New energy generation power prediction method and system based on meteorological data

Abstract

The invention provides a new energy power generation power prediction method and system based on meteorological data, and relates to the technical field of new energy power generation, and the method comprises the steps: firstly obtaining a meteorological data set (including continuous collection data of various meteorological elements) and a working condition data set (including operation parameter data of the equipment in different operation states) corresponding to new energy power generation equipment; then, correlation analysis is carried out on the meteorological data set, and a meteorological element time sequence correlation graph is established to represent a meteorological element time dimension mutual influence relation; and then carrying out association mapping on the meteorological element time sequence association diagram and the working condition data set to generate working condition response association features. And constructing a generated power linkage prediction model based on the working condition response correlation characteristics, and predicting the real-time meteorological element data to obtain an initial prediction result. And finally, performing association correction on the initial prediction result and the historical generated power data to generate a final prediction result, thereby realizing accurate prediction of the new energy generated power.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, and more specifically, to a method and system for predicting new energy power generation capacity based on meteorological data. Background Technology

[0002] In the field of new energy power generation, such as wind power and solar power, accurate prediction of power generation capacity is crucial for the stable operation of the power grid, the rational allocation of power resources, and the efficient utilization of new energy power generation equipment. Traditional methods for predicting new energy power generation capacity mainly rely on modeling and prediction based on single meteorological data or simple equipment operating parameters.

[0003] On the one hand, considering only a single meteorological data point can lead to an inaccurate overall understanding of meteorological conditions due to the complex interactions and temporal relationships among meteorological elements, thus affecting the accuracy of power generation forecasts. For example, in wind power generation, considering only wind speed while ignoring the synergistic effects of other meteorological elements such as wind direction and air pressure makes it difficult to accurately predict the actual power generation of wind turbine generators.

[0004] On the other hand, while forecasting based solely on equipment operating parameters can reflect the current operating status of the equipment, it cannot fully consider the potential impact of changes in meteorological conditions on equipment operation. Moreover, traditional methods lack in-depth mining and effective utilization of historical data during the forecasting process, and fail to correlate and correct real-time forecast results with historical data. This results in a significant deviation between the forecast results and the actual power generation, failing to meet the high-precision requirements of modern power grids for forecasting new energy power generation. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for predicting the power generation of new energy sources based on meteorological data.

[0006] Furthermore, embodiments of the present invention also provide a new energy power generation prediction system based on meteorological data.

[0007] Based on the above, this embodiment of the invention acquires meteorological data sets and operating condition data sets corresponding to new energy power generation equipment, performs correlation analysis on the meteorological data sets, and establishes a time-series correlation diagram of meteorological elements. This depicts the mutual influence relationship between different meteorological elements in the time dimension. The time-series correlation diagram of meteorological elements is then mapped to the operating condition data sets to generate operating condition response correlation features. This effectively reveals the inherent correspondence between changes in meteorological elements and changes in equipment operating parameters, achieving a deep integration of meteorological conditions and equipment operation. A power generation linkage prediction model is constructed based on the operating condition response correlation features, which can fully utilize the synergistic effect of meteorological elements and equipment operating conditions to predict power generation, improving the accuracy and reliability of the prediction. Finally, the initial prediction results of new energy power generation are correlated and corrected with historical power generation data, further eliminating prediction errors and generating a final prediction result of new energy power generation that is closer to reality, significantly improving the accuracy of new energy power generation prediction. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the new energy power generation prediction method based on meteorological data provided in the embodiments of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of a new energy power generation prediction system based on meteorological data provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a new energy power generation prediction method based on meteorological data, provided in one embodiment of the present invention. The following is a detailed description of this new energy power generation prediction method based on meteorological data.

[0011] Step S110: Obtain the meteorological data set corresponding to the new energy power generation equipment and the operating condition data set of the new energy power generation equipment. The meteorological data set contains continuously collected data of multiple meteorological elements, and the operating condition data set contains the operating parameter data of the new energy power generation equipment under different operating conditions.

[0012] This embodiment uses a large-scale ground-mounted solar photovoltaic power station as an application scenario. In this scenario, the new energy power generation equipment includes photovoltaic power generation arrays and supporting equipment such as inverters and combiner boxes. Meteorological data is acquired through meteorological monitoring stations deployed within the photovoltaic power station. These monitoring stations are set up in different areas of the power station according to a preset spatial distribution rule to ensure that the collected data can comprehensively reflect the overall meteorological conditions of the power station.

[0013] The meteorological dataset includes various meteorological elements such as solar irradiance, ambient temperature, wind speed, wind direction, relative humidity, and atmospheric pressure. Solar irradiance is continuously collected using a total radiation meter; ambient temperature and relative humidity are obtained using temperature and humidity sensors; wind speed and direction are monitored by wind speed and direction sensors; and atmospheric pressure is collected using a barometric pressure sensor. All meteorological element data are continuously recorded at fixed time intervals, forming a continuous data sequence.

[0014] Step S120: Perform correlation analysis on the continuously collected data of multiple meteorological elements in the meteorological data set to establish a time-series correlation diagram of meteorological elements. The time-series correlation diagram of meteorological elements is used to characterize the mutual influence relationship of different meteorological elements in the time dimension.

[0015] First, the continuously collected data of various meteorological elements are processed to remove obvious data collection errors, such as values ​​exceeding the physical reasonable range. Then, based on time series analysis methods, the correlations between different meteorological elements at the same time point and between different time points are studied. By calculating relevant statistics, the degree and direction of influence among various meteorological elements are determined, and a time series correlation diagram reflecting these relationships is constructed.

[0016] Step S121: Extract the time series information of the continuous collection data of multiple meteorological elements in the meteorological data set, determine the time collection interval and time series length of the continuous collection data of each meteorological element, and mark the abnormal time nodes in the time series information. The abnormal time nodes are the time nodes corresponding to the interruption of data collection or the absence of data.

[0017] Continuous data collection of various meteorological elements, such as solar irradiance, ambient temperature, and wind speed, is extracted from the meteorological data set to form their respective time series information. For solar irradiance data, the time collection interval is set to 1 minute, that is, the solar irradiance value is recorded once every minute. By statistically analyzing the start and end times of the recorded data, the length of the time series can be determined. For example, if a time series contains 720 consecutive data records, the corresponding time series length is 720 minutes.

[0018] Step S122: Complete the marked time series information by supplementing the meteorological element data corresponding to the abnormal time node according to the change trend of meteorological element data before and after the abnormal time node, and forming a complete time series information.

[0019] For the marked abnormal time nodes, interpolation is used to complete the data. Taking ambient temperature data as an example, if 10:05-10:07 is an abnormal time node, the ambient temperature at 10:04 is 25℃, and the ambient temperature at 10:08 is 26℃. Moreover, by observing the temperature change trend before 10:04, it can be seen that the temperature is slowly rising.

[0020] Step S123: Based on the completed time series information, calculate the correlation parameter between the continuously collected data of different meteorological elements at the same time node. The correlation parameter is used to quantify the degree of mutual influence between different meteorological elements.

[0021] Using the completed time series information, we selected multiple combinations of meteorological elements, including solar irradiance and ambient temperature, solar irradiance and wind speed, and ambient temperature and atmospheric pressure. For each combination of meteorological elements, we obtained the values ​​of the two types of meteorological elements at each same time point.

[0022] The correlation between these values ​​is calculated using statistical methods to obtain a correlation parameter. For example, at multiple time points during midday, when solar irradiance is high, ambient temperature is also high. Calculations show that the correlation parameter between solar irradiance and ambient temperature is high at this time, indicating that the two have a significant mutual influence during this period.

[0023] Step S1231: Extract all the same time nodes from the completed time series information, determine the values ​​of continuously collected data of different meteorological elements corresponding to each same time node, and establish a time node-meteorological element value correspondence table.

[0024] By iterating through the completed time series information of various meteorological elements, the common time nodes of all meteorological elements are extracted. These time nodes are the same time nodes. For example, meteorological elements such as solar irradiance, ambient temperature, and wind speed all have data records at 00:00, 00:01, ..., 23:59 every day, and these time nodes are the same time nodes.

[0025] Step S1232: Select two different types of continuously collected meteorological data as a group of analysis objects. Select the continuous collected data of wind meteorological elements and the continuous collected data of light meteorological elements as the first group of analysis objects. Select the continuous collected data of wind meteorological elements and the continuous collected data of air pressure meteorological elements as the second group of analysis objects. Select the continuous collected data of light meteorological elements and the continuous collected data of air pressure meteorological elements as the third group of analysis objects.

[0026] In this embodiment, the continuously collected meteorological data for wind-related elements is wind speed data, the continuously collected meteorological data for light-related elements is solar irradiance data, and the continuously collected meteorological data for air pressure-related elements is atmospheric pressure data. The first set of analysis objects is wind speed data and solar irradiance data. Through the analysis of this set of data, the correlation between changes in wind speed and changes in solar irradiance is explored.

[0027] The second set of analyses focuses on wind speed and atmospheric pressure data to study the interaction between them. The third set of analyses focuses on solar irradiance and atmospheric pressure data to analyze the relationship between changes in solar irradiance and atmospheric pressure.

[0028] Step S1233: For each group of analysis objects, extract the values ​​of the group of analysis objects at each same time node from the time node-meteorological element value correspondence table to form a value sequence pair of the group of analysis objects.

[0029] Taking the first set of analysis objects (wind speed data and solar irradiance data) as an example, the wind speed value and solar irradiance value corresponding to each time node are extracted sequentially from the time node-meteorological element value correspondence table. The wind speed value and solar irradiance value at the same time node are paired together and arranged in chronological order of the time nodes to form a wind speed-solar irradiance value sequence pair.

[0030] Similarly, for the second set of analytical objects (wind speed data and atmospheric pressure data), the wind speed and atmospheric pressure values ​​at each time point are extracted to form a wind speed-atmospheric pressure numerical sequence pair. The third set of analytical objects forms a solar irradiance-atmospheric pressure numerical sequence pair.

[0031] Step S1234: Calculate the covariance of each set of numerical sequences of the analysis objects at each same time node. The covariance is used to reflect the consistency of the changing trend of the two types of meteorological element data values. The validity of the calculated covariance is judged, and covariance values ​​with absolute values ​​exceeding the preset covariance range are removed.

[0032] For the first set of analysis objects, the wind speed-solar irradiance numerical sequence pairs are analyzed. At each same time node, the covariance is calculated based on the wind speed value, solar irradiance value, and the average value of these two factors across all time nodes. The calculation logic for the covariance is as follows: multiply (wind speed value - average wind speed) and (solar irradiance value - average solar irradiance) at each time node, sum the results, and then divide by the number of time nodes minus one.

[0033] Step S1235: Based on the covariance after validity assessment, further calculate the correlation coefficient. The correlation coefficient is the ratio of the covariance to the product of the standard deviations of the two types of meteorological element data. Record the correlation coefficient of each group of analysis objects at each same time node.

[0034] After obtaining the covariance values ​​for validity assessment, the standard deviations of wind speed data and solar irradiance data are calculated separately. The calculation logic for the standard deviation is as follows: sum the squares of (feature value - feature average) at each time point, divide by the number of time points minus one, and then take the square root.

[0035] Then, the covariance value at each time point is divided by the product of the standard deviation of the wind speed data and the standard deviation of the solar irradiance data to obtain the correlation coefficient at that time point. Following this method, the correlation coefficient of each group of analyzed objects at each same time point is calculated and recorded.

[0036] Step S1236: Normalize the correlation coefficient by using linear normalization to convert the range of the correlation coefficient to between 0 and 1. The normalized correlation coefficient is the correlation degree parameter. Following the above steps, calculate the correlation degree parameter of all groups of analysis objects at each same time node to form a correlation degree parameter sequence for each group of analysis objects.

[0037] For each group of analyzed objects, determine the minimum and maximum values ​​of the correlation coefficients. Use a linear normalization formula to convert each correlation coefficient into a value between 0 and 1. The logic of the normalization formula is: (correlation coefficient - minimum value) ÷ (maximum value - minimum value). The normalized value is the correlation parameter.

[0038] Following the same steps, calculate the correlation parameters for the second and third groups of analysis objects at each same time point. Arrange the correlation parameters of each group of analysis objects in chronological order to form their respective correlation parameter sequences.

[0039] Step S124: Construct an initial meteorological element association diagram based on the association degree parameter. The nodes in the initial meteorological element association diagram are different types of meteorological elements, and the lines between the nodes are the corresponding association degree parameters. Delete invalid lines in the initial meteorological element association diagram. Invalid lines are those whose association degree parameters are lower than a preset association threshold.

[0040] Different types of meteorological elements, such as solar irradiance, ambient temperature, wind speed, and atmospheric pressure, are used as nodes in the initial meteorological element correlation diagram. For every two sets of meteorological element nodes, a line is drawn between the nodes based on the correlation parameters calculated earlier, and the correlation parameters are labeled on the line.

[0041] Step S125: Perform time-series dimension expansion processing on the initial meteorological element association map after deleting invalid connections, and connect the initial meteorological element association maps at different time nodes in chronological order to form a meteorological element time-series association framework containing time dimension information.

[0042] The day is divided into multiple time nodes, and each time node has a corresponding initial meteorological element correlation diagram after removing invalid connections. The above initial meteorological element correlation diagrams are arranged sequentially according to the time nodes from morning to night.

[0043] During the arrangement process, the initial meteorological element correlation diagrams of adjacent time nodes are connected by a time axis to form a framework structure. This framework structure can reflect the changes in the correlation relationships of meteorological elements at different time nodes, thus constituting a time-series correlation framework of meteorological elements containing time dimension information.

[0044] Step S126: In the meteorological element time series association framework, according to the changing trend of the meteorological element association degree parameter between different time nodes, supplement the time dimension association edge, and label the corresponding time span information for each time dimension association edge. The time dimension association edge is used to characterize the influence relationship of the same meteorological element between different time nodes.

[0045] Observe the changes in the correlation parameters of the same meteorological element at different time points within the temporal correlation framework. For example, observe the changing trends of the correlation parameters between solar irradiance and other meteorological elements at time points such as 9:00, 10:00, and 11:00. When it is found that the correlation characteristics of the same meteorological element are continuous or influential between adjacent time points or time intervals, add time-dimensional correlation edges.

[0046] Step S127: Integrate the initial meteorological element association graph, the meteorological element temporal association framework, and the time dimension association edges to generate the meteorological element temporal association graph. The meteorological element temporal association graph includes node type information, node association degree parameter information, temporal association edge information, and time span information.

[0047] The initial meteorological element correlation diagram, after removing invalid connections, is used as a basis. A time-dimensional structure from the meteorological element temporal correlation framework is incorporated, and supplementary time-dimensional correlation edges are added. In the integrated diagram, the type of each node is clearly labeled, such as solar irradiance nodes, ambient temperature nodes, etc.

[0048] For the connections between nodes, the corresponding correlation parameters are labeled; for time-dimensional correlation edges, the corresponding time span information is labeled. Through the above integration, a complete meteorological element time-series correlation diagram is formed, which can comprehensively reflect the mutual influence relationship and time span of different meteorological elements at different time nodes.

[0049] Step S130: Perform correlation mapping processing between the meteorological element time series correlation diagram and the operating condition data set to generate the operating condition response correlation feature of the new energy power generation equipment. The operating condition response correlation feature is used to characterize the correspondence between changes in meteorological elements and changes in equipment operating parameters.

[0050] In this embodiment, after obtaining the time-series correlation diagram of meteorological elements, it is necessary to establish a correlation between it and the operating condition data set of the solar photovoltaic power station. The time-series correlation diagram of meteorological elements clearly shows the mutual influence of meteorological elements such as solar irradiance, ambient temperature, and wind speed over time, while the operating condition data set covers the operating parameters of the photovoltaic array, such as DC-side voltage, DC-side current, inverter conversion efficiency, and panel temperature. Through correlation mapping processing, it can be clearly determined how these operating parameters change accordingly when meteorological elements change.

[0051] Step S131: Extract the operating parameter change information from the operating condition data set, and filter the abnormal change data in the operating parameter change information. Abnormal change data is data whose numerical change range exceeds a preset range. The operating parameter change information includes the numerical change range and the numerical change rate of the operating parameters. The numerical change range is the difference between the operating parameter values ​​at different time points, and the numerical change rate is the ratio of the numerical change range to the time interval.

[0052] When extracting operational parameter change information from the operating data set, the focus is first on the key operating parameters of the photovoltaic array. Taking DC-side voltage as an example, voltage values ​​at different time points are collected, and the magnitude of the change between adjacent time points is calculated, i.e., the voltage value at the later time point minus the voltage value at the previous time point. The rate of change is the magnitude of change divided by the time interval between the two time points. For example, if the adjacent acquisition interval is 5 minutes, and the voltage change is 20 volts, then the rate of change is 4 volts per minute.

[0053] Step S132: Calculate the rate of change of meteorological element values ​​and the rate of change of operating parameter values, wherein the rate of change is defined as the relative change of the values ​​with respect to the reference values.

[0054] We will take solar irradiance from meteorological elements and DC-side current from operational parameters as examples for calculation. We will set the baseline value of solar irradiance as a value under a certain steady state, such as 1000 watts per square meter. At a certain point in time, the actual value of solar irradiance is 800 watts per square meter. Therefore, its rate of change is (800-1000)÷1000=-0.2, or -20%.

[0055] For the DC side current, the reference value is set to 20 amperes. When the actual value is 18 amperes, its rate of change is (18-20)÷20=-0.1, that is, the rate of change is -10%. Through the above calculation method, the changes in meteorological elements and operating parameters are converted into the rate of change relative to the reference value, eliminating the difference in dimensions and facilitating subsequent correlation and matching.

[0056] Step S133: Establish a mapping relationship table, match the rate of change of meteorological element values ​​with the rate of change of operating parameter values, and label the corresponding time span information for each matching item.

[0057] A mapping table is established using the rate of change of solar irradiance, the rate of change of ambient temperature, the rate of change of DC-side voltage, and the rate of change of inverter conversion efficiency as examples. Within a certain time period, the rate of change of solar irradiance is -15%, the rate of change of ambient temperature is 5%, the corresponding rate of change of DC-side voltage is -10%, and the rate of change of inverter conversion efficiency is -2%. This set of matching relationships is recorded.

[0058] Step S134: Optimize the mapping relationship table, calculate the matching degree of each matching item, delete matching items with a matching degree lower than the preset matching threshold, and obtain the optimized mapping relationship table. The matching degree is the degree of correspondence between the rate of change of meteorological element values ​​and the rate of change of operating parameter values.

[0059] When calculating the matching degree, the overall degree of fit between the combination of meteorological factor change rates and the combination of operating parameter change rates is comprehensively considered. For example, for a certain matching item, the solar irradiance change rate is -12%, the ambient temperature change rate is 4%, the corresponding DC side voltage change rate is -8%, and the inverter conversion efficiency change rate is -1.5%. The degree of fit between the two is calculated using a preset algorithm. If the matching degree of this matching item is 0.85, then the matching degree is considered to be within the range of 0.85.

[0060] A preset matching threshold of 0.7 was set. Since 0.85 is higher than the threshold, the match was retained. If the matching degree of a match is 0.6, which is lower than the threshold, it was deleted from the mapping table. After the above optimization, the matches in the mapping table all have a high degree of fit, improving the reliability of subsequent analysis.

[0061] Step S135: Based on the optimized mapping table, calculate the probability distribution of the operating parameter value change rate corresponding to the value change rate of each meteorological element. The probability distribution is used to represent the possibility that the operating parameter will have different value change rates under a specific meteorological element value change rate.

[0062] Taking the rate of change of solar irradiance as an example, all matching items with a rate of change of solar irradiance ranging from -10% to -15% were extracted from the optimized mapping table. The different intervals of the rate of change of DC current and their corresponding numbers within these matching items were then statistically analyzed.

[0063] Assuming that within the range of solar irradiance variation rate, there are 20 matching terms for the DC-side current variation rate in the -8% to -12% range, and 10 matching terms in the -12% to -16% range, for a total of 50 matching terms. Then, the probability of the DC-side current variation rate being in the -8% to -12% range is 20 ÷ 50 = 0.4, and the probability of it being in the -12% to -16% range is 10 ÷ 50 = 0.2. Through the above calculations, the probability distribution of the DC-side current variation rate under a specific solar irradiance variation rate is obtained.

[0064] Step S136: Based on the probability distribution and the correlation parameters in the time series correlation diagram of the meteorological elements, the change information of the operating parameters is weighted and processed to obtain weighted change information of the operating parameters, with the weight values ​​being the corresponding correlation parameters.

[0065] Taking the impact of ambient temperature changes on inverter conversion efficiency as an example, the correlation parameter between ambient temperature and solar irradiance in the meteorological element time-series correlation diagram is 0.6, and the correlation parameter between ambient temperature and wind speed is 0.3. In the operating parameter change information, the inverter conversion efficiency change rate is -3%.

[0066] The inverter conversion efficiency change is weighted based on the correlation parameters. If the correlation between ambient temperature and solar irradiance is the primary consideration, the weighted change is -3% × 0.6 = -1.8%. If the correlation with wind speed is also considered, a combined weighting can be applied based on the actual situation. This weighting process makes the changes in operating parameters more reflective of their importance in relation to meteorological factors.

[0067] Step S137: Combine the weighted operating parameter change information with the time series association direction and time span information in the meteorological element time series association diagram to generate the operating condition response association feature containing the time dimension change trend. The operating condition response association feature includes the meteorological element change sequence, the corresponding operating parameter change sequence, the association weight sequence of the two, and the time series span sequence.

[0068] Collect solar irradiance variation sequences over a period of time, such as [-5%, -10%, -15%, -12%], with corresponding weighted DC-side voltage variation sequences of [-3%, -7%, -11%, -9%]. The correlation weight sequences are determined based on the correlation parameters in the meteorological element time-series correlation diagram, such as [0.7, 0.65, 0.75, 0.68]. The time-series span sequences are the time intervals between each period, such as [10 minutes, 10 minutes, 10 minutes].

[0069] By combining the above sequences, each value in the solar irradiance variation sequence corresponds to a value in the weighted DC side voltage variation sequence. At the same time, the associated weight sequence and time span sequence provide information on the correlation strength and time interval. The resulting operating condition response correlation feature fully presents the changing trends and correlations between meteorological elements and operating parameters in the time dimension.

[0070] Step S140: Construct a power generation linkage prediction model based on the operating condition response correlation features, and perform prediction analysis on the real-time meteorological element data in the meteorological data set through the power generation linkage prediction model to obtain the initial prediction result of new energy power generation.

[0071] A power generation prediction model is constructed using the obtained operating condition response correlation features. This model takes changes in meteorological elements as input and, combined with the correlations inherent in the operating condition response correlation features, predicts the power generation of the photovoltaic power plant. When real-time meteorological data is input into the model, the model analyzes the real-time changes in meteorological elements and predicts the corresponding changes in power generation based on the correlation features, thus obtaining the initial power generation prediction result.

[0072] Step S141: Extract the meteorological element change sequence, operating parameter change sequence, correlation weight sequence, and time series span sequence from the working condition response correlation features. Use the meteorological element change sequence as the model input feature, the operating parameter change sequence as the model intermediate feature, the correlation weight sequence as the model weight parameter, and the time series span sequence as the model time dimension adjustment parameter.

[0073] The various sequences are separated from the operating condition response correlation features. The meteorological element change sequences, including changes in solar irradiance and ambient temperature, are set as input features of the model to receive real-time meteorological data changes. The operating parameter change sequences, such as changes in DC current and voltage, are used as intermediate features of the model and participate in the calculations within the model.

[0074] The correlation weight sequence determines the importance of the association between different meteorological elements and operational parameters, and as a model weight parameter, it affects the proportion of each feature in the calculation. The time span sequence is used to adjust the model's sensitivity to data changes at different time intervals, enabling the model to adapt to changes in the time dimension.

[0075] Step S142: Construct the network structure of the power generation linkage prediction model. The network structure includes an input layer, an intermediate feature processing layer, a time dimension adjustment layer, and an output layer. The input layer is used to receive the input features of the model. The intermediate feature processing layer is used to perform correlation operations on the intermediate features of the model. The time dimension adjustment layer is used to adjust the sensitivity of the model to time dimension information according to the time dimension adjustment parameters. The output layer is used to output the predicted power generation value.

[0076] The input layer contains multiple neurons, the number of which matches the dimensionality of the meteorological element change sequence. These neurons are responsible for receiving the input data from the meteorological element change sequence. The intermediate feature processing layer contains multiple processing units that perform correlation operations on the operating parameter change sequence. It then combines the correlation weight sequence to weight different features, strengthening the influence of important correlated features.

[0077] The time-dimension adjustment layer adjusts the focus on data at different time points according to the time series sequence, and gives appropriate weight adjustments to changes with large time spans. The output layer integrates the processing results of each layer and outputs the corresponding power generation prediction value, completing the prediction process from input to output.

[0078] Step S143: Import the associated weight sequence into the intermediate feature processing layer as a weighting coefficient in the intermediate feature operation process, so that when the intermediate feature processing layer processes the sequence of changes in running parameters, it performs feature enhancement or weakening according to the weight value of the associated weight sequence.

[0079] The values ​​in the associated weight sequence are distributed to the various computational units of the intermediate feature processing layer. For example, if the associated weight corresponding to a certain sequence of changes in operating parameters is 0.8, then when the intermediate feature processing layer performs operations on that sequence of changes in operating parameters, its value is multiplied by 0.8 to strengthen the influence of that feature; if the associated weight is 0.3, then it is multiplied by 0.3 to weaken its influence.

[0080] Through the above weighted processing, the intermediate feature processing layer can highlight the features with high correlation between meteorological elements and operational parameters, reduce the role of features with low correlation in the calculation, make the model pay more attention to important correlation information, and improve the accuracy of prediction.

[0081] Step S144: Import the time span sequence into the time dimension adjustment layer, and adjust the model's response weights to changes in meteorological elements under different time spans through the time dimension adjustment layer.

[0082] The time interval information in the time-series sequence is input into the time dimension adjustment layer, where corresponding adjustment mechanisms are set. For sequences with shorter time spans, it indicates that meteorological elements change more frequently, and the model sets a higher weight for the changes within that time period to capture rapidly changing trends; for sequences with longer time spans, meteorological elements change relatively slowly, and the response weights are set lower.

[0083] For example, a sequence with a time span of 5 minutes corresponds to a response weight of 0.9, while a sequence with a time span of 30 minutes corresponds to a response weight of 0.5. Through these adjustments, the time dimension adjustment layer enables the model to allocate attention reasonably according to the time span, better adapting to changes in meteorological elements at different time scales.

[0084] Step S145: Train the power generation prediction model. The training data consists of historical meteorological element data and corresponding historical power generation data in the meteorological data set. The training objective is to ensure that the deviation between the power generation prediction value output by the model and the historical power generation data is within a preset range.

[0085] Historical meteorological data from the past year, such as solar irradiance, ambient temperature, and wind speed, along with corresponding historical power generation data from photovoltaic power plants, are selected as training data. This historical meteorological data is then input into the power generation prediction model, which outputs the predicted power generation value.

[0086] The deviation between the predicted value and historical power generation data is calculated. By adjusting the model's network parameters, such as the weighting coefficients of the intermediate feature processing layer and the response weights of the time dimension adjustment layer, the deviation is gradually reduced. The preset deviation range is ±5%. When the deviation between the model's output predicted value and the historical data is mostly within this range, the training has achieved its goal.

[0087] Step S146: After training, the real-time meteorological element data in the meteorological data set is input into the input layer of the power generation linkage prediction model. The intermediate feature processing layer performs calculations on the changes in the operating parameters corresponding to the real-time meteorological element data. The time dimension adjustment layer adjusts the time dimension information during the calculation process. The output layer outputs the corresponding power generation prediction value. The power generation prediction values ​​corresponding to multiple real-time meteorological element data are arranged in chronological order to form the initial prediction result of the new energy power generation.

[0088] The trained model is capable of processing real-time data, inputting real-time meteorological data such as solar irradiance and ambient temperature into the input layer. The intermediate feature processing layer performs calculations on the changes in corresponding operating parameters based on the real-time meteorological data and the correlations obtained during model training, yielding intermediate calculation results.

[0089] The time-dimension adjustment layer adjusts the intermediate calculation results in the time dimension based on the time span information of the real-time data. Finally, the output layer integrates the adjusted results and outputs the power generation prediction value corresponding to each time node. These prediction values ​​are arranged in chronological order, such as one prediction value every 10 minutes, to form the initial prediction results for new energy power generation.

[0090] Step S1451: Extract historical meteorological element data from the meteorological data set. The historical meteorological element data is continuously collected meteorological element data within a preset historical time period. Divide the historical meteorological element data into training dataset, validation dataset, and test dataset according to the time sequence. The division ratio is determined according to the preset dataset allocation ratio.

[0091] The historical time period is preset to the past two years. Continuously collected data such as solar irradiance, ambient temperature, and wind speed from these two years are extracted from the meteorological data set as historical meteorological element data. The historical meteorological element data is divided into training, validation, and test datasets according to a preset dataset allocation ratio of 7:2:1.

[0092] For example, if the total data volume is 730 days (two years), the training dataset contains 511 days of data, the validation dataset contains 146 days of data, and the test dataset contains 73 days of data. The partitioning process strictly follows the chronological order to ensure the temporal continuity of each dataset and avoid data corruption affecting the model training effect.

[0093] Step S1452: Extract historical power generation data corresponding to the historical meteorological element data from the operating condition data set, so that the historical power generation data and the historical meteorological element data correspond one-to-one at the time node, and divide the dataset into training label set, verification label set and test label set according to the dataset allocation ratio.

[0094] Within the operating data set, locate the historical power generation data corresponding to the time points of the historical meteorological data, ensuring that there is corresponding power generation data for each meteorological data time point. For example, if the historical meteorological data has a record value at 8:00 on a certain day, then the corresponding historical power generation data for 8:00 on that day also needs to be extracted.

[0095] Following the same 7:2:1 ratio as historical meteorological data, historical power generation data is divided into training label set, validation label set, and test label set, which correspond to the training dataset, validation dataset, and test dataset, respectively, for label comparison and effect verification during model training.

[0096] Step S1453: Perform data augmentation processing on the training dataset and training label set. By perturbing the meteorological element data in the training dataset, new training data and corresponding new training labels are generated, thereby expanding the scale of the training dataset and training label set.

[0097] The solar irradiance data in the training dataset is perturbed by randomly increasing or decreasing the values ​​according to a certain proportion, such as within ±5%. For example, if the original solar irradiance data is 900 watts per square meter, it may become 855 watts per square meter or 945 watts per square meter after perturbing.

[0098] Based on the disturbed meteorological data and historical correlations, corresponding new training labels are calculated, namely, newly added historical power generation data. In this way, the size of the training dataset and training label set is expanded, allowing the model to access more diverse data and improving its generalization ability.

[0099] Step S1454: Input the expanded training dataset into the power generation linkage prediction model, and perform calculations through the input layer, intermediate feature processing layer, time dimension adjustment layer and output layer of the power generation linkage prediction model to obtain the power generation prediction value during the training process.

[0100] The meteorological element data in the expanded training dataset is input into the model's input layer, which then passes the data to the intermediate feature processing layer. The intermediate feature processing layer performs weighted calculations on the data according to preset correlation weights, fully considering the correlation between meteorological elements and operational parameters during the processing.

[0101] The time dimension adjustment layer adjusts the time information during the computation process to adapt to the data characteristics under different time spans. Finally, the output layer outputs the predicted power generation values ​​during the training process. These predicted values ​​will be compared with the actual power generation data in the training label set to adjust the model parameters.

[0102] Step S1455: Calculate the deviation between the predicted power generation value during the training process and the corresponding historical power generation data in the expanded training label set. The deviation value is the absolute difference between the predicted value and the label value. Calculate the average deviation value during the training process.

[0103] For each predicted power generation value output during training, the deviation value is calculated by comparing it with the corresponding historical power generation data in the training label set. For example, if the predicted value is 500 kW and the label value is 520 kW, the deviation value is 20 kW.

[0104] Adding all the deviation values ​​together and dividing by the number of deviation values ​​yields the average deviation value. By statistically analyzing the average deviation value, we can intuitively understand the overall prediction error of the model during the training process.

[0105] Step S1456: Adjust the network parameters of the power generation linkage prediction model according to the average deviation value and the deviation value of each predicted value. The adjustment objects include the weighting coefficient of the intermediate feature processing layer, the sensitivity parameter of the time dimension adjustment layer, and the operation parameters of the output layer. The adjustment direction is to reduce the average deviation value.

[0106] After obtaining the average deviation value and the deviation value of each predicted value, the parameters are adjusted based on the influence mechanism of the parameters of each layer of the model on the prediction results. For the weighting coefficients of the intermediate feature processing layer, if the prediction deviation of the correlation between a certain type of meteorological element and the operating parameters is large, such as the prediction deviation of power generation corresponding to changes in solar irradiance being consistently high, the weighting coefficient ratio of this type of meteorological element in the intermediate feature processing layer is reduced, while the weighting coefficients of other meteorological elements with more accurate correlation are appropriately increased.

[0107] The sensitivity parameters of the time-dimension adjustment layer need to be adjusted in conjunction with the distribution of the deviation values ​​over time. If the prediction deviation is large over a short time span, such as inaccurate prediction of power fluctuations within 1 hour, the sensitivity parameters of the time-dimension adjustment layer to changes in meteorological elements over a short time span should be increased. If the prediction deviation is more significant over a long time span, such as deviation in trend predictions over 6 hours, the parameters should be adjusted to enhance the model's response to changes in long-term meteorological elements.

[0108] The output layer's operational parameters are adjusted to address the overall prediction result's deviation. When the average deviation is systematically high or low, such as when the overall predicted values ​​are generally higher than historical power generation data, the linear transformation parameters of the output layer need to be adjusted. This is done by changing the weights and biases of the output layer neurons to bring the prediction results closer to the trend of historical data. After each parameter adjustment, the training dataset is re-inputted into the model for computation, and the deviation and average deviation are recalculated until the average deviation shows a continuously decreasing trend.

[0109] Step S1457: After completing one parameter adjustment, input the verification dataset into the adjusted power generation linkage prediction model to obtain the power generation prediction value during the verification process, calculate the deviation value between the prediction value during the verification process and the verification label set, and obtain the average verification deviation value.

[0110] After parameter adjustments are complete, the training dataset is discontinued, and instead, the validation dataset is input into the adjusted power generation linkage prediction model. The validation dataset contains historical meteorological data that were not used in model training. This data comes from the same time period as the training dataset but belongs to different time periods, effectively testing the model's generalization ability.

[0111] The model processes each meteorological element data in the validation dataset. After input layer reception, intermediate feature processing layer operation, time dimension adjustment layer adjustment, and output layer transformation, it generates the corresponding predicted power generation value. The predicted value is then compared one by one with the corresponding historical power generation data in the validation label set. The deviation value of each validation sample is obtained by calculating the absolute difference between the predicted value and the label value, using the same calculation method as in the training process.

[0112] The mean validation bias is obtained by summing all the validation bias values ​​and dividing by the number of validation samples. The mean validation bias reflects the model's predictive performance on unseen data, preventing the model from losing generalization ability due to overfitting the training data.

[0113] Step S1458: Determine whether the verification average deviation value is less than the preset deviation threshold. If it is less, continue to use the test dataset to test the model. If it is not less, return to continue adjusting the model network parameters until the verification average deviation value is less than the preset deviation threshold.

[0114] The preset deviation threshold is determined based on the power fluctuation characteristics of new energy power generation equipment and actual application requirements. For photovoltaic power plants, this preset deviation threshold is usually set to 10%-15% of the average fluctuation range of historical power generation data. The calculated verification average deviation value is compared with the preset deviation threshold. If the verification average deviation value is less than the preset threshold, it means that after parameter adjustment, the model can still maintain good prediction accuracy on untrained data and has a certain generalization ability, and can proceed to the next testing stage.

[0115] If the average deviation value is greater than or equal to the preset deviation threshold, it indicates that the adjusted parameters of the model have not yet reached an ideal state, and there may be overfitting or improper parameter adjustment. In this case, it is necessary to return to the parameter adjustment step, re-analyze the distribution characteristics of the deviation values, and readjust the parameters of the intermediate feature processing layer, time dimension adjustment layer, and output layer for the sample types with larger deviations in the validation set. Repeat the parameter adjustment and validation process until the average deviation value is less than the preset deviation threshold.

[0116] Step S1459: Input the test dataset into the validated power generation linkage prediction model to obtain the predicted power generation value during the test process, calculate the deviation between the predicted value during the test process and the test label set, and obtain the average deviation value of the test.

[0117] Once the validation mean deviation value meets the requirements, the test dataset is input into the validated power generation linkage prediction model. The test dataset is a third set of data independent of the training and validation datasets, with a wider coverage of time span and meteorological conditions, enabling a comprehensive evaluation of the model's final prediction performance.

[0118] The model processes the meteorological element data in the test dataset through a complete prediction process to generate corresponding predicted power generation values. These predicted values ​​are then compared with historical power generation data in the test label set, and the deviation value for each test sample is calculated, i.e., the absolute difference between the predicted value and the label value.

[0119] All test deviations are aggregated, and their average value is calculated to obtain the test mean deviation. The test mean deviation is a key indicator for measuring the final performance of the model, reflecting the model's predictive accuracy in real-world application scenarios.

[0120] Step S1460: Determine whether the average test deviation value is less than the preset deviation threshold. If it is less, the model training is complete. If it is not less, readjust the dataset partitioning ratio and data augmentation method, and repeat the above training steps until the average test deviation value is less than the preset deviation threshold.

[0121] The average test deviation is compared with a preset deviation threshold. If the average test deviation is less than the preset threshold, it indicates that the model has achieved the expected prediction accuracy and generalization ability after the entire process of training, validation, and testing, and the model training is complete. At this point, the model can be solidified for subsequent real-time power generation prediction.

[0122] If the average deviation value of the test is greater than or equal to the preset deviation threshold, the dataset splitting and data augmentation process need to be re-examined. Possible reasons include an unreasonable ratio of training, validation, and test datasets, such as an excessively high proportion of training data leading to overfitting of the model to data from a specific time period; or the data augmentation methods failing to effectively cover real-world weather change scenarios, such as not considering weather data disturbances under extreme weather conditions.

[0123] The dataset splitting ratio was readjusted, such as increasing the sample size of the test dataset, to better reflect actual prediction needs. Simultaneously, data augmentation methods were optimized by increasing the types of perturbations to extreme weather data, such as simulating changes in meteorological elements under scenarios like strong sunlight or sudden cloud cover. After adjustments, the model training, validation, and testing steps were re-executed until the average test deviation was less than the preset deviation threshold.

[0124] Step S150: Perform correlation correction processing on the initial prediction result of the new energy power generation and the historical power generation data in the operating condition data set to generate the final prediction result of the new energy power generation.

[0125] After obtaining the initial prediction results of new energy power generation, the actual power generation process involves many complex factors, such as equipment aging and real-time maintenance status, which may cause the initial prediction results to deviate from the actual situation. By performing correlation analysis between the initial prediction results and historical power generation data in the operating condition data set, the patterns in the historical data can be used to correct the initial prediction results and improve the prediction accuracy.

[0126] The correlation correction process requires establishing a mapping relationship between the initial forecast results and historical data, analyzing the deviation characteristics and trends between the two, and constructing a targeted correction mechanism. The correction process must consider consistency over time to ensure that the corrected forecast results accurately reflect the power generation variation patterns in different time periods, ultimately generating a final forecast result for new energy power generation that can be directly applied to actual operation and scheduling.

[0127] Step S151: Extract the predicted power sequence from the initial prediction result of the new energy power generation, and mark the abnormal prediction values ​​in the predicted power sequence. The predicted power sequence contains power generation prediction values ​​corresponding to multiple time nodes, and the abnormal prediction values ​​are values ​​that exceed the preset power prediction range.

[0128] Extract the predicted power sequence arranged in chronological order from the initial prediction results of new energy power generation. Each element in the predicted power sequence corresponds to the predicted power generation value at a specific time node. The interval between time nodes is consistent with the collection interval of meteorological data, such as a predicted value every 15 minutes.

[0129] The preset power prediction range is determined based on the rated power, historical maximum and minimum power generation, and equipment operating status of the new energy power generation equipment. For photovoltaic power plants, the preset power prediction range is typically set to 0 to 1.1 times the rated power, while also adjusting for historical power fluctuations during the same period. Each predicted value in the predicted power sequence is iterated through, and values ​​exceeding this range are marked as abnormal prediction values.

[0130] For example, if the rated power of a photovoltaic power station is 1000 kW and the preset power prediction range is 0 to 1100 kW, when the predicted value at a certain time point is 1200 kW or -50 kW, the above values ​​are marked as abnormal prediction values. During the marking process, the time point and specific value of the abnormal prediction value need to be recorded.

[0131] Step S152: Extract historical power generation data from the operating condition data set to form a historical power sequence. Remove abnormal historical values ​​from the historical power sequence. The historical power sequence contains multiple historical power generation values ​​at the same time interval as the predicted power sequence. Abnormal historical values ​​are power generation values ​​recorded under equipment failure conditions.

[0132] Historical power generation data corresponding to the time range of the predicted power series are selected from the operating condition data set and arranged according to the same time intervals as the predicted power series to form the historical power series. For example, if the predicted power series covers the next 24 hours with a predicted value every 15 minutes, then the historical power series selects historical power generation data of the same time period and time interval in the past.

[0133] Anomaly detection is performed on historical power sequences to identify power generation records under equipment failure conditions. Equipment failure conditions can be determined through fault codes, maintenance records, or power mutation characteristics in the operating condition data set. For example, if the power generation suddenly drops to 0 in a short period of time without any abnormal changes in meteorological factors, or if the power fluctuation exceeds three times the normal operating range, the corresponding power values ​​are all identified as abnormal historical values.

[0134] Identified abnormal historical values ​​are removed from the historical power series to prevent these invalid data from interfering with subsequent correlation corrections. After removal, if there are missing data in the historical power series, linear interpolation can be performed to supplement the missing data based on the trend of normal historical values ​​before and after the missing data, ensuring the integrity of the historical power series.

[0135] Step S153: Perform time dimension alignment processing on the predicted power sequence and the historical power sequence, and delete the marked abnormal predicted values ​​and corresponding historical power generation values ​​in the predicted power sequence after alignment.

[0136] Because the predicted power series and historical power series may have inconsistent time bases—for example, the predicted time points are based on real-time clocks, while historical data records have time deviations—time dimension alignment processing is required. Using a standard timestamp as a reference, the time node markers in the two series are adjusted to ensure that each predicted power value matches its corresponding historical power generation value precisely in time.

[0137] After time alignment is completed, these outliers are deleted from the predicted power series based on the previously marked locations of the outliers. Simultaneously, the historical power generation values ​​for the corresponding time points in the historical power series are also deleted. For example, if the predicted power series value at 10:00 is an outlier, then both the predicted value and the historical power value for that time point are deleted.

[0138] The above processing eliminates the influence of outlier data on deviation analysis, allowing subsequent deviation parameters to better reflect the prediction error characteristics under normal operating conditions. The processed predicted power sequence maintains the same length as the historical power sequence, and the data at each time point is valid.

[0139] Step S154: Calculate the deviation parameter between each predicted power value and the corresponding historical power generation value in the aligned predicted power sequence, and statistically analyze the distribution of all deviation parameters.

[0140] For the aligned predicted power series and historical power series, the deviation parameter is calculated for each time point. The deviation parameter is calculated as the difference between the predicted power value and the historical power generation value, i.e., deviation parameter = predicted power value - historical power generation value. This difference can be positive or negative; a positive value indicates that the predicted value is higher than the historical value, and a negative value indicates that the predicted value is lower than the historical value.

[0141] The deviation parameters at all time points are summarized, and their distribution is statistically analyzed. This distribution analysis includes the maximum, minimum, median, and average values ​​of the deviation parameters, as well as the percentage of samples within each deviation interval. For example, the number of deviation parameters falling within the intervals of -50 kW to -30 kW, -30 kW to 0 kW, 0 kW to 30 kW, and 30 kW to 50 kW is calculated, along with the proportion of each interval to the total sample size.

[0142] By statistically analyzing the distribution, we can clearly understand the overall distribution characteristics of the deviation parameters and determine whether the prediction results are generally too high, too low, or exhibit a random distribution.

[0143] Step S155: Analyze the variation pattern of the deviation parameter in the time dimension, determine the trend characteristics of the deviation parameter changing with time, and mark the corresponding time interval for each trend characteristic. The trend characteristics include the increasing trend, decreasing trend or fluctuation trend of the deviation parameter.

[0144] Plot the deviation parameter over time using time points as the horizontal axis and the deviation parameter as the vertical axis. Analyze the changing pattern of the deviation parameter over time by observing the shape of the curve. When the curve slopes upwards, indicating that the deviation parameter gradually increases over time, it is considered an increasing trend; when the curve slopes downwards, indicating that the deviation parameter gradually decreases over time, it is considered a decreasing trend; when the curve fluctuates within a certain range without a clear overall upward or downward trend, it is considered a fluctuating trend.

[0145] During the analysis, a corresponding time interval was labeled for each trend characteristic. For example, from 8:00 to 10:00 AM, the deviation parameter gradually increased from 10 kW to 40 kW, showing an increasing trend, and this time period was labeled as the increasing trend interval; from 10:00 to 2:00 PM, the deviation parameter fluctuated between 20 kW and 40 kW without a clear overall trend, and this period was labeled as the fluctuating trend interval; from 2:00 to 5:00 PM, the deviation parameter gradually decreased from 35 kW to 5 kW, showing a decreasing trend, and this period was labeled as the decreasing trend interval.

[0146] The division of trend characteristics needs to be combined with the operating characteristics of new energy power generation equipment. For example, the power of a photovoltaic power station increases from sunrise to noon and decreases from noon to sunset throughout the day. The trend characteristics of the deviation parameters may be related to this pattern and need to be fully considered when marking the time interval.

[0147] Step S156: Construct a deviation correction function based on the trend characteristics and the corresponding time interval. The input of the deviation correction function is the predicted power value, the corresponding time node information and the trend characteristic type, and the output is the corrected power value.

[0148] Based on different trend characteristics and corresponding time intervals, deviation correction functions are constructed respectively. For an increasing trend interval, since the deviation parameter gradually increases over time, the correction function needs to include a correction term that increases over time, allowing the predicted value to be adjusted downwards or upwards more significantly over time to offset the increasing trend of deviation. For example, for cases where predicted values ​​are generally high and the deviation is increasing, the correction function can be set as: Corrected power value = Predicted power value - (Base correction amount + Time coefficient × Time interval), where the time coefficient is a positive number, and the correction amount gradually increases with the increase of the time interval.

[0149] For a decreasing trend range, the deviation parameter gradually decreases over time, and the correction term of the correction function must also decrease over time. If the predicted value is too high and the deviation is decreasing, the correction function can be set as: Corrected power value = Predicted power value - (Base correction amount - Time coefficient × Time interval), and the correction amount gradually decreases over time.

[0150] For fluctuation trend ranges where the deviation parameter does not show a clear overall trend, the correction function can adopt a fixed correction method based on the historical average deviation, such as corrected power value = predicted power value - average deviation parameter within the range. Simultaneously, the function input must include the predicted power value, the corresponding time node information (to determine the time range), and the trend characteristic type (to select the corresponding correction formula), ensuring that the function can output accurate corrected power values ​​based on different input conditions.

[0151] Step S157: Input each predicted power value, the corresponding time node information, and the matching trend feature type from the initial prediction result of the new energy power generation into the deviation correction function to obtain the corrected power value.

[0152] Each predicted power value in the initial prediction results of new energy power generation is iterated through, and its corresponding time node information is extracted. Based on the time node information, the time interval to which the predicted value belongs is determined, and then the corresponding trend feature type is matched. For example, if the time node corresponding to a certain predicted power value is 9:30, which belongs to the previously marked increasing trend interval from 8:00 to 10:00, then the matched trend feature type is increasing trend.

[0153] The predicted power value, time node information (such as a specific timestamp), and trend characteristic type are used as input parameters and substituted into the corresponding deviation correction function for calculation. During the calculation process, the function determines the time interval based on the time node information, selects the correction formula based on the trend characteristic type, and calculates the corrected power value according to the formula.

[0154] For example, for a predicted power value of 500 kW within an increasing trend interval, with a time node of 9:30 and a time interval of 1.5 hours from the start time of the interval at 8:00, a base correction of 10 kW, and a time coefficient of 5 kW / hour, the corrected power value = 500 - (10 + 5 × 1.5) = 500 - 17.5 = 482.5 kW. Perform this operation for each predicted power value to obtain a series of corrected power values.

[0155] Step S158: Arrange all the corrected power values ​​in chronological order to form the final prediction result of the new energy power generation.

[0156] Arrange all corrected power values ​​in chronological order according to their corresponding time points, ensuring that the corrected power value for each time point is in the correct position in the sequence. After arrangement, perform an integrity check on the corrected power sequence to confirm that all time points have corresponding corrected power values, with no missing or duplicates.

[0157] During the inspection, if the corrected power value at a certain time point is still found to be abnormal, such as exceeding the actual possible power generation range of the equipment, the predicted power value, time interval matching, and correction function calculation process for that time point need to be rechecked to identify the problem and recalculate. After confirming that all corrected power values ​​are reasonable, a complete final prediction result of new energy power generation is formed.

[0158] This final prediction result can more accurately reflect the changes in the power generation capacity of new energy power generation equipment in the future.

[0159] Step S1561: Classify and encode the trend features, assign unique trend codes to increasing trends, decreasing trends and fluctuating trends respectively, and establish a correspondence table between trend features and trend codes.

[0160] The three identified trend characteristics are classified and coded, with each trend characteristic assigned a unique identifier. For example, an increasing trend is assigned the code "T01", a decreasing trend is assigned the code "T02", and a fluctuating trend is assigned the code "T03". The codes use an alphanumeric format for easy recognition and processing by the computer system.

[0161] Establish a mapping table between trend features and trend codes. The table contains two columns: "Trend Feature Name" and "Trend Code," specifying the code value for each trend feature. For example, the table records "Increasing Trend - T01," "Decreasing Trend - T02," and "Fluctuating Trend - T03." This mapping table is stored in the system database. During the subsequent construction and invocation of deviation correction functions, the corresponding code can be queried through the trend feature, or the corresponding trend feature can be retrieved through the code.

[0162] Step S1562: Divide the time intervals corresponding to the trend characteristics into multiple time sub-intervals, ensuring that the variation pattern of the deviation parameters within each time sub-interval remains consistent, and label each time sub-interval with a corresponding time code.

[0163] Within each time interval corresponding to a trend feature, the variation pattern of the deviation parameter is analyzed using the sliding window method. The duration of the sliding window is determined based on the historical data collection interval, for example, using 1 hour as the base window length, and the fluctuation of the deviation parameter is analyzed by moving the window one window at a time. When the variation amplitude of the deviation parameter within the window is within a preset consistency threshold range, the time period corresponding to that window is marked as a potential time sub-interval.

[0164] Adjacent potential time intervals are merged and verified. If the difference in the slope of the deviation parameter change trend between two adjacent windows is less than a set threshold, they are merged into one time interval. For example, in a time interval with an increasing trend, the deviation parameter increases by 2% per hour for the first 3 hours and by 2.1% per hour for the next 2 hours. If the slope difference threshold is set to 0.2%, these two time periods can be merged into one time interval.

[0165] After merging all potential sub-intervals, the final time sub-interval division is determined. Each time sub-interval includes a start timestamp and an end timestamp. For example, if an increasing trend corresponds to a time interval of 8:00-18:00, it is divided into three sub-intervals: 8:00-12:00, 12:00-15:00, and 15:00-18:00. A unique time code is assigned to each time sub-interval. The coding rule uses a combination of numbers from the start time of the interval. For example, 8001200 represents the time sub-interval of 8:00-12:00, 12001500 represents the time sub-interval of 12:00-15:00, and so on.

[0166] The correspondence between time codes and time sub-intervals is stored in a code mapping table. This table also records the trend characteristic type, start time, end time, and deviation parameter change pattern for each time sub-interval. For example, the record corresponding to time code 8001200 is: trend characteristic type is increasing trend, start time 8:00, end time 12:00, and the change pattern is that the deviation parameter increases by approximately 2% per hour. Using this method, when subsequently calling the deviation correction function, the corresponding time code can be quickly matched based on the time node information.

[0167] Step S1563: Collect the deviation parameters, corresponding predicted power values ​​and historical power generation data for each time sub-interval, and establish a corresponding database.

[0168] The start and end times of each time sub-interval are retrieved from the time-encoding mapping table. Based on the time range, all deviation parameters within that interval are extracted from the deviation parameter records. Simultaneously, the corresponding predicted power values ​​and historical power generation data for these deviation parameters are extracted, ensuring a one-to-one correspondence between the three at each time point. For example, within the time sub-interval of 8:00-12:00, the deviation parameters, predicted power values, and historical power generation data for five time points—8:00, 9:00, 10:00, 11:00, and 12:00—are extracted.

[0169] The extracted data undergoes integrity verification to check for missing data or mismatched time points. If a predicted power value or historical power generation data is missing for a certain time point, it is supplemented by linear interpolation from data of adjacent time points. For example, if data for the 9:30 time point is missing, supplementary data for 9:30 can be calculated based on data from 9:00 and 10:00.

[0170] The validated dataset is stored by time sub-interval, with each time sub-interval corresponding to a data subset. Each data subset contains a time node field, a deviation parameter field, a predicted power value field, and a historical power generation data field. A time-coded label and a trend feature label are added to each data subset; for example, the data subset with time code 800-1200 is labeled with an "increasing trend" label.

[0171] When establishing the corresponding database, a hierarchical storage structure is adopted. The first layer divides the data into folders according to trend feature types, and the second layer divides the data into subfolders according to time codes. Each subfolder contains the dataset files for the corresponding time sub-interval. The dataset files are stored in a structured format, supporting data retrieval by time node, deviation parameter range, and other conditions, facilitating quick data retrieval when constructing the deviation correction sample set later.

[0172] Step S1564: Extract multiple sets of deviation parameters, predicted power values ​​and historical power generation data under the same time sub-interval and the same trend characteristics from the corresponding database, and construct a deviation correction sample set. Each sample in the sample set contains a predicted power value, time node information, trend code and corresponding correction target value. The correction target value is historical power generation data.

[0173] Based on the trend characteristic type and time code, a subset of data that meets the criteria is selected from the corresponding database. For example, a subset of data with an "increasing trend" and a time code of 8001200 is selected. This subset contains relevant data for all time points within this time interval.

[0174] From the selected data subset, the predicted power value, time node information, trend code, and historical power generation data for each individual time node are extracted and combined to form a sample. The trend code is determined based on the trend characteristic type; for example, "increasing trend" corresponds to code "T01," "decreasing trend" corresponds to code "T02," and "fluctuating trend" corresponds to code "T03." The target value for correction is directly derived from the historical power generation data for that time node.

[0175] Iterate through all time points in the data subset as described above to generate multiple samples. Deduplicate the samples; if two samples have identical predicted power values, time point information, and trend codes, retain one of them. Divide the deduplicated samples into training and validation samples according to a preset ratio, for example, 80% as training samples and 20% as validation samples, used to construct the initial bias correction function and validate the function performance, respectively.

[0176] The partitioned sample set is stored as a sample set file. Each sample in the file is arranged in a fixed format and contains, in sequence, the predicted power value, time node information, trend code, and correction target value. At the same time, a sample set description file is generated, recording information such as the total number of samples, the number of training samples, the number of validation samples, the distribution of trend features, and the coverage of time sub-intervals.

[0177] Step S1565: Model the deviation correction sample set using regression analysis, using predicted power value, time node information and trend code as input variables, and correction target value as output variable, to construct an initial deviation correction function.

[0178] Multiple linear regression was chosen as the modeling basis, with predicted power values, time node information, and trend codes as independent variables, and the corrected target value as the dependent variable. The input variables were preprocessed, converting the time node information into numerical variables, such as converting the time of day into minutes (8:00 to 480 minutes, 12:00 to 720 minutes, etc.). One-hot encoding was performed on the trend codes, converting "T01," "T02," and "T03" into three binary variables, each representing whether the trend type exists.

[0179] Substitute the preprocessed input and output variables into the multiple linear regression model and calculate the model parameters using the least squares method. The model parameters include the coefficients and constants for each input variable. For example, the model expression can be represented as: Corrected target value = Coefficient 1 × Predicted power value + Coefficient 2 × Time node value + Coefficient 3 × Trend encoding variable 1 + Coefficient 4 × Trend encoding variable 2 + Coefficient 5 × Trend encoding variable 3 + Constant term.

[0180] During the calculation process, the model parameters are iteratively optimized. By adjusting the coefficient values, the error between the model's predicted value and the corrected target value is gradually reduced. The error value is recorded for each iteration. The iteration stops when the error value no longer decreases significantly or reaches the preset number of iterations, yielding the parameter values ​​of the initial deviation correction function. These parameter values ​​are then substituted into the model expression to form the initial deviation correction function.

[0181] The initial bias correction function is initially validated by substituting the input variables from the validation sample set into the function to calculate the predicted correction value. This predicted correction value is then compared with the target correction value in the validation sample set to calculate the error rate. If the error rate is within the preset range, the initial bias correction function is successfully constructed. If the error rate exceeds the preset range, the model parameters are readjusted or another regression analysis method, such as multinomial regression, is selected, and modeling is performed again.

[0182] Step S1566: Optimize the parameters of the initial deviation correction function by minimizing the sum of squared errors between the function's predicted value and the correction target value, and adjust the coefficient parameters of the initial deviation correction function to obtain the optimized deviation correction function.

[0183] The optimization objective is defined as minimizing the sum of squared errors between the predicted and corrected target values. This sum of squared errors is calculated by squaring the difference between the predicted and corrected values ​​for each sample and then summing the results. The optimization problem is constructed using the coefficients of the initial bias correction function as the optimization variables.

[0184] The gradient descent method is used to solve the optimization problem. First, the partial derivatives of the sum of squared errors with respect to each coefficient parameter are calculated to obtain the gradient vector. The coefficient parameter values ​​are then adjusted according to the direction of the gradient vector. The step size for each adjustment is determined by the learning rate, which is set through a trial-and-error method to ensure stable convergence of the parameter adjustment process.

[0185] During the iterative optimization process, the sum of squared errors is calculated after each iteration, and its trend is observed. If the sum of squared errors continues to decrease, the iteration continues; if the sum of squared errors shows an increasing trend, the learning rate is reduced and the iteration is restarted. Simultaneously, an iteration stopping condition is set: when the sum of squared errors is less than a preset threshold or the maximum number of iterations is reached, the iteration stops, and the optimized coefficient parameter values ​​are obtained.

[0186] Substituting the optimized coefficient parameter values ​​into the initial bias correction function yields the optimized bias correction function. The optimized function is then validated by calculating its sum of squared errors on the validation sample set and comparing it with the sum of squared errors before optimization to confirm the optimization effect. If the optimized sum of squared errors is significantly reduced, the optimized function is accepted; otherwise, the optimization algorithm or learning rate is readjusted, and parameter optimization is performed again.

[0187] Step S1567: Integrate the optimized deviation correction functions corresponding to different time sub-intervals and different trend characteristics, and establish a function index table. The index items contain time codes and trend codes. Locate the corresponding deviation correction function through the index items.

[0188] A unique identifier is assigned to the optimized deviation correction function corresponding to each time sub-interval and trend feature combination. The identifier is composed of a time code and a trend code; for example, the time code 8001200 and the trend code T01 combine to form the identifier "8001200-T01". The function identifier and the corresponding deviation correction function are stored in a function library. The function library adopts a key-value pair structure, where the key is the function identifier and the value is the parameter and expression of the deviation correction function.

[0189] A function index table is constructed, with each row containing three fields: time code, trend code, and function identifier. The corresponding function identifier can be found in the index table using the time code and trend code, and then the corresponding deviation correction function can be retrieved from the function library based on the function identifier. For example, inputting the time code 8001200 and the trend code T01, the function identifier "8001200-T01" is found through the index table, and the corresponding deviation correction function is then retrieved from the function library.

[0190] The function index table is optimized by creating a combined index of time codes and trend codes to improve query efficiency. When a deviation correction function needs to be called, the corresponding time code is determined based on the time node information, and the corresponding trend code is determined based on the trend characteristics. The function identifier is quickly located through the combined index, enabling rapid retrieval of the deviation correction function. Simultaneously, the function index table and function library are updated regularly. When new time sub-intervals or trend characteristics are added, the corresponding functions and index entries are added promptly.

[0191] Step S1568: Perform performance verification on the integrated deviation correction function. Select a verification sample set to input the deviation correction function, calculate the deviation between the corrected power value output by the deviation correction function and the historical power generation data in the verification sample set. If the deviation is less than the preset verification deviation, the deviation correction function is completed. If the deviation is not less than the preset verification deviation, the parameters of the deviation correction function are re-optimized until the deviation is less than the preset verification deviation.

[0192] A validation sample set is selected from the sample set, containing samples with different time sub-intervals and different trend characteristics. The predicted power values, time node information, and trend codes from the validation sample set are input into the integrated deviation correction function. The function is then used to calculate the corrected power values ​​based on the time and trend codes.

[0193] Calculate the deviation between the corrected power value of each sample and the historical power generation data in the validation sample set. The deviation is calculated as the absolute difference between the two. Statistically analyze the deviations of all samples and calculate the average and maximum deviations. Compare the average deviation with the preset validation deviation. If the average deviation is less than the preset validation deviation and the maximum deviation is within an acceptable range, the deviation correction function performance validation is considered successful, and the system is considered complete.

[0194] If the average deviation or maximum deviation does not meet the requirements, samples with larger deviations are selected, and their corresponding time sub-intervals and trend characteristics are analyzed. For the time sub-intervals and trend characteristics of these samples, the corresponding data subsets are retrieved again, the number of samples is increased or the data preprocessing method is adjusted, the deviation correction function is reconstructed, and the parameters are optimized.

[0195] Repeat the performance verification process until the deviation between the corrected power value output by the deviation correction function and the historical power generation data is less than the preset verification deviation. After successful verification, store the final deviation correction function and its index table in the model library.

[0196] Figure 2 The illustration shows exemplary hardware and software components of a meteorological data-based new energy power generation forecasting system 100, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the meteorological data-based new energy power generation forecasting system 100 and to perform the functions in this application.

[0197] For example, a new energy power generation forecasting system 100 based on meteorological data may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the new energy power generation forecasting system 100 based on meteorological data may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The new energy power generation forecasting system 100 based on meteorological data also includes an I / O interface 150 between the computer and other input / output devices.

[0198] Furthermore, this embodiment of the invention also provides a readable storage medium, which has computer-executable instructions pre-set in it. When the processor executes the computer-executable instructions, the above-mentioned new energy power generation prediction method based on meteorological data is implemented.

Claims

1. A method for predicting the power generation of new energy sources based on meteorological data, characterized in that, The method includes: Acquire meteorological data sets corresponding to the new energy power generation equipment and operating condition data sets of the new energy power generation equipment. The meteorological data sets contain continuously collected data of multiple meteorological elements, and the operating condition data sets contain operating parameter data of the new energy power generation equipment under different operating conditions. The meteorological data set contains a collection of meteorological elements. The continuous collection data of various meteorological elements are analyzed and processed to establish a time series correlation diagram of meteorological elements. The time series correlation diagram of meteorological elements is used to characterize the mutual influence relationship of different meteorological elements in the time dimension. The meteorological element time series correlation diagram is correlated and mapped with the operating condition data set to generate the operating condition response correlation feature of the new energy power generation equipment. The operating condition response correlation feature is used to characterize the correspondence between changes in meteorological elements and changes in equipment operating parameters. Based on the operating condition response correlation features, a power generation linkage prediction model is constructed. The power generation linkage prediction model is used to predict and analyze the real-time meteorological element data in the meteorological data set to obtain the initial prediction result of new energy power generation. The initial prediction result of the new energy power generation is correlated and corrected with the historical power generation data in the operating condition data set to generate the final prediction result of the new energy power generation.

2. The new energy power generation prediction method based on meteorological data according to claim 1, characterized in that, The step of performing correlation analysis on continuously collected data of multiple meteorological elements in the meteorological data set to establish a time-series correlation diagram of meteorological elements includes: Extract the time series information of continuously collected data of multiple meteorological elements in the meteorological data set, determine the time collection interval and time series length of the continuously collected data of each meteorological element, and mark the abnormal time nodes in the time series information. The abnormal time nodes are the time nodes corresponding to data collection interruption or data missing. The marked time series information is supplemented by adding meteorological element data corresponding to the abnormal time nodes based on the changing trends of meteorological element data before and after the abnormal time nodes, thus forming complete time series information. Based on the completed time series information, the correlation parameter between continuously collected data of different meteorological elements at the same time node is calculated. The correlation parameter is used to quantify the degree of mutual influence between different meteorological elements. Based on the correlation parameter, an initial meteorological element correlation diagram is constructed. The nodes in the initial meteorological element correlation diagram are different types of meteorological elements, and the lines between the nodes are the corresponding correlation parameters. Invalid lines in the initial meteorological element correlation diagram are deleted. Invalid lines are those whose correlation parameters are lower than a preset correlation threshold. The initial meteorological element association map after deleting invalid connections is subjected to time-series dimension expansion processing. The initial meteorological element association maps at different time points are connected in chronological order to form a meteorological element time-series association framework containing time-dimensional information. In the meteorological element time-series correlation framework, based on the changing trend of the correlation degree parameter of meteorological elements between different time nodes, time dimension correlation edges are added, and the corresponding time span information is labeled for each time dimension correlation edge. The time dimension correlation edges are used to characterize the influence relationship of the same meteorological element between different time nodes. By integrating the initial meteorological element association graph, the meteorological element temporal association framework, and the time-dimensional association edges, the meteorological element temporal association graph is generated. The meteorological element temporal association graph includes node type information, node correlation degree parameter information, temporal association edge information, and time span information.

3. The method for predicting new energy power generation based on meteorological data according to claim 2, characterized in that, The calculation of the correlation parameter between continuously collected data of different meteorological elements at the same time node based on the completed time series information includes: Extract all the same time nodes from the completed time series information, determine the values ​​of continuously collected data of different meteorological elements corresponding to each same time node, and establish a time node-meteorological element value correspondence table. Two different types of continuously collected meteorological data were selected as a group of analysis objects. The continuous collected data of wind meteorological elements and light meteorological elements were selected as the first group of analysis objects. The continuous collected data of wind meteorological elements and air pressure meteorological elements were selected as the second group of analysis objects. The continuous collected data of light meteorological elements and air pressure meteorological elements were selected as the third group of analysis objects. For each group of analysis objects, extract the values ​​of the group of analysis objects at each same time node from the time node-meteorological element value correspondence table to form a value sequence pair of the group of analysis objects; Calculate the covariance of each set of numerical sequences of the analysis objects at each same time node. The covariance is used to reflect the consistency of the changing trend of the two types of meteorological element data. The validity of the calculated covariance is judged, and the covariance values ​​with absolute values ​​exceeding the preset covariance range are removed. Based on the covariance after the validity judgment, the correlation coefficient is further calculated. The correlation coefficient is the ratio of the covariance to the product of the standard deviations of the two types of meteorological element data. The correlation coefficient of each group of analysis objects at each same time node is recorded. The correlation coefficient is normalized by using a linear normalization method to convert the range of the correlation coefficient to between 0 and 1. The normalized correlation coefficient is the correlation degree parameter. Following the steps above, calculate the correlation parameters of all groups of analyzed objects at each same time point, forming a sequence of correlation parameters for each group of analyzed objects.

4. The method for predicting new energy power generation based on meteorological data according to claim 1, characterized in that, The step of performing association mapping processing between the time-series correlation diagram of the meteorological elements and the operating condition data set to generate the operating condition response correlation features of the new energy power generation equipment includes: Extract the operating parameter change information from the operating condition data set, filter the abnormal change data in the operating parameter change information, the abnormal change data is the data whose numerical change range exceeds the preset range, the operating parameter change information includes the numerical change range and the numerical change rate of the operating parameter, the numerical change range is the difference of the operating parameter values ​​at different time points, and the numerical change rate is the ratio of the numerical change range to the time interval. Calculate the rate of change of meteorological element values ​​and the rate of change of operational parameter values, whereby the rate of change is defined as the relative change of the values ​​with respect to the reference values. Establish a mapping table to match the rate of change of meteorological element values ​​with the rate of change of operational parameter values, and label the corresponding time span information for each matching item; The mapping table is optimized by calculating the matching degree of each matching item and deleting matching items with a matching degree lower than a preset matching threshold to obtain the optimized mapping table. The matching degree is the degree of correspondence between the rate of change of meteorological element values ​​and the rate of change of operating parameter values. Based on the optimized mapping table, the probability distribution of the change rate of the operating parameter corresponding to the change rate of the meteorological element value is calculated. The probability distribution is used to represent the possibility that the operating parameter will have different change rates under a specific meteorological element value change rate. Based on the probability distribution and the correlation parameters in the time series correlation diagram of the meteorological elements, the change information of the operating parameters is weighted to obtain weighted change information of the operating parameters, and the weight values ​​are the corresponding correlation parameters. The weighted operating parameter change information is combined with the time series association direction and time span information in the meteorological element time series association diagram to generate the operating condition response association feature containing the time dimension change trend. The operating condition response association feature includes the meteorological element change sequence, the corresponding operating parameter change sequence, the association weight sequence of the two, and the time series span sequence.

5. The new energy power generation prediction method based on meteorological data according to claim 4, characterized in that, The calculation of the probability distribution of the range of changes in operating parameters corresponding to the range of changes in the numerical values ​​of each meteorological element, based on the optimized mapping table, includes: Extract a meteorological element value change rate from the optimized mapping table as the current analysis change rate, determine all operating parameter value change rates corresponding to the current analysis change rate, and form a set of operating parameter value change rates of the current analysis change rate. The set of operating parameter value change rates is grouped and divided into multiple change rate intervals according to the equal intervals of the change rate magnitude. Count the number of operating parameter value change rates contained in each change rate interval, and calculate the proportion of the number in each change rate interval to the total number of operating parameter value change rates. This proportion is the probability value of the operating parameter value change rate being in that change rate interval. Record each rate of change interval and its corresponding probability value to form a probability distribution table corresponding to the rate of change of the meteorological element. All probability distribution tables are integrated, and each probability distribution table is labeled with the corresponding meteorological element type and time span information to form a probability distribution set. The probability distribution set contains the probability distribution information, meteorological element type information and time span information corresponding to the rate of change of each meteorological element value.

6. The method for predicting new energy power generation based on meteorological data according to claim 1, characterized in that, The power generation linkage prediction model is constructed based on the operating condition response correlation features. This model is then used to predict and analyze real-time meteorological element data in the meteorological dataset to obtain initial prediction results for new energy power generation, including: Extract the meteorological element change sequence, operating parameter change sequence, correlation weight sequence, and time series span sequence from the working condition response correlation features. Use the meteorological element change sequence as the model input feature, the operating parameter change sequence as the model intermediate feature, the correlation weight sequence as the model weight parameter, and the time series span sequence as the model time dimension adjustment parameter. A network structure for constructing a power generation linkage prediction model is provided. The network structure includes an input layer, an intermediate feature processing layer, a time dimension adjustment layer, and an output layer. The input layer is used to receive the input features of the model. The intermediate feature processing layer is used to perform correlation operations on the intermediate features of the model. The time dimension adjustment layer is used to adjust the sensitivity of the model to time dimension information according to the time dimension adjustment parameters. The output layer is used to output the predicted power generation value. The associated weight sequence is imported into the intermediate feature processing layer as a weighting coefficient in the intermediate feature operation process, so that when the intermediate feature processing layer processes the sequence of changes in running parameters, it performs feature enhancement or weakening according to the weight value of the associated weight sequence. The time-series span sequence is imported into the time dimension adjustment layer, and the response weights of the model to changes in meteorological elements under different time spans are adjusted through the time dimension adjustment layer. The power generation prediction model is trained using historical meteorological data and corresponding historical power generation data from the meteorological data set. The training objective is to ensure that the deviation between the predicted power generation value output by the model and the historical power generation data is within a preset range. After training, the real-time meteorological element data in the meteorological data set is input into the input layer of the power generation linkage prediction model. The intermediate feature processing layer processes the changes in the operating parameters corresponding to the real-time meteorological element data. The time dimension adjustment layer adjusts the time dimension information during the calculation process. The output layer outputs the corresponding power generation prediction value. The power generation prediction values ​​corresponding to multiple real-time meteorological element data are arranged in chronological order to form the initial prediction result of the new energy power generation.

7. The method for predicting new energy power generation based on meteorological data according to claim 6, characterized in that, The training process for the power generation prediction model involves using historical meteorological data and corresponding historical power generation data from the meteorological data set. The training objective is to ensure that the deviation between the model's output power generation prediction and the historical power generation data is within a preset range, including: Historical meteorological element data is extracted from the meteorological data set. The historical meteorological element data is continuously collected meteorological element data within a preset historical time period. The historical meteorological element data is divided into training dataset, validation dataset and test dataset according to the time order. The division ratio is determined according to the preset dataset allocation ratio. Historical power generation data corresponding to the historical meteorological element data are extracted from the operating condition data set, so that the historical power generation data and the historical meteorological element data correspond one-to-one at the time node. Similarly, the dataset is divided into training label set, verification label set and test label set according to the allocation ratio of the dataset. Data augmentation processing is performed on the training dataset and training label set. By perturbing the meteorological element data in the training dataset, new training data and corresponding new training labels are generated, thereby expanding the scale of the training dataset and training label set. The expanded training dataset is input into the power generation prediction model. The power generation prediction model is then processed through its input layer, intermediate feature processing layer, time dimension adjustment layer, and output layer to obtain the predicted power generation value during the training process. The deviation between the predicted power generation value during the training process and the historical power generation data corresponding to the expanded training label set is calculated. The deviation value is the absolute difference between the predicted value and the label value. The average deviation value during the training process is statistically analyzed. The network parameters of the power generation linkage prediction model are adjusted based on the average deviation value and the deviation value of each predicted value. The adjustment objects include the weighting coefficients of the intermediate feature processing layer, the sensitivity parameters of the time dimension adjustment layer, and the operation parameters of the output layer. The adjustment direction is to reduce the average deviation value. After completing one parameter adjustment, the verification dataset is input into the adjusted power generation linkage prediction model to obtain the power generation prediction value during the verification process. The deviation value between the prediction value during the verification process and the verification label set is calculated to obtain the average verification deviation value. Determine whether the average verification deviation value is less than a preset deviation threshold. If it is less, continue to use the test dataset to test the model. If it is not less, return to continue adjusting the model network parameters until the average verification deviation value is less than the preset deviation threshold. Input the test dataset into the validated power generation prediction model to obtain the predicted power generation value during the test process. Calculate the deviation between the predicted value during the test process and the test label set to obtain the average deviation value. Determine whether the average test deviation value is less than a preset deviation threshold. If it is less, the model training is complete. If it is not less, readjust the dataset partitioning ratio and data augmentation method, and repeat the above training steps until the average test deviation value is less than the preset deviation threshold.

8. The method for predicting new energy power generation based on meteorological data according to claim 1, characterized in that, The step of correlating and correcting the initial prediction result of the new energy power generation with the historical power generation data in the operating condition data set to generate the final prediction result of the new energy power generation includes: Extract the predicted power sequence from the initial prediction results of the new energy power generation, and mark the abnormal prediction values ​​in the predicted power sequence. The predicted power sequence contains power generation prediction values ​​corresponding to multiple time nodes, and the abnormal prediction values ​​are values ​​that exceed the preset power prediction range. Historical power generation data is extracted from the operating condition data set to form a historical power sequence. Abnormal historical values ​​in the historical power sequence are removed. The historical power sequence contains multiple historical power generation values ​​with the same time interval as the predicted power sequence. Abnormal historical values ​​are power generation values ​​recorded under equipment failure conditions. The predicted power sequence and the historical power sequence are aligned in the time dimension. After alignment, the abnormal predicted values ​​and the corresponding historical power generation values ​​marked in the predicted power sequence are deleted. Calculate the deviation parameter between each predicted power value and the corresponding historical power generation value in the aligned predicted power sequence, and statistically analyze the distribution of all deviation parameters; Analyze the variation pattern of the deviation parameter over time, determine the trend characteristics of the deviation parameter over time, and label the corresponding time interval for each trend characteristic. The trend characteristics include the increasing trend, decreasing trend, or fluctuation trend of the deviation parameter. A deviation correction function is constructed based on the trend characteristics and the corresponding time intervals. The inputs of the deviation correction function are the predicted power value, the corresponding time node information, and the trend characteristic type, and the output is the corrected power value. Each predicted power value, the corresponding time node information, and the matching trend feature type in the initial prediction result of the new energy power generation are input into the deviation correction function to obtain the corrected power value. All the corrected power values ​​are arranged in chronological order to form the final prediction result of the new energy power generation.

9. The method for predicting new energy power generation based on meteorological data according to claim 8, characterized in that, The deviation correction function is constructed based on the trend characteristics and the corresponding time interval. The inputs to the deviation correction function are the predicted power value, the corresponding time node information, and the trend characteristic type. The output is the corrected power value, including: The trend features are classified and coded, and unique trend codes are assigned to increasing trends, decreasing trends and fluctuating trends respectively. A correspondence table between trend features and trend codes is established. Based on the time interval corresponding to the trend characteristics, multiple time sub-intervals are divided. The variation pattern of the deviation parameter in each time sub-interval remains consistent, and each time sub-interval is labeled with a corresponding time code. Collect the deviation parameters, corresponding predicted power values, and historical power generation data for each time sub-interval, and establish a corresponding database; Extract multiple sets of deviation parameters, predicted power values ​​and historical power generation data under the same time sub-interval and the same trend characteristics from the corresponding database to construct a deviation correction sample set. Each sample in the sample set contains a predicted power value, time node information, trend code and corresponding correction target value. The correction target value is historical power generation data. The deviation correction sample set is modeled using regression analysis, with predicted power value, time node information and trend code as input variables and correction target value as output variable, to construct an initial deviation correction function; The initial deviation correction function is optimized by minimizing the sum of squared errors between the predicted value and the correction target value, and adjusting the coefficient parameters of the initial deviation correction function to obtain the optimized deviation correction function. The optimized deviation correction functions corresponding to different time sub-intervals and different trend characteristics are integrated to establish a function index table. The index items contain time codes and trend codes, and the corresponding deviation correction functions are located by indexing the items. The performance of the integrated deviation correction function is verified. A verification sample set is selected as the input to the deviation correction function, and the deviation between the corrected power value output by the deviation correction function and the historical power generation data in the verification sample set is calculated. If the deviation is less than the preset verification deviation, the deviation correction function is completed. If the deviation is not less than the preset verification deviation, the parameters of the deviation correction function are re-optimized until the deviation is less than the preset verification deviation.

10. A new energy power generation prediction system based on meteorological data, characterized in that, The method includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the new energy power generation prediction method based on meteorological data as described in any one of claims 1-9.

Images