Multi-scene new energy generation power prediction method and system

Abstract

The invention discloses a multi-scene new energy power generation power prediction method and system, and belongs to the technical field of power generation power prediction, and the method comprises the steps: collecting multi-source data, carrying out the feature extraction of the multi-source data to determine a current operation scene, and carrying out the matching of an optimal prediction model based on the current operation scene, therefore, the problem that a traditional single model is insufficient in adaptation to a complex scene is avoided, the problems that an existing model is rigid and solidified and needs too high computing resources are solved, finally, a final prediction result is obtained through a multi-model dynamic weight fusion mechanism, prediction deviation of the single model is effectively counteracted, and the prediction accuracy is improved. Therefore, the stability and the accuracy of the prediction result are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of power generation prediction technology, specifically to a method and system for predicting new energy power generation in multiple scenarios. Background Technology

[0002] New energy power generation is characterized by significant intermittency, volatility, and uncertainty, with its output influenced by a combination of factors including meteorological conditions, geographical environment, and equipment status. This characteristic poses a severe challenge to the safe and stable operation of the power grid and the large-scale consumption of new energy. Traditional power forecasting methods mainly include statistical forecasting methods based on time series analysis, numerical weather prediction methods based on physical principles, and modern intelligent forecasting technologies. However, statistical forecasting methods require high data stability and struggle to effectively handle the nonlinear characteristics of new energy power generation. Numerical weather prediction methods extrapolate changes in meteorological elements through atmospheric physical models, but these methods are computationally complex and have limited ability to capture local micro-meteorological phenomena. Modern intelligent forecasting technologies can handle nonlinear relationships to some extent, but feature engineering is complex, model generalization ability needs improvement, and model training requires a large amount of labeled data and high computational resources.

[0003] In summary, traditional forecasting methods are mainly designed for specific operating conditions. When weather conditions, seasonal changes, or equipment status change, the forecasting accuracy drops significantly, failing to meet the stability and accuracy requirements of actual power grid dispatching. Furthermore, most of them adopt static model architectures, lacking online learning and adaptive adjustment mechanisms, which makes it impossible for the system to continuously optimize with changes in the operating environment, and long-term operating performance is difficult to guarantee.

[0004] Chinese Patent, Publication No. CN118783396A, Publication Date: October 15, 2024, discloses a method for predicting low-voltage distributed photovoltaic power in a local balance scenario. This method involves collecting time-series and non-time-series data, cleaning and formatting the data, extracting time-series and non-time-series features, and evaluating feature importance. Then, it performs personalized model training and fusion to construct a prediction model, performs model fusion and optimization, and finally evaluates and optimizes prediction performance, verifies historical data, and analyzes performance indicators. Based on the analysis results, it adapts the model and updates it in real time. However, it is difficult to adjust model parameters promptly when meteorological conditions, seasonal changes, or equipment status change, thus making it difficult to guarantee prediction accuracy. Summary of the Invention

[0005] This invention addresses the problems of existing new energy power generation prediction methods, which struggle to adapt to changes in scenarios and suffer from low accuracy due to a lack of adaptive adjustment mechanisms. It provides a multi-scenario new energy power generation prediction method and system. By collecting multi-source data and extracting features, the current operating scenario is determined. Based on this scenario, the optimal prediction model is matched, thus avoiding the inadequacy of traditional single-model adaptation to complex scenarios and resolving the pain points of rigid and inflexible existing models and excessive computational resource requirements. Finally, a multi-model dynamic weight fusion mechanism is used to obtain the final prediction result, effectively offsetting the prediction bias of a single model and significantly improving the stability and accuracy of the prediction results.

[0006] In a first aspect, one technical solution provided in this embodiment of the invention is: a method for predicting the power generation of new energy sources in multiple scenarios, comprising the following steps: S1. Simultaneously collect meteorological data, power data, and equipment status data, and preprocess them to obtain raw multi-source data; S2. Extract features from the original multi-source data and arrange them according to a fixed dimension to obtain multi-source feature vectors; analyze the multi-source feature vectors based on a scene matching mechanism to determine the current scene; S3. Based on the scene model matching mechanism, match the corresponding prediction model for the current scene and output the corresponding power prediction value; S4. Perform a comprehensive evaluation of the power prediction values ​​and output the power prediction results in a structured form.

[0007] This solution ensures data integrity and consistency by simultaneously collecting and preprocessing three core data types: meteorological, power, and equipment status. This addresses the issue of insufficient feature representation from a single data source, laying a solid foundation of high-quality data for subsequent predictions. By extracting and structuring key features in the time domain, frequency domain, and time series, and combining this with a fuzzy clustering scene matching mechanism, the solution accurately determines the current operating scenario. This allows the system to adapt to changes in weather, season, and equipment status, avoiding the shortcomings of traditional single-model scenario adaptation. By accurately mapping scenarios and models to match the optimal dedicated model, and relying on edge computing to achieve rapid local access, the solution addresses model rigidity, reduces computational resource consumption, and balances prediction accuracy and real-time performance, meeting the rapid response requirements of power grid dispatch. Through a multi-model dynamic weight fusion mechanism, the solution comprehensively evaluates the predicted values, offsetting the bias of a single model, and outputs the results in a standardized and structured form. This not only adapts to external system interfaces but also provides a basis for model closed-loop optimization, thereby supporting multi-energy collaborative dispatch.

[0008] Preferably, the meteorological data includes solar irradiance, ambient temperature, wind speed, wind direction, and cloud cover. The power data includes the instantaneous output power of photovoltaic modules, wind turbines, and energy storage devices, as well as the cumulative power generation. The device status data includes the remaining power of the energy storage device, the device operating temperature, and component fault warning signals. The timestamps of the meteorological data, power data, and equipment status data are consistent.

[0009] This solution achieves comprehensive and synchronous collection of multi-dimensional data by covering core meteorological indicators such as solar irradiance, wind, solar and energy storage power data, and key parameters of equipment status, while ensuring that the timestamps of the three types of data are consistent. This solves the problems of incomplete expression of single data features and misalignment of time sequence, providing high-quality data support for subsequent accurate feature extraction and scene recognition. It effectively improves the scene adaptability and power prediction accuracy of the prediction model, and lays a solid data foundation for multi-energy collaborative scheduling.

[0010] As a preferred embodiment, in S2, feature extraction is performed on the original multi-source data and the multi-source feature vector is obtained by arranging it according to a fixed dimension, including the following steps; The temporal characteristics of the original multi-source data are obtained by analyzing the fluctuations of power data within a fixed time period. Frequency domain analysis of meteorological data yields the dominant frequency component, which serves as the frequency domain characteristic of the original multi-source data. The changing trends and cycles of power data are analyzed, and the power status characteristics are correlated with equipment status data to obtain power status features as the time-series features of the original multi-source data; The time-domain features, frequency-domain features, and time-series features are arranged in a fixed dimension to generate a data matrix corresponding to the timestamp, which is then used as a multi-source feature vector.

[0011] This solution extracts time-domain features from power fluctuations, analyzes the frequency domain dominant components from meteorological data, and obtains time-series features by correlating power trend cycles with equipment status. This enables the capture of core influencing factors of new energy power generation from multiple dimensions and perspectives, thus overcoming the limitation that a single feature cannot characterize nonlinearity and volatility. By generating a standardized data matrix arranged in fixed dimensions, the structure and consistency of feature vectors are ensured, providing accurate input for subsequent fuzzy clustering scene recognition. It also helps the system quickly match the optimal prediction model, effectively improving the accuracy of scene adaptation and power prediction, and laying a reliable feature foundation for the final closed-loop optimization.

[0012] Preferably, in S2, the current scene is determined by analyzing multi-source feature vectors based on a scene matching mechanism, including the following steps: Historical multi-source data is collected and features are extracted from the historical multi-source data to obtain historical multi-source feature data; Using historical multi-source feature data as training samples, a clustering algorithm is used to train the cluster centers of typical scenes to obtain a scene recognition model; The typical scenarios include photovoltaic-dominated scenarios, wind power-dominated scenarios, hybrid scenarios, and extreme weather scenarios; The multi-source feature vector is used as the input to the scene recognition model. The Euclidean distance between the multi-source feature vector and each cluster center is calculated. The scene type corresponding to the cluster center with the smallest Euclidean distance is selected as the matching result of the current scene.

[0013] In this solution, a clustering algorithm is used to construct an identification model covering photovoltaic, wind power, hybrid, and extreme weather scenarios by using historical multi-source feature data as training samples. This achieves comprehensive coverage of typical scenarios and overcomes the limitations of traditional methods in terms of inaccurate scenario segmentation and incomplete coverage. By calculating the Euclidean distance between real-time feature vectors and cluster centers, the objectivity and efficiency of scenario identification are ensured, thereby quickly and accurately identifying the current operating scenario. This provides a reliable basis for subsequent adaptation to dedicated prediction models, effectively improving the system's adaptability to complex and ever-changing operating environments and ensuring the scenario adaptation accuracy of power prediction from the source.

[0014] Preferably, in S3, the prediction model corresponding to the current scene is matched based on the scene model matching mechanism, and the corresponding power prediction value is output, including the following steps: If the current scenario is dominated by photovoltaics, then the photovoltaic prediction model will be matched. If the current scenario is a wind power-dominated scenario, then match the wind power prediction model; If the current scenario is a mixed scenario, then multiple types of prediction models will be matched; If the current scenario is an extreme weather scenario, then a physical-coordinated prediction model is matched; Output the power prediction value for the corresponding model.

[0015] This solution establishes a precise matching mechanism between scenarios and dedicated models, comprehensively covering core scenarios such as photovoltaic, wind power, hybrid, and extreme weather, overcoming the limitations of insufficient adaptability and incomplete scenario coverage of traditional single models. By matching dedicated prediction models to different scenarios, the solution fully leverages the predictive advantages of each model in its corresponding scenario, thereby avoiding the weakness of generalized models. Furthermore, the matching logic is objective and efficient, not only quickly identifying the optimal model for the current scenario and providing key support for accurate prediction, but also significantly improving the system's adaptability to complex and ever-changing operating environments, ensuring prediction accuracy and stability under different operating conditions.

[0016] Preferably, the multi-type prediction model selects the corresponding single prediction model based on the scene type present in the mixed scene to obtain the single prediction value, and assigns model weights based on the root mean square error of the last n prediction results of each single prediction model, and performs a weighted summation of the single prediction values ​​based on the model weights to obtain the power prediction value. The power prediction result is obtained by outputting the power prediction value, prediction period, current scene type and model matching record in a structured form; After obtaining the power prediction results, the parameters of the prediction model are optimized through a closed-loop feedback mechanism.

[0017] In this solution, by objectively assigning model weights to each individual model in mixed scenarios and weightedly fusing the prediction results of scenario-specific models, the advantages of different models can be fully utilized, the bias of a single model can be effectively offset, the limitation of weak generalization ability of general models can be overcome, and the accuracy and stability of prediction results can be improved. By outputting core information in a structured form, it is clear, complete and compatible with external system interfaces, which facilitates application scheduling and subsequent traceability. At the same time, combined with a closed-loop feedback mechanism to continuously optimize model parameters, it can avoid model rigidity, allow the system to dynamically evolve with running data, further enhance its adaptability to complex environments, and ensure prediction performance in long-term operation.

[0018] Preferably, after obtaining the power prediction results, the parameters of the prediction model are optimized through a closed-loop feedback mechanism, including the following steps: After the forecast period ends, the actual power generation during that period is collected and matched one-to-one with the target forecast value according to the timestamp; The prediction accuracy is obtained by comparing the actual power generation with the target prediction value. If the prediction accuracy of the target prediction value is lower than the preset threshold for n consecutive rounds, it is determined that the accuracy is not up to standard, and the prediction model is retrained.

[0019] In this scheme, the prediction accuracy is objectively quantified by precisely matching the actual power generation with the target prediction value according to the timestamp, thus avoiding performance evaluation bias. By setting a rule to trigger retraining for n consecutive rounds of failure to meet the target, the model adaptation failure problem can be captured in a timely manner. This overcomes the limitations of traditional models being rigid and lacking dynamic optimization, realizes continuous iterative upgrades of prediction model parameters, strengthens the system's adaptability to environmental changes, and ensures the prediction accuracy and stability under different operating conditions during long-term operation, providing reliable support for grid dispatch and multi-energy collaborative optimization.

[0020] Secondly, one technical solution provided in this embodiment of the invention is: a multi-scenario new energy power generation prediction system, including a data acquisition module, a feature extraction module, a scenario analysis module, a model matching module, and a comprehensive prediction module; The data acquisition module synchronously collects meteorological data, power data, and equipment status data, and performs preprocessing to obtain raw multi-source data. The feature extraction module extracts features from the original multi-source data and arranges them according to a fixed dimension to obtain a multi-source feature vector; it then analyzes the multi-source feature vector based on a scene matching mechanism to determine the current scene. The scenario analysis module is equipped with a scenario model matching mechanism, which matches the corresponding prediction model to the current scenario and outputs the corresponding power prediction value. The integrated prediction module comprehensively evaluates the power prediction values ​​and outputs the power prediction results in a structured form.

[0021] In this solution, a corresponding system is built to integrate the power prediction method, thereby enabling human-computer interaction and improving the user experience.

[0022] Thirdly, one technical solution provided in this embodiment of the invention is: a computer device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to execute the program stored in the memory to implement the steps of a multi-scenario new energy power generation prediction method.

[0023] Fourthly, one technical solution provided in this embodiment of the invention is: a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of a multi-scenario new energy power generation prediction method.

[0024] The beneficial effects of this invention are as follows: This invention determines the current operating scenario by collecting multi-source data and extracting its features, and then matches the optimal prediction model based on the current operating scenario. This not only avoids the problem of insufficient adaptation of traditional single models to complex scenarios, but also solves the pain points of rigidity and excessive computational resource requirements of existing models. Finally, the final prediction result is obtained through a multi-model dynamic weight fusion mechanism, which effectively offsets the prediction bias of a single model, thereby significantly improving the stability and accuracy of the prediction results.

[0025] The above description of the invention is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0027] Figure 1 This is a flowchart of a multi-scenario new energy power generation prediction method according to the present invention; Figure 2 This is a schematic diagram of a multi-scenario new energy power generation prediction system according to the present invention; Figure 3 This is a schematic diagram of a computer device provided by the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0029] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures; the process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0030] Example 1: As Figure 1 As shown, in order to address the problems of existing new energy power generation prediction methods being unable to adapt to the impact of changing scenarios on prediction accuracy, and the lack of adaptive adjustment mechanisms leading to low prediction accuracy, this embodiment provides a multi-scenario new energy power generation prediction method, including the following steps: S1: Simultaneously collect meteorological data, power data, and equipment status data, and preprocess them to obtain raw multi-source data.

[0031] In this embodiment, the meteorological data includes solar irradiance, ambient temperature, wind speed, wind direction, and cloud cover. The power data includes the instantaneous output power of photovoltaic modules, wind turbines, and energy storage devices, as well as the cumulative power generation. The device status data includes the remaining power of the energy storage device, the device operating temperature, and component fault warning signals. The timestamps of the meteorological data, power data, and equipment status data are consistent.

[0032] Specifically, during data collection, data is generally collected automatically at fixed intervals, which can be configured to be 1 minute / round or 15 minutes / round, etc., according to scheduling needs. In emergency scheduling scenarios, a new round of forecasts can be triggered manually. Meteorological data can be collected through devices such as irradiance sensors, temperature sensors, wind speed and direction instruments, and cloud cover monitoring cameras. Power data can be collected through devices such as current transformers, voltage transformers, and power transmitters. Equipment status data can be collected through devices such as energy storage status monitoring devices and equipment operation and maintenance sensors.

[0033] This embodiment achieves comprehensive and synchronous collection of multi-dimensional data by covering core meteorological indicators such as solar irradiance, wind, solar and energy storage power data, and key parameters of equipment status, while ensuring that the timestamps of the three types of data are consistent. This solves the problems of incomplete expression of single data features and misalignment of time sequence, and provides high-quality data support for subsequent accurate feature extraction and scene recognition. It effectively improves the scene adaptability and power prediction accuracy of the prediction model, and lays a solid data foundation for multi-energy collaborative scheduling.

[0034] S2: Extract features from the original multi-source data and arrange them in a fixed dimension to obtain multi-source feature vectors; analyze the multi-source feature vectors based on the scene matching mechanism to determine the current scene.

[0035] In this embodiment, feature extraction is performed on the original multi-source data and the multi-source feature vector is obtained by arranging it according to a fixed dimension, including the following steps; The temporal characteristics of the original multi-source data are obtained by analyzing the fluctuations of power data within a fixed time period. Frequency domain analysis of meteorological data yields the dominant frequency component, which serves as the frequency domain characteristic of the original multi-source data. The changing trends and cycles of power data are analyzed, and the power status characteristics are correlated with equipment status data to obtain power status features as the time-series features of the original multi-source data; The time-domain features, frequency-domain features, and time-series features are arranged in a fixed dimension to generate a data matrix corresponding to the timestamp, which is then used as a multi-source feature vector.

[0036] Specifically, the mean, variance, peak, and trough of power data over the past hour (the time can be set according to requirements) can be calculated to reflect the intensity of power fluctuations as the time-domain feature of the original multi-source data. The dominant frequency component of meteorological data is extracted using Fourier transform to capture periodic variation patterns as the frequency-domain feature of the original multi-source data. The power trend and periodic terms of the power data are extracted using a sliding window method, and combined with equipment status data to generate "status-power" correlation features as the time-series features of the original multi-source data, such as the correlation coefficient between SOC and energy storage output power. By arranging the extracted features according to a fixed dimension, a data matrix corresponding to the feature and timestamp is generated as the multi-source feature vector.

[0037] This embodiment extracts time-domain features from power fluctuations, analyzes the dominant frequency components from meteorological data, and obtains time-series features by correlating power trend cycles with equipment status. This enables the capture of core influencing factors of new energy power generation from multiple dimensions and perspectives, thus overcoming the limitation that a single feature cannot characterize nonlinearity and volatility. By generating a standardized data matrix arranged in fixed dimensions, the structure and consistency of feature vectors are ensured, providing accurate input for subsequent fuzzy clustering scene recognition. It also helps the system quickly match the optimal prediction model, effectively improving the accuracy of scene adaptation and power prediction, and laying a reliable feature foundation for the final closed-loop optimization.

[0038] In this embodiment, the current scene is determined by analyzing multi-source feature vectors based on a scene matching mechanism, including the following steps: Historical multi-source data is collected and features are extracted from the historical multi-source data to obtain historical multi-source feature data; Using historical multi-source feature data as training samples, a clustering algorithm is used to train the cluster centers of typical scenes to obtain a scene recognition model; The typical scenarios include photovoltaic-dominated scenarios, wind power-dominated scenarios, hybrid scenarios, and extreme weather scenarios; The multi-source feature vector is used as the input to the scene recognition model. The Euclidean distance between the multi-source feature vector and each cluster center is calculated. The scene type corresponding to the cluster center with the smallest Euclidean distance is selected as the matching result of the current scene.

[0039] Specifically, a fuzzy C-means clustering algorithm can be used, with feature data from the past three months as training samples, to pre-train cluster centers for typical scenarios, thus obtaining a scene recognition model. The multi-source feature vector is used as input to the scene recognition model, and the Euclidean distance between this vector and each scene cluster center is calculated. The smaller the distance, the closer the current scene is to the scene corresponding to that cluster center, thus obtaining matching results. For example, a strong wind scene is dominated by wind power, a cloudy scene by solar power, and a hybrid scene by wind, solar, and energy storage synergy. Simultaneously, scene feature descriptions, such as average wind speed and irradiance fluctuation range, are output.

[0040] This embodiment uses historical multi-source feature data as training samples and employs a clustering algorithm to construct an identification model covering photovoltaic, wind power, hybrid, and extreme weather scenarios. This achieves comprehensive coverage of typical scenarios and overcomes the limitations of traditional methods in terms of inaccurate scenario segmentation and incomplete coverage. By calculating the Euclidean distance between real-time feature vectors and cluster centers, the objectivity and efficiency of scenario identification are ensured, thereby quickly and accurately identifying the current operating scenario. This provides a reliable basis for subsequent adaptation to dedicated prediction models, effectively improving the system's adaptability to complex and ever-changing operating environments and ensuring the scenario adaptation accuracy of power prediction from the source.

[0041] S3: Based on the scene model matching mechanism, match the corresponding prediction model for the current scene and output the corresponding power prediction value.

[0042] In this embodiment, the prediction model corresponding to the current scene is matched based on the scene model matching mechanism, and the corresponding power prediction value is output, including the following steps: If the current scenario is dominated by photovoltaics, then the photovoltaic prediction model will be matched. If the current scenario is a wind power-dominated scenario, then match the wind power prediction model; If the current scenario is a mixed scenario, then multiple types of prediction models will be matched; If the current scenario is an extreme weather scenario, then a physical-coordinated prediction model is matched; Output the power prediction value for the corresponding model.

[0043] Specifically, photovoltaic-dominated scenarios mainly involve sunny or cloudy conditions that affect photovoltaic power generation, and a pre-trained photovoltaic prediction model with a long short-term memory artificial neural network as its core structure is selected. Wind-dominated scenarios mainly involve strong winds or no wind that affect wind power generation, and a pre-trained wind power prediction model with a convolutional neural network as its core result is selected. Hybrid scenarios mainly involve the coordinated operation of wind, solar, and energy storage, and a multi-type prediction model that uses LSTM, ensemble learning, and CNN in combination to predict power generation is selected. Extreme weather scenarios, such as heavy rain and thunderstorms, use a physical collaborative prediction model that combines physical models and machine learning for prediction.

[0044] This embodiment establishes a precise matching mechanism between scenarios and dedicated models, comprehensively covering core scenarios such as photovoltaic, wind power, hybrid, and extreme weather, overcoming the limitations of insufficient adaptability and incomplete scenario coverage of traditional single models. By matching dedicated prediction models for different scenarios, it fully leverages the prediction advantages of each model in the corresponding scenario, thereby avoiding the weakness of generalized models. Moreover, the matching logic is objective and efficient, which can not only quickly lock the optimal model for the current scenario and provide key support for accurate prediction, but also significantly improve the system's adaptability to complex and ever-changing operating environments, ensuring the prediction accuracy and stability under different operating conditions.

[0045] S4: Perform a comprehensive evaluation of the power prediction values ​​and output the power prediction results in a structured form. Specifically, the confidence level of the output results of each model can be evaluated. For example, the confidence level of the prediction results can be evaluated based on parameters such as the root mean square error of the prediction results in the last 30 rounds. If the confidence level is greater than or equal to the confidence threshold, the prediction value is output; if it is less than the confidence threshold, the prediction is re-performed.

[0046] In this embodiment, the multi-type prediction model selects the corresponding single prediction model based on the scene type present in the mixed scene to obtain the single prediction value, and assigns model weights based on the root mean square error of the last n prediction results of each single prediction model. Based on the model weights, the single prediction values ​​are weighted and summed to obtain the power prediction value. The power prediction result is obtained by outputting the power prediction value, prediction period, current scene type and model matching record in a structured form; After obtaining the power prediction results, the parameters of the prediction model are optimized through a closed-loop feedback mechanism.

[0047] This embodiment, in mixed scenarios, objectively allocates model weights and weightedly fuses the prediction results of scenario-specific models. This fully leverages the advantages of different models, effectively offsets the bias of a single model, overcomes the limitation of weak generalization ability of general models, and improves the accuracy and stability of prediction results. By outputting core information in a structured form, it is clear, complete, and compatible with external system interfaces, facilitating application scheduling and subsequent traceability. At the same time, combined with a closed-loop feedback mechanism to continuously optimize model parameters, it can avoid model rigidity, allowing the system to dynamically evolve with running data, further enhancing its adaptability to complex environments and ensuring prediction performance in long-term operation.

[0048] In this embodiment, after obtaining the power prediction results, the parameters of the prediction model are optimized through a closed-loop feedback mechanism, including the following steps: After the forecast period ends, the actual power generation during that period is collected and matched one-to-one with the target forecast value according to the timestamp; The prediction accuracy is obtained by comparing the actual power generation with the target prediction value. If the prediction accuracy of the target prediction value is lower than the preset threshold for n consecutive rounds, it is determined that the accuracy is not up to standard, and the prediction model is retrained.

[0049] This embodiment accurately correlates actual power generation with target predicted values ​​using timestamps, objectively quantifying prediction accuracy and avoiding performance evaluation bias. By setting a rule to trigger retraining for n consecutive rounds of failure to meet targets, it can promptly capture model adaptation failures, thereby overcoming the limitations of traditional models' rigidity and lack of dynamic optimization. This enables continuous iterative upgrades of prediction model parameters, strengthens the system's adaptability to environmental changes, and ensures prediction accuracy and stability under different operating conditions during long-term operation, providing reliable support for grid dispatch and multi-energy collaborative optimization.

[0050] Example 2: Figure 2 As shown, this embodiment also provides a multi-scenario new energy power generation prediction system, including a data acquisition module, a feature extraction module, a scenario analysis module, a model matching module, and a comprehensive prediction module; The data acquisition module synchronously collects meteorological data, power data, and equipment status data, and performs preprocessing to obtain raw multi-source data. The feature extraction module extracts features from the original multi-source data and arranges them according to a fixed dimension to obtain a multi-source feature vector; it then analyzes the multi-source feature vector based on a scene matching mechanism to determine the current scene. The scenario analysis module is equipped with a scenario model matching mechanism, which matches the corresponding prediction model to the current scenario and outputs the corresponding power prediction value. The comprehensive prediction performs a comprehensive evaluation of the power prediction values ​​and outputs the power prediction results in a structured form.

[0051] This embodiment implements the power prediction method in this solution by constructing a corresponding system, thereby realizing human-computer interaction and improving the user experience.

[0052] This embodiment also provides a computer device, such as... Figure 3 As shown, it includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory is used to store computer programs; When the processor executes the program stored in memory, it implements a multi-scenario new energy power generation prediction method.

[0053] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0054] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0055] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0056] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0057] This application also provides a computer-readable storage medium storing a computer program, which is executed by a processor to provide a method for predicting the power generation of new energy sources in multiple scenarios.

[0058] As can be seen from the above embodiments, it has at least the following substantial effects: (1) This invention ensures data integrity and consistency by simultaneously collecting and preprocessing three types of core data: meteorological, power and equipment status, and solves the problem of insufficient feature expression from a single data source, thus laying a solid foundation of high-quality data for subsequent forecasting. (2) This invention extracts key features in the time domain, frequency domain and time series and arranges them in a structured manner. Combined with the fuzzy clustering scene matching mechanism, it accurately determines the current operating scene, allowing the system to adapt to changes in weather, season and equipment status, thus avoiding the shortcomings of traditional single model scene adaptation. (3) This invention accurately maps and matches the optimal dedicated model to the scene and model, and relies on edge computing to realize local fast calling, which not only solves the problem of model rigidity, but also reduces the consumption of computing resources, and takes into account both prediction accuracy and real-time performance, thus meeting the needs of rapid response of power grid dispatch. (4) This invention comprehensively evaluates the predicted value through a multi-model dynamic weight fusion mechanism, which offsets the bias of a single model and outputs the results in a standardized and structured form. It is compatible with external system interfaces and provides a basis for model closed-loop optimization, thereby supporting multi-energy collaborative scheduling.

[0059] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for predicting the power generation of new energy sources in multiple scenarios, characterized in that: Includes the following steps: S1. Simultaneously collect meteorological data, power data, and equipment status data, and preprocess them to obtain raw multi-source data; S2. Extract features from the original multi-source data and arrange them according to a fixed dimension to obtain multi-source feature vectors; analyze the multi-source feature vectors based on a scene matching mechanism to determine the current scene; S3. Based on the scene model matching mechanism, match the corresponding prediction model for the current scene and output the corresponding power prediction value; S4. Perform a comprehensive evaluation of the power prediction values ​​and output the power prediction results in a structured form.

2. The multi-scenario new energy power generation prediction method according to claim 1, characterized in that: The meteorological data includes solar irradiance, ambient temperature, wind speed, wind direction, and cloud cover. The power data includes the instantaneous output power of photovoltaic modules, wind turbines, and energy storage devices, as well as the cumulative power generation. The device status data includes the remaining power of the energy storage device, the device operating temperature, and component fault warning signals. The timestamps of the meteorological data, power data, and equipment status data are consistent.

3. The multi-scenario new energy power generation prediction method according to claim 1, characterized in that: In S2, feature extraction is performed on the original multi-source data and the multi-source feature vector is obtained by arranging it according to a fixed dimension, including the following steps; The temporal characteristics of the original multi-source data are obtained by analyzing the fluctuations of power data within a fixed time period. Frequency domain analysis of meteorological data yields the dominant frequency component, which serves as the frequency domain characteristic of the original multi-source data. The changing trends and cycles of power data are analyzed, and the power status characteristics are correlated with equipment status data to obtain power status features as the time-series features of the original multi-source data; The time-domain features, frequency-domain features, and time-series features are arranged in a fixed dimension to generate a data matrix corresponding to the timestamp, which is then used as a multi-source feature vector.

4. The multi-scenario new energy power generation prediction method according to claim 1, characterized in that: In S2, the current scene is determined by analyzing multi-source feature vectors based on a scene matching mechanism, including the following steps: Historical multi-source data is collected and features are extracted from the historical multi-source data to obtain historical multi-source feature data; Using historical multi-source feature data as training samples, a clustering algorithm is used to train the cluster centers of typical scenes to obtain a scene recognition model; The typical scenarios include photovoltaic-dominated scenarios, wind power-dominated scenarios, hybrid scenarios, and extreme weather scenarios; The multi-source feature vector is used as the input to the scene recognition model. The Euclidean distance between the multi-source feature vector and each cluster center is calculated. The scene type corresponding to the cluster center with the smallest Euclidean distance is selected as the matching result of the current scene.

5. The multi-scenario new energy power generation prediction method according to claim 4, characterized in that: In S3, the prediction model corresponding to the current scene is matched based on the scene model matching mechanism, and the corresponding power prediction value is output, including the following steps: If the current scenario is dominated by photovoltaics, then the photovoltaic prediction model will be matched. If the current scenario is a wind power-dominated scenario, then match the wind power prediction model; If the current scenario is a mixed scenario, then multiple types of prediction models will be matched; If the current scenario is an extreme weather scenario, then a physical-coordinated prediction model is matched; Output the power prediction value for the corresponding model.

6. The multi-scenario new energy power generation prediction method according to claim 4, characterized in that: The multi-type prediction model selects the corresponding single prediction model based on the scene type present in the mixed scene to obtain the single prediction value, and assigns model weights based on the root mean square error of the last n prediction results of each single prediction model. Based on the model weights, the single prediction values ​​are weighted and summed to obtain the power prediction value. The power prediction result is obtained by outputting the power prediction value, prediction period, current scene type and model matching record in a structured form; After obtaining the power prediction results, the parameters of the prediction model are optimized through a closed-loop feedback mechanism.

7. The multi-scenario new energy power generation prediction method according to claim 6, characterized in that: After obtaining the power prediction results, the parameters of the prediction model are optimized through a closed-loop feedback mechanism, including the following steps: After the forecast period ends, the actual power generation during that period is collected and matched one-to-one with the target forecast value according to the timestamp; The prediction accuracy is obtained by comparing the actual power generation with the target prediction value. If the prediction accuracy of the target prediction value is lower than the preset threshold for n consecutive rounds, it is determined that the accuracy is not up to standard, and the prediction model is retrained.

8. A multi-scenario new energy power generation prediction system, applicable to the multi-scenario new energy power generation prediction method according to any one of claims 1-7, characterized in that: It includes a data acquisition module, a feature extraction module, a scene analysis module, a model matching module, and a comprehensive prediction module; The data acquisition module synchronously collects meteorological data, power data, and equipment status data, and performs preprocessing to obtain raw multi-source data. The feature extraction module extracts features from the original multi-source data and arranges them according to a fixed dimension to obtain a multi-source feature vector; it then analyzes the multi-source feature vector based on a scene matching mechanism to determine the current scene. The scenario analysis module is equipped with a scenario model matching mechanism, which matches the corresponding prediction model to the current scenario and outputs the corresponding power prediction value. The integrated prediction module comprehensively evaluates the power prediction values ​​and outputs the power prediction results in a structured form.

9. A computer device, characterized in that: It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the steps of the multi-scenario new energy power generation prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-scenario new energy power generation prediction method as described in any one of claims 1-7.

Images