a mutually dependent triangular relationship. The data fusion process anomaly index comprehensively assesses process anomalies by calculating the deviations of these three parameters from their respective preset benchmark values: noise covariance exceeding limits, coverage falling below standard, or error covariance exceeding limits all contribute to an increased index. A higher index indicates a more severe interconnected set of problems—noise interference, missing information, and result accuracy deviations—that collectively lead to significant deviations from expectations in the fusion process, resulting in lower reliability and effectiveness of the data fusion.

[0062] The metric ratio corresponding to the noise covariance in the data fusion process represents the relative proportion between the noise covariance in the data fusion process and the predefined noise covariance, and its impact on the anomaly index of the data fusion process. This ratio quantifies the contribution weight of the noise covariance in the overall anomaly assessment, reflecting the key role of data noise fluctuations in detecting deviations in the fusion results. The metric ratio corresponding to the coverage in the data fusion process represents the relative proportion between the coverage in the data fusion process and the predefined coverage, and its impact on the anomaly index. This value determines the importance of coverage in the comprehensive assessment and is used to measure the sensitivity of data coverage to the consistency of the fusion results. The metric ratio corresponding to the error covariance in the data fusion process represents the relative proportion between the error covariance in the data fusion process and the predefined error covariance, and its impact on the anomaly index. It is used to assess the diagnostic value of error fluctuations in fusion anomaly detection, highlighting the significant impact of error stability on the overall quantification results.

[0063] The database stores preset evaluation benchmark parameters for the data fusion process, including the boundary noise covariance, boundary coverage, and boundary error covariance. These parameters are dynamically bound to key indicators of the real-time data fusion process and preset anomaly evaluation benchmarks through a structured parameter mapping table, forming a complete anomaly evaluation system for the data fusion process. When it is necessary to calculate the anomaly index of a specific data fusion process, the system extracts the data fusion process parameters, including the noise covariance, coverage, and error covariance. Subsequently, based on a preset rule base, historical data similarity matching, or statistical analysis model, these indicators are compared or calculated with the parameter mapping table in the database. Finally, the system dynamically outputs the boundary noise covariance, boundary coverage, and boundary error covariance applicable to the fusion process, along with the corresponding weight coefficients. Among them, the A1, A2, and A3 measurement ratios are used as weight coefficients, and their values ​​range from 0 to 1.

[0064] Specifically, the model error indicators are collected and analyzed. The specific analysis process is as follows: The model error indicators include the mean absolute error of the power generation prediction model, the root mean square error of the power generation prediction model, and the mean square error of the power generation prediction model.

[0065] The error indices of power generation prediction models should be determined using the following steps: First, collect historical model predictions and corresponding historical actual observations, ensuring the data timeframe is consistent and complete. Calculate the historical model mean absolute error (MAE): sum the absolute differences between the historical model prediction and the corresponding historical actual value at each time point, and then average the sums to reflect the overall magnitude of the prediction deviation. Calculate the power generation prediction model mean square error (MSE): square each error and then average the results. This amplifies the impact of larger errors and is often used for gradient optimization. Calculate the power generation prediction model root mean square error (RMSE): take the square root of the MAE of the power generation prediction model; its dimensions are consistent with the actual values, and it is used to assess the stability of the prediction.

[0066] Obtain the abnormal final value of the data fusion process and match the abnormal increment of the data fusion process from the database.

[0067] The data fusion process anomaly index is dynamically generated by analyzing the latest data generated in real time during the data fusion process and using specific calculation methods. This index provides a clear and quantitative representation of the current health status and degree of anomalies in the data fusion process. The data fusion process anomaly index is continuously updated over time and with the arrival of new data.

[0068] The calculation of outliers in the prediction model evaluation indicators is based on the proportional relationships between the mean absolute error and the defined mean absolute error of the power generation prediction model, the root mean square error and the defined root mean square error of the power generation prediction model, and the mean square error and the defined mean square error of the power generation prediction model. The specific process is as follows: using the defined mean absolute error, defined root mean square error, and defined mean square error as benchmarks, the proportional relationships between the mean absolute error, root mean square error, and mean square error of the power generation prediction model and their corresponding defined values ​​are calculated. Different measurement ratios are assigned to the degree of influence of the outliers in the prediction model evaluation indicators according to each proportional relationship, and the results are summarized. Then, the outlier increment from the data fusion process is added to obtain the outliers in the prediction model evaluation indicators.

[0069] Outliers in the prediction model evaluation metrics are used to quantify the degree of abnormality in the overall performance of the prediction model.

[0070] ; PGZ represents outliers in the prediction model evaluation index; MPW represents the mean absolute error of the power generation prediction model; MJC represents the root mean square error of the power generation prediction model; MFC represents the mean square error of the power generation prediction model; H represents the abnormal increment during the data fusion process; MPW_L represents the preset defined mean absolute error in the database; MJC_L represents the preset defined root mean square error in the database; MFC_L represents the preset defined mean square error in the database; C1 represents the preset measurement ratio corresponding to the mean absolute error in the database; C2 represents the preset measurement ratio corresponding to the root mean square error in the database; and C3 represents the preset measurement ratio corresponding to the mean square error in the database.

[0071] The mean absolute error is defined to characterize the upper limit of the mean absolute error of the power generation prediction model; the root mean square error is defined to characterize the upper limit of the root mean square error of the power generation prediction model; and the mean square error is defined to characterize the upper limit of the mean square error of the power generation prediction model.

[0072] The mean absolute error, root mean square error, and mean square error of a power generation prediction model reflect its accuracy and stability. Anomalies in the data fusion process capture additional uncertainties introduced during data preprocessing or integration. Both factors jointly determine the magnitude of outliers: a significant increase in the model error ratio or fusion increment leads to a corresponding increase in outliers, indicating potential problems with model performance or data quality. The anomaly increment in the data fusion process quantifies the additional errors introduced by information integration bias or noise during data fusion. This increment enhances the robustness of outlier assessment, ensuring the final result better reflects the actual fluctuations in model performance under complex scenarios. This is particularly effective when data sources are heterogeneous or the fusion algorithm has limitations, highlighting potential problems. When the mean absolute error of a power generation prediction model increases, it typically signifies an increase in overall prediction bias, directly leading to a significant increase in the mean square error and consequently, a higher root mean square error. When the root mean square error (RMSE) of a power generation prediction model increases, it leads to a significant quadratic increase in both the overall RMSE and the mean absolute error (MAE). Since the RMSE is the square root of the total RMSE, it inevitably increases, and the magnitude of this increase reflects the degree of rise in the RMSE. Simultaneously, the MAE of the power generation prediction model typically also increases with the increase in the RMSE, as both reflect the overall level of error.

[0073] The metric proportion corresponding to the mean absolute error (MAE) of the power generation prediction model represents the relative proportion between the MAE and the predefined MAE, indicating the degree of influence on outliers in the prediction model's evaluation indicators. This proportion quantifies the contribution weight of the MAE in the overall evaluation, reflecting the relative importance of prediction accuracy deviation in anomaly detection. The metric proportion corresponding to the root mean square (RMS) error of the power generation prediction model represents the relative proportion between the RMS error and the predefined RMS error, indicating the degree of influence on outliers. This value determines the importance of the RMS error in the comprehensive evaluation, used to assess the crucial role of prediction stability in anomaly identification. The metric proportion corresponding to the mean square error (MSE) of the power generation prediction model represents the relative proportion between the MMS error and the predefined MMS error, indicating the degree of influence on outliers. This measure assesses the sensitivity of the MSE in anomaly detection, highlighting the diagnostic value of prediction error dispersion in overall quantification.

[0074] The database stores preset benchmark parameters for power generation prediction models, including defined mean absolute error, defined root mean square error, and defined mean square error. These parameters are dynamically bound to the error indicators output by the power generation prediction model and the preset outlier evaluation benchmarks through a structured parameter mapping table, forming a complete outlier calculation system for the prediction model evaluation indicators. When it is necessary to calculate outliers for a specific prediction model, the system extracts the error indicators of the current model and the abnormal increments during the data fusion process. Subsequently, based on a preset rule base or historical data matching, the error indicators are compared with the parameter mapping table in the database. Finally, the system dynamically outputs the defined mean absolute error, defined root mean square error, and defined mean square error applicable to the model, along with the corresponding weight coefficients. Among them, C1, C2, and C3 measure the proportion values ​​as weight coefficients, and their values ​​range from 0 to 1.

[0075] Figure 5 The flowchart for adjusting the prediction model of this invention is as follows: The process begins by constructing spatiotemporal correlation features and collecting and analyzing model error indicators based on these features to obtain outliers of the prediction model evaluation indicators; then, a judgment is made: if the outlier of the prediction model evaluation indicators is less than or equal to the set outlier threshold, the current prediction model parameters are retained and the predicted power generation value is directly output; if the outlier of the prediction model evaluation indicators is greater than the set outlier threshold, the relevant spatiotemporal correlation features of multi-source data are stored in the outlier area of ​​the database, then the prediction model parameters are adjusted and the adjusted outlier of the prediction model evaluation indicators is recalculated, and a judgment is made based on the adjusted outlier of the prediction model evaluation indicators whether further secondary adjustment is needed. Figure 6 The flowchart below shows the secondary adjustment process of the prediction model of this invention. It determines whether secondary adjustment is needed. If the abnormal value of the adjusted evaluation index does not exceed the abnormal threshold, the model is directly used to output the predicted value. If the abnormal value of the adjusted evaluation index exceeds the abnormal threshold, the relevant data is stored in the database's abnormal area and secondary adjustment is performed. After secondary adjustment, the model is evaluated again to obtain the abnormal value of the readjusted prediction model evaluation index. If the abnormal value of the readjusted prediction model evaluation index does not exceed the abnormal threshold, the prediction result is output. If the abnormal value of the readjusted prediction model evaluation index exceeds the abnormal threshold, an early warning signal is generated to ensure that the final predicted value is output only when the model's reliability meets the requirements.

[0076] Specifically, the process involves determining whether to adjust the parameters of the prediction model. The specific process is as follows: collect and analyze the model error index to obtain outliers in the prediction model evaluation index.

[0077] The outliers of the prediction model evaluation index are compared with the outlier threshold of the evaluation index. If the outlier of the prediction model evaluation index is less than or equal to the outlier threshold of the evaluation index, the parameters of the prediction model are retained, and the predicted power generation value output by the prediction model is obtained.

[0078] If the outlier of the prediction model evaluation index is greater than the outlier threshold, the spatiotemporal correlation features constructed from the multi-source dataset are stored in the outlier region of the database. Then, based on the outlier of the prediction model evaluation index, the learning rate in the prediction model is reduced. The prediction model is retrained and validated by constructing spatiotemporal correlation features from the multi-source dataset. The adjusted outlier of the prediction model evaluation index is obtained, and it is determined whether the parameters of the prediction model need to be adjusted a second time.

[0079] The evaluation indicator anomaly threshold is used to characterize the maximum upper limit of abnormal values ​​of the evaluation indicators of the prediction model. It is determined by comprehensively analyzing the historical performance of the model and business needs. It is usually based on the historical evaluation indicator data of the model during stable operation, calculating its high-order statistical value, and setting it together with the maximum acceptable error range of the business, and stored in the database.

[0080] Based on the evaluation index of the prediction model, the learning rate in the prediction model is reduced by the outlier. The outlier index of the data fusion process is divided into different level intervals in the database. Each level interval corresponds to a different reduction value in the learning rate in the prediction model. The learning rate in the next prediction model is obtained by subtracting the reduction value in the learning rate in the current prediction model.

[0081] The strategy of reducing the learning rate based on outliers in the predictive model evaluation metrics has the core advantage of enhancing the model's convergence stability in outlier data regions by dynamically reducing the model parameter update step size. When the outliers in the predictive model evaluation metrics are high, appropriately reducing the learning rate can prevent the parameters from getting stuck in local optima or diverging due to drastic gradient fluctuations, prompting the model to learn more refined effective patterns in spatiotemporal correlation features. This adjustment directly suppresses the risk of overfitting or underfitting corresponding to outliers. Through iterative retraining, the outliers in the evaluation metrics gradually converge towards the outlier threshold, avoiding training oscillations caused by aggressive adjustments and improving the model's ability to capture complex relationships in multi-source data through gradual optimization, ultimately reducing and stabilizing outliers within a reasonable range. By adaptively reducing the parameter update step size, the robustness of the model to noise, errors, and inherent drastic fluctuations in multi-source data is significantly enhanced. When an increase in outliers in the predictive model evaluation metrics is detected, this strategy suppresses the risk of local optima or divergence caused by drastic gradient fluctuations, prompting the model to capture more refined effective patterns in complex spatiotemporal correlation features and avoiding overfitting problems caused by outliers. By progressively optimizing the balance between training speed and accuracy, this mechanism ensures the stability of model convergence in a multi-source heterogeneous data environment, systematically driving the prediction error to steadily converge toward the threshold range, thereby improving the generalization ability and reliability of power generation prediction and providing precise support for grid dispatch and energy management.

[0082] By constructing a systematic error monitoring and dynamic parameter adjustment mechanism, the adaptability and accuracy of the prediction model have been significantly improved, providing reliable support for wind and solar power generation prediction. Its core benefits are highlighted in three aspects: First, the automated judgment process based on anomaly threshold comparison can quickly identify model performance degradation, effectively avoiding the lag of manual intervention and ensuring timely response to complex variables in wind and solar power generation. Second, when anomalies are detected, the targeted reduction of the learning rate and retraining prevents overfitting risks caused by drastic parameter fluctuations and progressively optimizes the model's performance in multi-source data fusion scenarios, improving the stability of power generation prediction. Finally, the design of automatically archiving abnormal data to the database's abnormal area not only provides rich historical reference samples for subsequent model iterations but also constructs a closed-loop optimization system. By continuously accumulating abnormal cases and spatiotemporal features related to wind and solar power generation, it continuously strengthens the model's generalization ability to complex operating conditions. This method, which combines real-time evaluation, parameter adjustment, and data accumulation, ensures the reliability of wind and solar power generation prediction results while enabling model self-regulation.

[0083] Furthermore, it is determined whether to perform secondary adjustments on the parameters of the prediction model. The specific determination process is as follows: if the abnormal value of the evaluation index of the adjusted prediction model is less than or equal to the abnormal threshold of the evaluation index, then the parameters of the prediction model are retained, and the predicted power generation value output by the prediction model is obtained.

[0084] If the outlier of the adjusted prediction model evaluation index is greater than the outlier threshold, the spatiotemporal correlation features constructed from the multi-source data set are re-stored in the outlier region of the database. Then, based on the outlier of the adjusted prediction model evaluation index, the learning rate and Dropout rate in the prediction model are reduced, thereby obtaining the readjusted outlier of the prediction model evaluation index.

[0085] If the outlier of the readjusted prediction model evaluation index is less than or equal to the outlier threshold of the evaluation index, the parameters of the prediction model will be retained, and the prediction model will output the predicted power generation value.

[0086] If the outlier of the readjusted prediction model's evaluation index exceeds the outlier threshold, an early warning signal will be generated.

[0087] It generates early warning signals to alert staff through lights and sounds.

[0088] This method, which reduces the learning rate and dropout rate of the prediction model by adjusting the outlier value of the prediction model evaluation index, effectively suppresses the interference of data fluctuations on the prediction model and improves the stability and accuracy of wind and solar power generation prediction in multi-source heterogeneous data fusion scenarios. The anomaly index of the data fusion process is divided into different level intervals in the database, with each level interval corresponding to a different reduction in the learning rate of the prediction model. Subtracting this reduction from the current reduction in the learning rate yields the learning rate for the next prediction model.

[0089] When adjusting the learning rate and dropout rate of a predictive model to reduce outliers in the evaluation metrics, adjusting the learning rate usually has a more significant impact than adjusting the dropout rate. This is because the learning rate directly controls the step size of parameter updates; an excessively high learning rate can lead to instability in the optimization process, causing the model to fluctuate wildly or even diverge during training, thus significantly affecting the evaluation metrics. The dropout rate, on the other hand, primarily improves the model's generalization ability indirectly by preventing overfitting, and its adjustment effect is relatively mild. Therefore, if the outliers in the predictive model's evaluation metrics are mainly caused by training instability, reducing the learning rate can more directly stabilize the model and reduce outliers. If the outliers originate from data noise or overfitting, adjusting the dropout rate may be more effective. However, overall, adjusting the learning rate has a more significant and direct short-term impact on the outliers in the predictive model's evaluation metrics.

[0090] Reducing the learning rate and dropout rate in the adjusted prediction model's evaluation metrics outliers can improve the model's stability and generalization ability through a more conservative parameter tuning strategy. Lowering the learning rate slows down the parameter update step size, avoiding abnormal fluctuations in the evaluation metrics caused by gradient oscillations, thus more robustly approaching the optimal solution. Lowering the dropout rate retains more network nodes for training, enhancing the model's ability to learn features, especially suitable for situations where outliers may be caused by overfitting or feature loss. This dual adjustment effectively suppresses the magnitude of outlier evaluation metrics, making them more likely to converge to the threshold range, while reducing the model's sensitivity to noise or outlier data. If the adjusted prediction model's evaluation metrics still exceed the outlier threshold, it indicates that the problem may stem from data quality or the model structure itself. In this case, triggering an early warning signal can guide further investigation into the root cause.

[0091] By constructing an adaptive, multi-stage dynamic parameter optimization system, precise control of model performance is achieved through a hierarchical adjustment strategy. Its core value lies in: employing a progressive optimization logic, firstly adjusting the learning rate for initial correction; if still insufficient, further co-optimizing the learning rate and Dropout rate ensures training stability while enhancing generalization ability; simultaneously establishing a closed-loop feedback mechanism for abnormal data, continuously accumulating spatiotemporal feature data to strengthen the model's adaptability; and finally setting an intelligent early warning threshold, promptly terminating redundant calculations and prompting intervention when two optimizations are ineffective, effectively balancing automation efficiency and result reliability. This structured adjustment process not only significantly improves the accuracy of power generation prediction but also forms a complete intelligent operation and maintenance closed loop from parameter optimization to data accumulation to abnormal early warning.

[0092] Figure 7 The flowchart illustrates the relationship between the power generation difference and the actual power generation of this invention. This flowchart describes the closed-loop feedback adjustment mechanism of the power generation prediction system: The system first outputs the predicted power generation value, then collects the actual power generation and calculates the difference between the two; if the difference is within the power generation difference range, the current model parameters remain unchanged; if the difference exceeds the power generation difference range, the predicted value is stored in the difference area of ​​the database, and the prediction model is adjusted accordingly to update its parameters to optimize future prediction accuracy, thereby forming a continuous closed-loop optimization mechanism.

[0093] Specifically, the actual total power generation is compared with the predicted power generation output by the prediction model. The comparison process involves collecting the actual total power generation from the regional new energy control center.

[0094] The difference between the actual total power generation and the predicted power generation value output by the prediction model is analyzed to obtain the power generation difference.

[0095] A difference analysis is performed between the actual total power generation and the predicted power generation value output by the prediction model, and the absolute difference between the two is calculated.

[0096] The difference in power generation is compared with the range of power generation difference to determine whether the parameters of the prediction model need to be readjusted.

[0097] By collecting real-time actual power generation data from regional renewable energy control centers and performing precise difference analysis with the output values ​​of prediction models, prediction deviations can be dynamically quantified, forming a closed-loop feedback mechanism. Intelligent comparison of power generation differences with their intervals avoids frequent parameter adjustments due to minor fluctuations and promptly identifies systematic deviations, providing an objective basis for parameter optimization. This method enhances the model's adaptive capabilities, continuously improving prediction accuracy through data-driven approaches, ultimately achieving a dual improvement in power generation planning efficiency and renewable energy absorption rate, while providing more reliable decision support for grid dispatch.

[0098] Specifically, the process for determining whether to readjust the prediction model parameters is as follows: if the difference in power generation falls within the power generation difference range, then the predicted power generation result is accurate, and it is determined that the prediction model parameters should not be adjusted.

[0099] If the power generation difference does not fall within the power generation difference range, the predicted power generation value output by the prediction model is stored in the difference range of the database. Then, the learning rate and dropout rate of the prediction model are adjusted based on the power generation difference to update the prediction model parameters.

[0100] The learning rate and dropout rate of the prediction model are adjusted based on the difference in power generation, thereby updating the prediction model parameters. The difference in power generation is divided into different level intervals in the database. Each level interval corresponds to a different reduction in the learning rate in the prediction model. The learning rate of the next prediction model is obtained by subtracting the reduction in the learning rate of the current prediction model. Similarly, the difference in power generation is divided into different level intervals in the database. Each level interval corresponds to a different reduction in the dropout rate in the prediction model. The dropout rate of the next prediction model is obtained by subtracting the reduction in the dropout rate of the current prediction model.

[0101] The power generation difference range is a quantitative range used to evaluate the accuracy of the prediction model. It is determined by combining historical data statistical analysis, business tolerance, and dynamic adjustment of model performance. If the difference between the prediction and the actual power generation is within this range, the result is considered accurate; otherwise, the model parameters are adjusted to achieve closed-loop optimization.

[0102] By setting a reasonable range for power generation difference, the system can effectively distinguish between normal fluctuations and significant deviations, avoiding unnecessary parameter adjustments to the prediction model and improving its stability. When the power generation difference exceeds the reasonable range, the system automatically stores the abnormal prediction value in the difference region of the database for subsequent analysis and model optimization. Simultaneously, it dynamically adjusts the learning rate and dropout rate, enabling the model to adaptively balance training speed and generalization ability, reducing the risk of overfitting or underfitting. This intelligent adjustment mechanism not only improves the adaptability and accuracy of the prediction model, ensuring high reliability of new energy power generation prediction in the long term, but also provides more accurate data support for grid dispatch and energy management.

[0103] Reference Figure 2 As shown, the second aspect of the present invention provides an intelligent prediction system for wind and solar power generation based on multi-source data fusion, comprising: a multi-source data fusion module, a prediction model module, an intelligent optimization module, and a database.

[0104] The multi-source data fusion module is connected to the prediction model module, the prediction model module is connected to the intelligent optimization module, and the multi-source data fusion module, the prediction model module, and the intelligent optimization module are all connected to the database.

[0105] A database is used to store parameters involved in a smart forecasting system for wind and solar power generation based on multi-source data fusion.

[0106] The multi-source data fusion module is used to collect power generation performance parameters, environmental data, and equipment operation status data monitored by the regional new energy control center in real time. It preprocesses the power generation performance parameters, environmental data, and equipment operation status data respectively, merges the preprocessed power generation performance parameters, environmental data, and equipment operation status data, collects and analyzes the parameters of the data fusion process, and determines whether the data fusion process needs to be adjusted, thereby obtaining an accurate multi-source data set.

[0107] The prediction model module is used to construct spatiotemporal correlation features through multi-source data sets and input them into the power generation prediction model. The prediction model outputs the predicted power generation value, collects and analyzes the model error index, and determines whether to adjust the parameters of the prediction model, thereby intelligently optimizing the accuracy of the power generation prediction model.

[0108] The spatiotemporal correlation feature represents modeling the equipment layout of wind farms and photovoltaic power stations as a graph structure, where nodes represent wind turbines or photovoltaic arrays and edges represent the spatial correlation between equipment.

[0109] The intelligent optimization module is used to collect the actual total power generation and compare it with the power generation prediction value output by the prediction model. Based on the comparison results, it determines whether to readjust the prediction model parameters, thereby optimizing the intelligent prediction results of wind and solar power generation in the prediction model.

[0110] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A smart prediction method for wind and solar power generation based on multi-source data fusion, characterized in that, include: Step 1: Real-time collection of power generation performance parameters, environmental data, and equipment operation status data monitored by the regional new energy control center; preprocessing of the power generation performance parameters, environmental data, and equipment operation status data; fusion of the preprocessed power generation performance parameters, environmental data, and equipment operation status data; collection and analysis of data fusion process parameters; determination of whether to adjust the data fusion process; thereby obtaining an accurate multi-source data set. Step 2: Construct spatiotemporal correlation features through multi-source data sets and input them into the power generation prediction model. The prediction model outputs the predicted power generation value. Collect and analyze the model error index to determine whether the parameters of the prediction model need to be adjusted, thereby intelligently optimizing the accuracy of the power generation prediction model. The aforementioned spatiotemporal correlation feature indicates that the equipment layout of wind farms and photovoltaic power stations is modeled as a graph structure, where nodes represent wind turbines or photovoltaic arrays and edges represent the spatial correlation between equipment; Step 3: Collect the actual total power generation and compare it with the power generation prediction value output by the prediction model. Based on the comparison results, determine whether to readjust the prediction model parameters to optimize the intelligent prediction results of wind and solar power generation in the prediction model.

2. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 1, characterized in that: The determination of whether to adjust the data fusion process is as follows: Collect and analyze parameters of the data fusion process to obtain the data fusion process anomaly index; The power generation performance parameters reflect the real-time power generation capacity and efficiency of the power generation equipment; The environmental data reflects the external environmental factors that affect power generation performance and the operation of power generation equipment; The equipment operation status data reflects the operating status of the power generation equipment; The abnormality index of the data fusion process is compared with the abnormality threshold of the data fusion process. If the abnormality index of the data fusion process is less than or equal to the abnormality threshold of the data fusion process, the multi-source data set is stored in the normal area of ​​the database, and the spatiotemporal correlation features of the multi-source data set are constructed. The multi-source data set is a comprehensive dataset formed by preprocessing, merging and optimizing the power generation performance parameters, environmental data and equipment operation status data monitored by the regional new energy control center. If the data fusion process anomaly index exceeds the data fusion process anomaly threshold, the current fusion is cancelled, and the preprocessed power generation performance parameters, environmental data, and equipment operating status data are stored in the anomaly area of ​​the database. Then, based on the data fusion process anomaly index, the observation noise covariance matrix in the data fusion algorithm is increased. The preprocessed power generation performance parameters, environmental data, and equipment operating status data stored in the database anomaly area are fused, and the data fusion process parameters are re-collected and analyzed to obtain the adjusted data fusion process anomaly index. Finally, it is determined whether to perform a second adjustment to the data fusion process.

3. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 2, characterized in that: The specific determination process for whether to perform secondary adjustments on the data fusion process is as follows: If the adjusted abnormal index of the data fusion process is less than or equal to the abnormal threshold of the data fusion process, the multi-source data set will be stored in the normal area of ​​the database, and spatiotemporal correlation features will be constructed through the multi-source data set. If the adjusted data fusion process anomaly index is greater than the data fusion process anomaly threshold, the preprocessed power generation performance parameters, environmental data, and equipment operating status data are stored again in the anomaly area of ​​the database. Based on the adjusted data fusion process anomaly index, the observation noise covariance matrix in the data fusion algorithm is increased and the buffer size inside the data fusion algorithm is increased. The power generation performance parameters, environmental data, and equipment operating status data in the anomaly area of ​​the database are then re-fused to obtain the readjusted data fusion process anomaly index. If the abnormal index of the readjusted data fusion process is less than or equal to the abnormal threshold of the data fusion process, the multi-source data set will be stored in the normal area of ​​the database, and spatiotemporal correlation features will be constructed through the multi-source data set. If the abnormal index of the readjusted data fusion process exceeds the abnormal threshold of the data fusion process, an early warning will be triggered.

4. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 1, characterized in that: The specific analysis process for collecting and analyzing the parameters of the data fusion process is as follows: The parameters of the data fusion process include the noise covariance of the data fusion process, the coverage of the data fusion process, and the error covariance of the data fusion process; The calculation of the data fusion process anomaly index is based on the proportional relationship between the noise covariance of the data fusion process and the boundary noise covariance, the proportional relationship between the boundary coverage and the coverage of the data fusion process, and the proportional relationship between the error covariance of the data fusion process and the boundary error covariance. Specifically, the boundary noise covariance, boundary coverage, and boundary error covariance contrast are used as benchmarks. The proportional relationships between the noise covariance, coverage, and error covariance of the data fusion process and the corresponding boundary values ​​are calculated respectively. Different measurement ratios are assigned according to the degree of influence of each proportional relationship on the data fusion process anomaly index, and the results are summarized to finally obtain the data fusion process anomaly index. The data fusion process anomaly index is used to quantify the overall deviation between the data fusion result and the expected normal state.

5. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 1, characterized in that: The specific process for determining whether to adjust the parameters of the prediction model is as follows: Collect and analyze model error indicators to obtain outliers in the prediction model evaluation indicators; The outliers of the prediction model evaluation index are compared with the outlier threshold of the evaluation index. If the outliers of the prediction model evaluation index are less than or equal to the outlier threshold of the evaluation index, the parameters of the prediction model are retained, and the predicted power generation value output by the prediction model is obtained. If the outlier of the prediction model evaluation index is greater than the outlier threshold, the spatiotemporal correlation features constructed from the multi-source dataset are stored in the outlier region of the database. Then, based on the outlier of the prediction model evaluation index, the learning rate in the prediction model is reduced. The prediction model is retrained and validated by constructing spatiotemporal correlation features from the multi-source dataset. The adjusted outlier of the prediction model evaluation index is obtained, and it is determined whether the parameters of the prediction model need to be adjusted a second time.

6. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 5, characterized in that: The specific process for determining whether to perform secondary adjustments to the parameters of the prediction model is as follows: If the outlier of the adjusted prediction model evaluation index is less than or equal to the outlier threshold of the evaluation index, the parameters of the prediction model are retained, and the predicted power generation value output by the prediction model is obtained. If the outlier of the adjusted prediction model evaluation index is greater than the outlier threshold, the spatiotemporal correlation features constructed from the multi-source data set are re-stored in the outlier region of the database. Then, the learning rate and Dropout rate in the prediction model are reduced based on the outlier of the adjusted prediction model evaluation index, thereby obtaining the readjusted outlier of the prediction model evaluation index. If the outlier of the evaluation index of the readjusted prediction model is less than or equal to the outlier threshold of the evaluation index, the parameters of the prediction model will be retained and the prediction model will output the predicted value of power generation. If the outlier of the readjusted prediction model's evaluation index exceeds the outlier threshold, an early warning signal will be generated.

7. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 1, characterized in that: The specific analysis process for collecting and analyzing model error indicators is as follows: Collect and analyze model error indicators, including the mean absolute error of the power generation prediction model, the root mean square error of the power generation prediction model, and the mean square error of the power generation prediction model. Obtain the abnormal final value of the data fusion process and match the abnormal increment of the data fusion process from the database; The calculation of outliers in the prediction model evaluation index is based on the proportional relationship between the mean absolute error and the defined mean absolute error of the power generation prediction model, the proportional relationship between the root mean square error and the defined root mean square error of the power generation prediction model, and the proportional relationship between the mean square error and the defined mean square error of the power generation prediction model. The specific process is as follows: using the defined mean absolute error, defined root mean square error, and defined mean square error as benchmarks, the proportional relationship between the mean absolute error, root mean square error, and mean square error of the power generation prediction model and the corresponding defined value is calculated. Different measurement ratios are assigned to the degree of influence of each proportional relationship on the outliers of the prediction model evaluation index, and the results are summarized. Then, the outlier increment of the data fusion process is added to obtain the outliers of the prediction model evaluation index. Outliers in the prediction model evaluation metrics are used to quantify the degree of abnormality in the overall performance of the prediction model.

8. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 1, characterized in that: The comparison is performed with the predicted power generation value output by the prediction model. The specific comparison process is as follows: The actual total power generation is collected from the regional new energy control center; The difference between the actual total power generation and the predicted power generation value output by the prediction model is analyzed to obtain the power generation difference. Compare the difference in power generation with the range of power generation difference to determine whether the parameters of the prediction model need to be readjusted.

9. The intelligent prediction method for wind and solar power generation based on multi-source data fusion according to claim 8, characterized in that: The specific process for determining whether to readjust the prediction model parameters is as follows: If the difference in power generation falls within the range of power generation difference, then the predicted power generation result is accurate, and it is determined that the parameters of the prediction model will not be adjusted. If the power generation difference does not fall within the power generation difference range, the predicted power generation value output by the prediction model is stored in the difference range of the database. Then, the learning rate and dropout rate of the prediction model are adjusted based on the power generation difference to update the prediction model parameters.

10. A multi-source data fusion-based intelligent forecasting system for wind and solar power generation, characterized in that: include: The multi-source data fusion module is used to collect power generation performance parameters, environmental data and equipment operation status data monitored by the regional new energy control center in real time. It preprocesses the power generation performance parameters, environmental data and equipment operation status data respectively, merges the preprocessed power generation performance parameters, environmental data and equipment operation status data, collects and analyzes the data fusion process parameters, and determines whether to adjust the data fusion process, so as to obtain an accurate multi-source data set. The prediction model module is used to construct spatiotemporal correlation features through multi-source data sets and input them into the power generation prediction model. The prediction model outputs the predicted power generation value, collects and analyzes the model error index, and determines whether the parameters of the prediction model need to be adjusted, thereby intelligently optimizing the accuracy of the power generation prediction model. The spatiotemporal correlation feature represents modeling the equipment layout of wind farms and photovoltaic power stations as a graph structure, where nodes represent wind turbines or photovoltaic arrays and edges represent the spatial correlation between equipment; The intelligent optimization module is used to collect the actual total power generation and compare it with the power generation prediction value output by the prediction model. Based on the comparison results, it determines whether to readjust the prediction model parameters, thereby optimizing the intelligent prediction results of wind and solar power generation in the prediction model.

Images