A multi-new energy station centralized power prediction and optimization evaluation method

Abstract

This invention discloses a method for centralized power prediction and evaluation of multiple renewable energy power plants, specifically relating to the field of centralized power prediction and evaluation technology. The method includes parsing and time-aligning multi-source power prediction files from multiple renewable energy power plants within the same prediction period to generate a set of prediction entries. Each prediction entry includes a power plant number, a prediction source identifier, a prediction time scale, and a predicted power sequence arranged chronologically. For each prediction entry, the method calls the measured power sequence of the corresponding power plant to perform hourly difference calculations to obtain a deviation sequence, and writes the deviation sequence into a deviation index structure in chronological order to form a historical deviation record that can be expanded over time. The method dynamically models the deviation evolution process between the multi-source power prediction results and the measured operational feedback within a sliding time window, and maps the prediction state to comparable confidence results to perform real-time optimization output.

Description

Technical Field

[0001] This invention relates to the field of centralized power prediction and evaluation technology, and more specifically, to a centralized power prediction and evaluation method for multiple renewable energy power plants. Background Technology

[0002] In the field of new energy power production and operation management, centralized power prediction results are an important basis for grid dispatch, planning and operation control. With the continuous expansion of the installed capacity of new energy, a single new energy power station often connects to multiple power prediction sources at the same time. Different prediction sources generate prediction results based on different meteorological models, calculation logic or historical data, resulting in multiple sets of parallel power prediction results in the same period. In existing technologies, the processing of multiple prediction results usually involves fixed rule filtering, static weighting, or simple accuracy ranking to determine the final prediction result. These methods generally assume that the reliability of the prediction model remains stable during operation. However, in the actual production and operation of new energy power, the prediction errors and stability of different prediction sources will continue to evolve over time due to rapid changes in meteorological conditions, differences in equipment operating status, and the geographical characteristics of the site. This makes it difficult for static evaluation methods to truly reflect the credibility of the prediction results under the current operating scenario. In addition, existing centralized power prediction management methods usually separate the evaluation of prediction accuracy from the operation management process, and lack a dynamic correction mechanism for prediction results in actual operation feedback. This means that once the prediction reliability drifts, prediction results that are not adapted to the current operating state may still be used, thereby affecting the accuracy and reliability of new energy power dispatch and operation management decisions. Therefore, there is an urgent need for a centralized power prediction optimization evaluation method that can dynamically evaluate the power prediction results of multiple new energy power plants during the production, operation and management of new energy power, and select the best target prediction results in real time based on the evolution of prediction reliability, so as to improve the scientificity and adaptability of new energy power production, operation and management. Summary of the Invention

[0003] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a method for centralized power prediction and evaluation of multiple new energy power plants. This method dynamically models the deviation evolution process between the power prediction results from multiple sources and the measured operational feedback within a sliding time window, and maps the predicted state to a comparable confidence result to perform real-time optimization output.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for centralized power prediction and selection evaluation of multiple renewable energy power plants, comprising: S1. Perform parsing and time alignment on the multi-source power prediction files from multiple renewable energy power plants within the same prediction period to generate a set of prediction entries. Each prediction entry includes the power plant number, prediction source identifier, prediction time scale, and prediction power sequence arranged in chronological order. S2. For each prediction item, call the measured power sequence of the corresponding station to perform hourly difference calculation to obtain the deviation sequence, and write the deviation sequence into the deviation index structure in time order to form a historical deviation record that can be expanded over time. S3. Perform dynamic modeling processing on the deviation index structure within the sliding time window to construct a dynamic model of prediction confidence. Dynamic modeling includes: calculating the trend of deviation accumulation over time based on historical deviation records, and simultaneously calculating the fluctuation characteristics of deviation between adjacent moments to form a prediction state model that evolves over time. S4. Call the prediction state model, and obtain the corresponding confidence results of the change trend and fluctuation characteristics of each prediction item at the current time through the corresponding confidence mapping table; perform the best calculation on the confidence results of different prediction sources, and determine the target prediction item as the lumped power prediction output. S5. When the measured deviation of the target prediction item in subsequent operation meets the preset offset condition, the current corresponding prediction state model is frozen, and the prediction state model is updated based on the newly added measured power data to form the updated prediction state model.

[0005] In a preferred embodiment, in S1, the process of generating the set of prediction entries includes: S1-1. Read the multi-source power prediction files from multiple new energy power plants on the centralized control side, extract the identification field used to identify the prediction attribution relationship in each power prediction file, and extract the predicted power data carried in the power prediction file. The identification field includes the power plant number, prediction source identification and prediction time scale. S1-2. Based on the prediction time scale, perform time axis alignment processing on the predicted power data in different power prediction files to form a predicted power data sequence under a unified time index. S1-3. Bind the time-aligned predicted power data sequence with the corresponding site number, prediction source identifier and prediction time scale to construct prediction entries, and then aggregate the prediction entries to form a prediction entry set.

[0006] In a preferred embodiment, in S2, the process of forming a historical deviation record that can be expanded over time includes: S2-1. For each prediction item, read the predicted power sequence in the prediction item and read the measured power sequence of the corresponding new energy power station. Alignment keys are constructed using the site number of the prediction entry, the prediction source identifier, the prediction time scale, and the time index in the prediction power sequence. Based on the alignment keys, the measured power values ​​under the same time index are located in the measured power sequence to form hourly paired prediction power values ​​and measured power values. S2-2. Perform a difference calculation on the predicted power value and the measured power value under each time index to generate a deviation sequence that corresponds one-to-one with the time index. While generating the deviation sequence, continuous segment identification is performed on the deviation sequence in chronological order to determine whether the deviation change of adjacent time indices in the deviation sequence exceeds a preset change threshold. If it does not exceed the threshold, adjacent deviations are merged into the same continuous deviation segment. If it exceeds the threshold, it is divided into new continuous deviation segments to output the start time index and end time index of the continuous deviation segment. S2-3. Write the start time index and end time index of the deviation sequence and continuous deviation segments into the deviation index structure in chronological order. Specifically, this includes: constructing a deviation record for each time index based on the alignment key. The deviation record includes at least the alignment key, time index, predicted power value, measured power value, and deviation value. Write the start time index and end time index of the continuous deviation segments as segment fields into the deviation record to form a searchable deviation index entry. Perform idempotent append storage using the alignment key as the write index key. If a deviation record with the same index key already exists in the deviation index structure, the corresponding deviation record is replaced. If no deviation record with the same index key exists in the deviation index structure, the corresponding deviation record is appended to form a historical deviation record that can be expanded over time.

[0007] In a preferred embodiment, in S3, the process of forming a predictive state model that evolves over time includes: S3-1. Determine the start time index and end time index of the current sliding time window in the deviation index structure, and extract the historical deviation records located within the sliding time window in chronological order. S3-2. For the historical deviation records within the sliding time window, read the deviation values ​​corresponding to each time index in chronological order, and use the cumulative deviation result corresponding to the previous time index as the recursive benchmark to perform a recursive cumulative operation on the deviation value corresponding to the current time index, forming a deviation cumulative sequence arranged in chronological order. After obtaining the cumulative deviation sequence, the cumulative deviation results corresponding to adjacent time indices are compared in chronological order to determine the increasing, decreasing, or alternating change state of the cumulative deviation sequence within the sliding time window, and the change state is determined as the trend of deviation accumulation over time. S3-3. While determining the trend of cumulative deviation, select the deviation values ​​corresponding to two adjacent time indices in the sliding time window in chronological order. Subtract the deviation value corresponding to the previous time index from the deviation value corresponding to the later time index to obtain the deviation change amount of the corresponding time index. Arrange the deviation change amounts in chronological order to form a deviation change amount sequence. After forming the deviation change sequence, the absolute change magnitudes of adjacent changes in the deviation change sequence are compared in chronological order. When the absolute change magnitudes do not exceed the preset stable threshold within the continuous time index range, it is determined that the deviation change magnitude remains stable within the continuous time index range. When the absolute change magnitudes exceed the preset drastic threshold at least once within the continuous time index range, it is determined that the deviation change magnitude has undergone drastic changes within the continuous time index range. The stability or drastic change of the determined deviation within the continuous time index range is taken as the fluctuation characteristic of the deviation between adjacent time points.

[0008] In a preferred embodiment, in S3, the process of forming a predictive state model that evolves over time further includes: S3-4. Combine the trend of the cumulative change of the deviation over time with the fluctuation characteristics of the deviation between adjacent time points to construct the prediction state model for the corresponding prediction item. S3-5. When the sliding time window moves backward and new historical deviation records are written into the deviation index structure, repeat S3-1 to S3-4 to update and build the prediction state model so that the prediction state model continues to evolve over time. The update build refers to: The first step is to move the start and end time indices of the sliding time window backward to the new time range, and then re-extract the historical deviation records located within the new sliding time window from the deviation index structure. The second step is to re-execute the recursive accumulation operation based on the re-extracted historical deviation records to generate a new deviation accumulation sequence, and redetermine the trend of deviation accumulation over time based on the new deviation accumulation sequence. The third step is to re-perform the differential calculation based on the re-extracted historical deviation records to generate a new sequence of deviation changes, and to redetermine the fluctuation characteristics of the deviation between adjacent time points based on the new absolute change magnitude determination results. The fourth step is to replace the trend and fluctuation characteristics corresponding to the previous sliding time window in the prediction state model with the newly determined trend and fluctuation characteristics, so that the prediction state model continues to evolve as the sliding time window moves forward.

[0009] In a preferred embodiment, in S4, the process of determining the target prediction entry as the lumped power prediction output includes: S4-1. Call the prediction state model. For each prediction item, read the change trend and fluctuation characteristics of the corresponding prediction item at the current time, and write the change trend and fluctuation characteristics into the state combination item field in the order of the preset state arrangement to form a state combination item for confidence calculation. S4-2. Based on the state combination item, perform a state matching operation in the credibility mapping table to determine the mapping item corresponding to the state combination item, and output the corresponding credibility result according to the mapping item. S4-3. Within the same new energy power station area, the credibility results from different prediction sources are sorted according to the preset comparison rules to obtain the credibility ranking results; The preset comparison rules include: performing pairwise comparisons based on the magnitude of the confidence results, and determining the sorting order of the confidence results based on the comparison results; S4-4. Based on the credibility ranking results, determine the prediction source that ranks first, and select the prediction item corresponding to the prediction source as the target prediction item. The target prediction item is determined as the lumped power prediction output.

[0010] In a preferred embodiment, in S5, the process of forming the updated predicted state model includes: S5-1. In the subsequent operation, for the target prediction item, the measured power data corresponding to the target prediction item is continuously read, and the deviation calculation is performed based on the measured power data and the predicted power data. When the calculated measured deviation meets the preset offset condition, the model update process is triggered. S5-2. After the model update process is triggered, the current corresponding prediction state model is frozen, the change trend and fluctuation characteristics of the prediction state model are solidified into the frozen model state, and the frozen model state is prevented from continuing to participate in the subsequent prediction state update. S5-3. After freezing the prediction state model, the newly added measured power data is written into the deviation index structure, and the deviation data within the sliding time window is re-extracted based on the updated historical deviation records as the data input for updating the model. S5-4. Based on the data input of the updated model, the deviation accumulation sequence is regenerated according to the recursive accumulation operation method of S3-2, and the deviation change sequence is regenerated according to the difference calculation and change magnitude determination method of S3-3. Then, the change trend and fluctuation characteristics in the prediction state model are re-determined to form the updated prediction state model. S5-5. Replace the frozen model state with the updated prediction state model, and use the updated prediction state model for the credibility mapping and optimization calculation of subsequent prediction periods to determine the new target prediction item as the lumped power prediction output.

[0011] The technical effects and advantages of this invention are as follows: This solution constructs a dynamic model of prediction reliability based on historical deviation records, explicitly incorporating the error evolution of multiple prediction sources at different operational stages into the evaluation process. This enables dynamic characterization and real-time optimization of the reliability of prediction results, overcoming the problem in existing technologies that rely on static rules or fixed weights and cannot reflect the changes in prediction reliability over time. This improves the adaptability and decision-making effectiveness of centralized power prediction results in actual operating scenarios. By introducing a deviation index structure and continuously writing hourly deviation records between predicted power and measured power with time index as the main line, the prediction error is transformed from discrete numerical values ​​into structured historical information that can be expanded over time. This provides a unified data foundation for subsequent stable and verifiable dynamic modeling within the sliding time window, avoiding computational redundancy and information fragmentation caused by repeatedly scanning the original sequence. Within the sliding time window, the trend of deviation accumulation over time and the fluctuation characteristics of deviation between adjacent time points are extracted respectively, so that the long-term drift behavior and short-term stability of the prediction results are characterized simultaneously. This enables the differentiation between different prediction states such as persistent inaccuracy and short-term disturbances, and improves the ability of the prediction state model to represent operational risks and reliability. The changing trends and fluctuation characteristics in the predicted state model are mapped to the credibility results under a unified standard, and different prediction sources are ranked and selected within the same new energy power station. This changes the selection process of prediction results from directly comparing numerical accuracy to a decision-making process based on state consistency, and increases the interpretability and auditability of centralized power prediction output in scenarios with multiple sources coexisting. Set up an offset triggering mechanism based on measured deviations, and freeze the current prediction state model after triggering. Then update the model based on the newly added measured data to form a complete operation mechanism from freezing to updating to re-evaluation. This prevents the distorted model from continuing to participate in subsequent decisions and avoids the prediction reliability drift from being amplified in the long term during operation. By integrating prediction and assessment, operational feedback, and model updates into a single execution process, the centralized power prediction results can continuously evolve with environmental changes during actual operation, rather than remaining at the post-event evaluation stage. This enhances the stability of the prediction results in supporting scheduling, planning, and control decisions during the production, operation, and management of new energy power. Attached Figure Description

[0012] Figure 1 This is a flowchart outlining the method steps of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Refer to the instruction manual appendix Figure 1 An embodiment of the present invention provides a method for centralized power prediction and selection evaluation of multiple renewable energy power plants, comprising: S1. On the centralized control side, the multi-source power prediction files from multiple renewable energy power plants within the same prediction period are parsed and time-aligned to generate a set of prediction entries. Each prediction entry includes the power plant number, prediction source identifier, prediction time scale, and prediction power sequence arranged in chronological order. S2. For each prediction item, call the measured power sequence of the corresponding station to perform hourly difference calculation to obtain the deviation sequence, and write the deviation sequence into the deviation index structure in time order to form a historical deviation record that can be expanded over time. S3. Perform dynamic modeling processing on the deviation index structure within the sliding time window to construct a dynamic model of prediction confidence. Dynamic modeling includes: calculating the trend of deviation accumulation over time based on historical deviation records, and simultaneously calculating the fluctuation characteristics of deviation between adjacent moments to form a prediction state model that evolves over time. S4. Call the prediction state model, and obtain the corresponding credibility results of the change trend and fluctuation characteristics of each prediction item at the current time through the corresponding credibility mapping table; perform the best calculation on the credibility results of different prediction sources within the same new energy power station area, and determine the target prediction item as the centralized power prediction output; S5. When the measured deviation of the target prediction item in subsequent operation meets the preset offset condition, the current corresponding prediction state model is frozen, and the prediction state model is updated based on the newly added measured power data to form an updated prediction state model, which is used for the credibility mapping and optimization evaluation of subsequent prediction periods.

[0015] In S1, the process of generating the set of prediction entries includes: S1-1. On the centralized control side, read the multi-source power prediction files from multiple new energy power plants, extract the identification field used to identify the prediction attribution relationship in each power prediction file, and extract the predicted power data carried in the power prediction file. The identification field includes the power plant number, prediction source identification and prediction time scale. The centralized control side refers to the centralized processing node used to centrally receive, parse and process the power prediction data of multiple new energy power plants. S1-2. Based on the prediction time scale, perform time axis alignment processing on the predicted power data in different power prediction files to form a predicted power data sequence under a unified time index. The time axis alignment processing refers to mapping each predicted power data to a unified time scale according to the prediction time scale, and rearranging the predicted power data in time order according to the unified time scale to form a predicted power data sequence under a unified time index. S1-3. Bind the time-aligned predicted power data sequence with the corresponding site number, prediction source identifier and prediction time scale to construct prediction entries, and collect the prediction entries to form a prediction entry set; where binding means establishing a one-to-one correspondence between the predicted power data sequence and the site number, prediction source identifier and prediction time scale in the same prediction entry.

[0016] In S2, the process of creating a historical deviation record that can be expanded over time includes: S2-1. For each prediction item, read the predicted power sequence in the prediction item and read the measured power sequence of the corresponding new energy power station. Alignment keys are constructed using the site number of the prediction entry, the prediction source identifier, the prediction time scale, and the time index in the prediction power sequence. Based on the alignment keys, the measured power values ​​under the same time index are located in the measured power sequence to form hourly paired prediction power values ​​and measured power values. The process of constructing alignment keys includes: The first step is to read the site number, prediction source identifier, and prediction time scale from the prediction entries and use them as fixed fields for the alignment key; The second step is to traverse the predicted power sequence and extract the time index corresponding to each predicted power value as the time field of the alignment key. The third step is to concatenate the fixed field and the time field according to the preset field order to form an alignment key, and write the alignment key into the paired index entries that correspond one-to-one with the corresponding time index, so as to locate the same time index in the measured power sequence later. S2-2. Perform a difference calculation on the predicted power value and the measured power value under each time index to generate a deviation sequence that corresponds one-to-one with the time index. While generating the deviation sequence, continuous segment identification is performed on the deviation sequence in chronological order to determine whether the deviation change of adjacent time indices in the deviation sequence exceeds a preset change threshold. If it does not exceed the threshold, adjacent deviations are merged into the same continuous deviation segment. If it exceeds the threshold, it is divided into new continuous deviation segments to output the start time index and end time index of the continuous deviation segment. The process of dividing the area into new continuous deviation segments includes: The first step is to set the time index corresponding to the first deviation in the deviation sequence as the starting time index of the current continuous deviation segment; The second step is to take two adjacent deviations in chronological order and calculate the difference between the latter deviation and the former deviation as the change in deviation. The third step is to keep the current continuous deviation segment unchanged when the deviation change does not exceed the preset change threshold, and update the time index corresponding to the next deviation to the end time index of the current continuous deviation segment. The fourth step is to solidify the time index corresponding to the previous deviation as the end time index of the previous continuous deviation segment and output the start time index and end time index of the previous continuous deviation segment when the time index corresponding to the next deviation is set as the start time index of the new continuous deviation segment, and continue to perform the identification of subsequent adjacent deviations. S2-3. Write the start time index and end time index of the deviation sequence and continuous deviation segments into the deviation index structure in chronological order. Specifically, this includes: constructing a deviation record for each time index based on the alignment key. The deviation record includes at least the alignment key, time index, predicted power value, measured power value, and deviation value. Write the start time index and end time index of the continuous deviation segment as segment fields into the deviation record to form a searchable deviation index entry, which is used to quickly locate the continuous deviation segment within the sliding time window. Perform idempotent append storage using the alignment key as the write index key. When a deviation record with the same index key already exists in the deviation index structure, the corresponding deviation record is replaced. When a deviation record with the same index key does not exist in the deviation index structure, the corresponding deviation record is appended to form a historical deviation record that can be expanded over time. The idempotent append storage process includes: The first step is to query the deviation index structure based on the write index key to see if there is an existing deviation record that matches the write index key. The second step is to write the currently constructed deviation record to the storage location of the existing deviation record when an existing deviation record exists, so as to replace the existing deviation record. The third step is to append the currently constructed deviation record to the end of the deviation index structure or the corresponding index position when there is no existing deviation record. The fourth step is to register the write index key as a written key after the write is completed. This ensures that the same deviation index result is obtained when the write is performed repeatedly on the same write index key, thereby maintaining the uniqueness of the historical deviation record and forming a historical deviation record that can be expanded over time.

[0017] In S3, the process of forming a predictive state model that evolves over time includes: S3-1. Determine the start time index and end time index of the current sliding time window in the deviation index structure, and extract the historical deviation records located within the sliding time window in chronological order. S3-2. For the historical deviation records within the sliding time window, read the deviation values ​​corresponding to each time index in chronological order, and use the cumulative deviation result corresponding to the previous time index as the recursive benchmark to perform a recursive cumulative operation on the deviation value corresponding to the current time index, forming a deviation cumulative sequence arranged in chronological order. After obtaining the cumulative deviation sequence, the cumulative deviation results corresponding to adjacent time indices are compared in chronological order to determine the increasing, decreasing, or alternating change state of the cumulative deviation sequence within the sliding time window, and the change state is determined as the trend of deviation accumulation over time. Incremental means that when comparing the cumulative deviation results corresponding to adjacent time indices in chronological order within the sliding time window, the cumulative deviation result corresponding to the later time index is greater than the cumulative deviation result corresponding to the previous time index in multiple consecutive comparisons, and the cumulative deviation results cover at least one consecutive time index range within the sliding time window in multiple consecutive comparisons. Decreasing means that when comparing the cumulative deviation results corresponding to adjacent time indices in chronological order within a sliding time window, multiple consecutive comparisons satisfy that the cumulative deviation result corresponding to the later time index is less than the cumulative deviation result corresponding to the previous time index, and multiple consecutive comparisons cover at least one consecutive time index range within the sliding time window. Alternating direction means that when comparing the cumulative deviation results corresponding to adjacent time indices in chronological order within a sliding time window, at least once the cumulative deviation result corresponding to the later time index is greater than the cumulative deviation result corresponding to the previous time index and the cumulative deviation result corresponding to the later time index is less than the cumulative deviation result corresponding to the previous time index alternates in adjacent comparisons, so that the cumulative deviation result shows an alternating state of rising and falling within the sliding time window. S3-3. While determining the trend of cumulative deviation, select the deviation values ​​corresponding to two adjacent time indices in the sliding time window in chronological order. Subtract the deviation value corresponding to the previous time index from the deviation value corresponding to the later time index to obtain the deviation change amount of the corresponding time index. Arrange the deviation change amounts in chronological order to form a deviation change amount sequence. After forming the deviation change sequence, the absolute change amplitudes of adjacent changes in the deviation change sequence are compared in chronological order. When the absolute change amplitude does not exceed a preset stable threshold within the continuous time index range, the deviation change is determined to be stable within the continuous time index range. When the absolute change amplitude exceeds a preset drastic threshold at least once within the continuous time index range, the deviation change is determined to have undergone drastic changes within the continuous time index range. The absolute change amplitude refers to the absolute change amplitude of the corresponding time index obtained by selecting two adjacent deviation changes in the deviation change sequence in chronological order, subtracting the previous deviation change from the latter, and taking the absolute value of the result. The stability or drastic change of the determined deviation within the continuous time index range is taken as the fluctuation characteristic of the deviation between adjacent time points.

[0018] In S3, the process of forming a predictive state model that evolves over time also includes: S3-4. Combine the trend of deviation accumulation over time with the fluctuation characteristics of deviation between adjacent time points to construct a prediction state model for the corresponding prediction item, which is used to characterize the running state of the prediction item within the current sliding time window. S3-5. When the sliding time window moves backward and new historical deviation records are written into the deviation index structure, repeat S3-1 to S3-4 to update and build the prediction state model, so that the prediction state model continues to evolve over time. The continuously evolving prediction state model constitutes the prediction reliability dynamic model. The update build refers to: The first step is to move the start and end time indices of the sliding time window backward to the new time range, and then re-extract the historical deviation records located within the new sliding time window from the deviation index structure. The second step is to re-execute the recursive accumulation operation based on the re-extracted historical deviation records to generate a new deviation accumulation sequence, and redetermine the trend of deviation accumulation over time based on the new deviation accumulation sequence. The third step is to re-execute the differential calculation based on the re-extracted historical deviation records to generate a new deviation change sequence, and redetermine the fluctuation characteristics of the deviation between adjacent time points based on the new absolute change amplitude determination results; where differential calculation refers to the calculation process of obtaining the deviation change sequence in S3-3. The fourth step is to replace the trend and fluctuation characteristics corresponding to the previous sliding time window in the prediction state model with the newly determined trend and fluctuation characteristics, so that the prediction state model continues to evolve as the sliding time window moves forward.

[0019] In S4, the process of determining the target prediction entry as the lumped power prediction output includes: S4-1. Call the prediction state model. For each prediction item, read the change trend and fluctuation characteristics of the corresponding prediction item at the current time, and write the change trend and fluctuation characteristics into the state combination item field in the order of the preset state arrangement to form a state combination item for confidence calculation. S4-2. Based on the state combination item, perform a state matching operation in the credibility mapping table to determine the mapping item corresponding to the state combination item, and output the corresponding credibility result according to the mapping item. The credibility mapping table refers to a pre-built table of correspondences between states and credibility. In the table, each mapping item includes at least: a state combination item identifier and its corresponding credibility result. The state matching operation refers to comparing each state combination item with the state combination item identifier in the credibility mapping table one by one. When a state combination item is completely consistent with a certain state combination item identifier, the mapping item corresponding to that state combination item identifier is determined, and the credibility result recorded in the mapping item is output. S4-3. Within the same new energy power station area, the credibility results from different prediction sources are sorted according to the preset comparison rules to obtain the credibility ranking results; The preset comparison rules include: performing pairwise comparisons based on the magnitude of the confidence results, and determining the sorting order of the confidence results based on the comparison results; S4-4. Based on the credibility ranking results, determine the prediction source that ranks first, and select the prediction item corresponding to the prediction source as the target prediction item. The target prediction item is determined as the lumped power prediction output.

[0020] In S5, the process of forming the updated predicted state model includes: S5-1. In the subsequent operation, for the target prediction item, the measured power data corresponding to the target prediction item is continuously read, and the deviation calculation is performed based on the measured power data and the predicted power data. When the calculated measured deviation meets the preset offset condition, the model update process is triggered. S5-2. After the model update process is triggered, the current corresponding prediction state model is frozen, the change trend and fluctuation characteristics of the prediction state model are solidified into the frozen model state, and the frozen model state is prevented from continuing to participate in the subsequent prediction state update. S5-3. After freezing the prediction state model, the newly added measured power data is written into the deviation index structure, and the deviation data within the sliding time window is re-extracted based on the updated historical deviation records as the data input for updating the model. S5-4. Based on the data input of the updated model, the deviation accumulation sequence is regenerated according to the recursive accumulation operation method of S3-2, and the deviation change sequence is regenerated according to the difference calculation and change magnitude determination method of S3-3. Then, the change trend and fluctuation characteristics in the prediction state model are re-determined to form the updated prediction state model. S5-5. Replace the frozen model state with the updated prediction state model, and use the updated prediction state model for the credibility mapping and optimization calculation of subsequent prediction periods to determine the new target prediction item as the lumped power prediction output.

[0021] It should be noted that, including but not limited to: This solution is based on the decision-making problems faced by centralized power prediction platforms in actual operation: During the same prediction period, the centralized control side often receives power prediction results from multiple renewable energy power plants and multiple prediction sources. These results are often close to each other in terms of numerical values, but their stability and reliability will change in different time periods. If only static sorting, fixed rules or one-time accuracy assessment are relied upon, it is easy to continuously use prediction results that are no longer suitable for the current scenario during operation, and there is a lack of clear correction path. Based on this premise, this solution first performs unified parsing and time alignment of multi-source power prediction files on the centralized control side. The purpose is not merely to organize the data, but to standardize prediction results from different prediction sources and time scales into a stable carrier of prediction entries. Each prediction entry is simultaneously bound to the site number, prediction source identifier, prediction time scale, and prediction power sequence arranged in chronological order. The key significance of this is that all subsequent operations related to bias, modeling, evaluation, and optimization always revolve around the same prediction entry, avoiding the problems of confusion or incomparability of multi-source and multi-scale data in subsequent steps. Based on this, a measured power sequence is introduced, and the deviation between the predicted power and the measured power is calculated hourly for each prediction entry. These deviations are then written into a deviation index structure with time index as the main line. The deviation index structure is deliberately used here because the focus is not on the magnitude of the error at a certain moment, but on how the error evolves continuously over time. By retaining continuous segment information during the writing process, the deviation record naturally has a structural basis for whether the deviation is stable or abruptly changes during this period. This provides directly usable historical material for subsequent dynamic modeling without the need to repeatedly scan the original sequence. Next, a fixed and verifiable modeling process is repeatedly executed within the sliding time window, based on the deviation index structure: On the one hand, by recursively accumulating historical deviations within the window, it is observed whether the deviation continuously expands, continuously converges, or frequently reverses direction over time, thus forming the trend of deviation accumulation over time; on the other hand, by comparing the changes in deviation at adjacent moments, it is determined whether the deviation remains stable or fluctuates drastically within a short time scale, thus forming the fluctuation characteristics of deviation between adjacent moments; these two types of results together constitute the prediction state model, which describes the state of the prediction item at the current stage of operation; as the sliding time window moves forward and historical deviations are continuously updated, this prediction state model will also be continuously updated, and its overall evolution process constitutes the dynamic model of prediction credibility; After obtaining the predicted state model, instead of drawing conclusions directly, a credibility mapping table is used to convert the combination of changing trends and fluctuation characteristics into a credibility result under a unified standard. The rationale for this approach is that changing trends and fluctuation characteristics are state descriptions and are not suitable for direct comparison. Through the mapping table, different state combinations can be mapped to explicit credibility results. Thus, within the same renewable energy power station, the credibility results from different prediction sources are ranked and calculated, and the prediction item corresponding to the highest-ranked prediction source is selected as the centralized power prediction output. At this point, the optimal result is no longer a score calculated in a certain instance, but a decision result supported by clear state basis and mapping path. Finally, to address the inevitable environmental changes and model failure risks during operation, an offset triggering mechanism was introduced after the optimal output. When the measured deviation of the target prediction item in subsequent operations meets the preset offset conditions, the system does not simply continue to correct the original model, but first freezes the current prediction state model, solidifying it into a frozen model state to prevent the distorted state from continuing to affect the update process. Subsequently, based on the newly added measured power data, the deviation index structure is rewritten, and the changing trend and fluctuation characteristics are regenerated according to the existing dynamic modeling process to form an updated prediction state model, which then participates in the credibility mapping and optimal evaluation again. This design of freezing to updating and then re-evaluating makes the whole scheme form a complete process: it can continuously utilize historical information and correct deviations in a timely manner at key offset points, avoiding the long-term accumulation of erroneous decisions.

[0022] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for centralized power prediction and selection evaluation of multiple renewable energy power plants, characterized in that, include: S1. Perform parsing and time alignment on the multi-source power prediction files from multiple renewable energy power plants within the same prediction period to generate a set of prediction entries. Each prediction entry includes the power plant number, prediction source identifier, prediction time scale, and prediction power sequence arranged in chronological order. S2. For each prediction item, call the measured power sequence of the corresponding station to perform hourly difference calculation to obtain the deviation sequence, and write the deviation sequence into the deviation index structure in time order to form a historical deviation record that can be expanded over time. S3. Perform dynamic modeling processing on the deviation index structure within the sliding time window to construct a dynamic model of prediction confidence. Dynamic modeling includes: calculating the trend of deviation accumulation over time based on historical deviation records, and simultaneously calculating the fluctuation characteristics of deviation between adjacent moments to form a prediction state model that evolves over time. S4. Call the prediction state model, and obtain the corresponding confidence results of the change trend and fluctuation characteristics of each prediction item at the current time through the corresponding confidence mapping table; perform the best calculation on the confidence results of different prediction sources, and determine the target prediction item as the lumped power prediction output. S5. When the measured deviation of the target prediction item in subsequent operation meets the preset offset condition, the current corresponding prediction state model is frozen, and the prediction state model is updated based on the newly added measured power data to form the updated prediction state model.

2. The method for centralized power prediction and selection evaluation of multiple new energy power plants according to claim 1, characterized in that: In S1, the process of generating the set of prediction entries includes: S1-1. Read the multi-source power prediction files from multiple new energy power plants on the centralized control side, extract the identification field used to identify the prediction attribution relationship in each power prediction file, and extract the predicted power data carried in the power prediction file. The identification field includes the power plant number, prediction source identification and prediction time scale. S1-2. Based on the prediction time scale, perform time axis alignment processing on the predicted power data in different power prediction files to form a predicted power data sequence under a unified time index. S1-3. Bind the time-aligned predicted power data sequence with the corresponding site number, prediction source identifier and prediction time scale to construct prediction entries, and then aggregate the prediction entries to form a prediction entry set.

3. The method for centralized power prediction and selection evaluation of multiple new energy power plants according to claim 2, characterized in that: In S2, the process of creating a historical deviation record that can be expanded over time includes: S2-1. For each prediction item, read the predicted power sequence in the prediction item and read the measured power sequence of the corresponding new energy power station. Alignment keys are constructed using the site number of the prediction entry, the prediction source identifier, the prediction time scale, and the time index in the prediction power sequence. Based on the alignment keys, the measured power values ​​under the same time index are located in the measured power sequence to form hourly paired prediction power values ​​and measured power values. S2-2. Perform a difference calculation on the predicted power value and the measured power value under each time index to generate a deviation sequence that corresponds one-to-one with the time index. While generating the deviation sequence, continuous segment identification is performed on the deviation sequence in chronological order to determine whether the deviation change of adjacent time indices in the deviation sequence exceeds a preset change threshold. If it does not exceed the threshold, adjacent deviations are merged into the same continuous deviation segment. If it exceeds the threshold, it is divided into new continuous deviation segments to output the start time index and end time index of the continuous deviation segment. S2-3. Write the start time index and end time index of the deviation sequence and continuous deviation segments into the deviation index structure in chronological order. Specifically, this includes: constructing a deviation record for each time index based on the alignment key. The deviation record includes at least the alignment key, time index, predicted power value, measured power value, and deviation value. Write the start time index and end time index of the continuous deviation segments as segment fields into the deviation record to form a searchable deviation index entry. Perform idempotent append storage using the alignment key as the write index key. If a deviation record with the same index key already exists in the deviation index structure, the corresponding deviation record is replaced. If no deviation record with the same index key exists in the deviation index structure, the corresponding deviation record is appended to form a historical deviation record that can be expanded over time.

4. The method for centralized power prediction and selection evaluation of multiple new energy power plants according to claim 3, characterized in that: In S3, the process of forming a predictive state model that evolves over time includes: S3-1. Determine the start time index and end time index of the current sliding time window in the deviation index structure, and extract the historical deviation records located within the sliding time window in chronological order. S3-2. For the historical deviation records within the sliding time window, read the deviation values ​​corresponding to each time index in chronological order, and use the cumulative deviation result corresponding to the previous time index as the recursive benchmark to perform a recursive cumulative operation on the deviation value corresponding to the current time index, forming a deviation cumulative sequence arranged in chronological order. After obtaining the cumulative deviation sequence, the cumulative deviation results corresponding to adjacent time indices are compared in chronological order to determine the increasing, decreasing, or alternating change state of the cumulative deviation sequence within the sliding time window, and the change state is determined as the trend of deviation accumulation over time. S3-3. While determining the trend of cumulative deviation, select the deviation values ​​corresponding to two adjacent time indices in the sliding time window in chronological order. Subtract the deviation value corresponding to the previous time index from the deviation value corresponding to the later time index to obtain the deviation change amount of the corresponding time index. Arrange the deviation change amounts in chronological order to form a deviation change amount sequence. After forming the deviation change sequence, the absolute change magnitudes of adjacent changes in the deviation change sequence are compared in chronological order. When the absolute change magnitudes do not exceed the preset stable threshold within the continuous time index range, it is determined that the deviation change magnitude remains stable within the continuous time index range. When the absolute change magnitudes exceed the preset drastic threshold at least once within the continuous time index range, it is determined that the deviation change magnitude has undergone drastic changes within the continuous time index range. The stability or drastic change of the determined deviation within the continuous time index range is taken as the fluctuation characteristic of the deviation between adjacent time points.

5. The method for centralized power prediction and selection evaluation of multiple new energy power plants according to claim 4, characterized in that: In S3, the process of forming a predictive state model that evolves over time also includes: S3-4. Combine the trend of the cumulative change of the deviation over time with the fluctuation characteristics of the deviation between adjacent time points to construct the prediction state model for the corresponding prediction item. S3-5. When the sliding time window moves backward and new historical deviation records are written into the deviation index structure, repeat S3-1 to S3-4 to update and build the prediction state model so that the prediction state model continues to evolve over time. The update build refers to: The first step is to move the start and end time indices of the sliding time window backward to the new time range, and then re-extract the historical deviation records located within the new sliding time window from the deviation index structure. The second step is to re-execute the recursive accumulation operation based on the re-extracted historical deviation records to generate a new deviation accumulation sequence, and redetermine the trend of deviation accumulation over time based on the new deviation accumulation sequence. The third step is to re-perform the differential calculation based on the re-extracted historical deviation records to generate a new sequence of deviation changes, and to redetermine the fluctuation characteristics of the deviation between adjacent time points based on the new absolute change magnitude determination results. The fourth step is to replace the trend and fluctuation characteristics corresponding to the previous sliding time window in the prediction state model with the newly determined trend and fluctuation characteristics, so that the prediction state model continues to evolve as the sliding time window moves forward.

6. The method for centralized power prediction and selection evaluation of multiple new energy power plants according to claim 5, characterized in that: In S4, the process of determining the target prediction entry as the lumped power prediction output includes: S4-1. Call the prediction state model. For each prediction item, read the change trend and fluctuation characteristics of the corresponding prediction item at the current time, and write the change trend and fluctuation characteristics into the state combination item field in the order of the preset state arrangement to form a state combination item for confidence calculation. S4-2. Based on the state combination item, perform a state matching operation in the credibility mapping table to determine the mapping item corresponding to the state combination item, and output the corresponding credibility result according to the mapping item. S4-3. Within the same new energy power station area, the credibility results from different prediction sources are sorted according to the preset comparison rules to obtain the credibility ranking results; The preset comparison rules include: performing pairwise comparisons based on the magnitude of the confidence results, and determining the sorting order of the confidence results based on the comparison results; S4-4. Based on the credibility ranking results, determine the prediction source that ranks first, and select the prediction item corresponding to the prediction source as the target prediction item. The target prediction item is determined as the lumped power prediction output.

7. The method for centralized power prediction and selection evaluation of multiple new energy power plants according to claim 6, characterized in that: In S5, the process of forming the updated predicted state model includes: S5-1. In the subsequent operation, for the target prediction item, the measured power data corresponding to the target prediction item is continuously read, and the deviation calculation is performed based on the measured power data and the predicted power data. When the calculated measured deviation meets the preset offset condition, the model update process is triggered. S5-2. After the model update process is triggered, the current corresponding prediction state model is frozen, the change trend and fluctuation characteristics of the prediction state model are solidified into the frozen model state, and the frozen model state is prevented from continuing to participate in the subsequent prediction state update. S5-3. After freezing the prediction state model, the newly added measured power data is written into the deviation index structure, and the deviation data within the sliding time window is re-extracted based on the updated historical deviation records as the data input for updating the model. S5-4. Based on the data input of the updated model, the deviation accumulation sequence is regenerated according to the recursive accumulation operation method of S3-2, and the deviation change sequence is regenerated according to the difference calculation and change magnitude determination method of S3-3. Then, the change trend and fluctuation characteristics in the prediction state model are re-determined to form the updated prediction state model. S5-5. Replace the frozen model state with the updated prediction state model, and use the updated prediction state model for the credibility mapping and optimization calculation of subsequent prediction periods to determine the new target prediction item as the lumped power prediction output.

Images