New energy output prediction device and method under multi-source data fusion analysis

Abstract

The invention provides a new energy output prediction device and method under multi-source data fusion analysis, and relates to the technical field of new energy output prediction, and the device comprises a data acquisition module which is used for collecting a multi-source data set of a target region, and carrying out the data fusion; the data labeling module is used for labeling the meteorological data in the multi-source data set through an abnormal weather classifier; the data updating module is used for updating the multi-modal fusion data set based on the weather labeling information; and the output prediction module is used for matching a plurality of corresponding output prediction channels in the new energy output prediction model according to the weather labeling information, and predicting the updated fusion data set to obtain a new energy output prediction result. Through the method, the technical problem that the accuracy of new energy output prediction is reduced due to the complexity of meteorological data in the prior art can be solved, and the accuracy of new energy output prediction is improved through multi-source data fusion and abnormal weather classification.

Description

Technical Field

[0001] This application relates to the technical field of new energy output prediction, and particularly to a new energy output prediction device and method under multi-source data fusion analysis. Background Art

[0002] New energy output prediction refers to predicting the power generation capacity of new energy (such as wind energy, solar energy, etc.) within a certain period according to information such as meteorological data and historical output data. Existing new energy output prediction methods have improved the accuracy of new energy output prediction to a certain extent. However, due to the complexity and diverse changes of meteorological data, the existing technology still faces many challenges. Meteorological data is highly non-linear and uncertain, and involves multiple variables (such as wind speed, temperature, humidity, etc.). There are not only complex spatio-temporal relationships between these variables, but also the changes in these data are often unable to be captured by simple linear models or statistical models. When meteorological conditions change drastically or extreme weather occurs, the prediction accuracy of existing methods will decrease significantly, resulting in a reduction in the accuracy of new energy output prediction.

[0003] In summary, there is a technical problem in the existing technology that the accuracy of new energy output prediction is reduced due to the complexity of meteorological data. Summary of the Invention

[0004] The purpose of this application is to provide a new energy output prediction device and method under multi-source data fusion analysis to solve the technical problem in the existing technology that the accuracy of new energy output prediction is reduced due to the complexity of meteorological data.

[0005] In view of the above problems, this application provides a new energy output prediction device and method under multi-source data fusion analysis.

[0006] In the first aspect, this application provides a new energy output prediction device under multi-source data fusion analysis. The new energy output prediction device under multi-source data fusion analysis includes: a data acquisition module, configured to collect a multi-source data set of a target area through multi-source sensors, perform data fusion on the multi-source data set to obtain a multi-modal fusion data set; a data annotation module, configured to annotate the meteorological data in the multi-source data set through an abnormal weather classifier to obtain weather annotation information; a data update module, configured to update the multi-modal fusion data set based on the weather annotation information to obtain an updated fusion data set; an output prediction module, configured to match corresponding multiple output prediction channels in a new energy output prediction model according to the weather annotation information, and predict the updated fusion data set to obtain a new energy output prediction result.

[0007] Second aspect, the present application also provides a new energy output prediction method under multi-source data fusion analysis. Among them, the new energy output prediction method under multi-source data fusion analysis includes: collecting a multi-source data set of a target area through multi-source sensors, performing data fusion on the multi-source data set to obtain a multi-modal fusion data set; annotating the meteorological data in the multi-source data set through an abnormal weather classifier to obtain weather annotation information; updating the multi-modal fusion data set based on the weather annotation information to obtain an updated fusion data set; and according to the weather annotation information, matching corresponding multiple output prediction channels in the new energy output prediction model, and predicting the updated fusion data set to obtain a new energy output prediction result.

[0008] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0009] A data acquisition module, configured to collect a multi-source data set of a target area through multi-source sensors, perform data fusion on the multi-source data set to obtain a multi-modal fusion data set; a data annotation module, configured to annotate the meteorological data in the multi-source data set through an abnormal weather classifier to obtain weather annotation information; a data update module, configured to update the multi-modal fusion data set based on the weather annotation information to obtain an updated fusion data set; and an output prediction module, configured to match corresponding multiple output prediction channels in the new energy output prediction model according to the weather annotation information, and predict the updated fusion data set to obtain a new energy output prediction result. That is to say, by collecting multi-source data and performing fusion, identifying and annotating abnormal weather, and selecting a suitable model for prediction according to different weather conditions, the error amplification of a single model under specific weather conditions is avoided, and the accuracy of new energy output prediction is improved.

[0010] The above description is only an overview of the technical solutions of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following specifically enumerates the specific embodiments of the present application. It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present application, nor is it used to limit the scope of the present application. Other features of the present application will become easily understandable through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings described below are only exemplary, and for those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0012] Figure 1 It is a schematic structural diagram of a new energy output prediction device under multi-source data fusion analysis of this application;

[0013] Figure 2 It is a schematic flow diagram of a new energy output prediction method under multi-source data fusion analysis of this application.

[0014] Explanation of reference numerals: data acquisition module 11, data annotation module 12, data update module 13, output prediction module 14. Detailed implementation manners

[0015] By providing a new energy output prediction device and method under multi-source data fusion analysis, this application solves the technical problem in the prior art that the accuracy of new energy output prediction is reduced due to the complexity of meteorological data. By collecting multi-source data and performing fusion, abnormal weather is identified and annotated, and a suitable model is selected for prediction according to different weather conditions, avoiding the error amplification of a single model under specific weather conditions and improving the accuracy of new energy output prediction.

[0016] Next, the technical solutions in this application will be described clearly and completely with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited by the example embodiments described here. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application. Additionally, it should be noted that for the sake of description, only the parts related to this application are shown in the accompanying drawings rather than all of them.

[0017] Embodiment 1, please refer to the attached Figure 1 This application provides a new energy output prediction device under multi-source data fusion analysis. Among them, the new energy output prediction device under multi-source data fusion analysis is used to implement the steps of the new energy output prediction method under multi-source data fusion analysis. The new energy output prediction device under multi-source data fusion analysis includes:

[0018] A data acquisition module 11, configured to collect a multi-source data set of a target area through multi-source sensors, and perform data fusion on the multi-source data set to obtain a multi-modal fusion data set.

[0019] Specifically, multiple sensors are used to collect various meteorological data within the target area, resulting in a multi-source dataset. Multi-source sensors refer to devices used to collect data from different sources and of different types, such as wind speed sensors, temperature sensors, humidity sensors, radiation sensors, etc. These sensors can be located at the same position or different positions to collect different meteorological information respectively. A multi-source dataset refers to a collection of raw data provided by different sensors or data sources, and each data source provides different types of information, including meteorological data (temperature, humidity, wind speed, etc.), equipment operation data (power, efficiency, equipment status, etc.), environmental data (geography, obstacles, etc.), satellite remote sensing data, historical weather data, extreme weather warning information, etc. The target area refers to a specific geographical area where new energy output prediction is required, such as a wind farm, a solar photovoltaic farm, etc. The selection of the target area is usually related to the specific energy prediction task.

[0020] The raw data comes from different sensors and has different formats and units, which need to be uniformly processed. The multi-source dataset is subjected to data cleaning to remove invalid or outlier values (such as abnormal data caused by sensor failures). The collection times of different sensors are different, and they are aligned according to timestamps, aligning all data to the same time step. Data fusion refers to integrating, processing, and analyzing data from different sensors or different data sources to obtain a unified and comprehensive dataset. The aligned dataset is subjected to data standardization and normalization processing to eliminate the dimensional differences in the data. The standardized and normalized datasets are merged to obtain a comprehensive multi-modal dataset that contains all the information from different sensors. By collecting meteorological data in the target area through multi-source sensors and performing data fusion, it is possible to comprehensively process data from different sensors and provide a more comprehensive and accurate prediction input.

[0021] The data annotation module 12 is used to annotate the meteorological data in the multi-source dataset through an abnormal weather classifier to obtain weather annotation information.

[0022] Specifically, the trained abnormal weather classifier is used to classify the meteorological data in the multi-source dataset, and a label (i.e., weather annotation information) is assigned to each piece of meteorological data. The abnormal weather classifier is a machine learning or deep learning model used to identify and classify weather events from meteorological data. By training on historical meteorological data, it learns how to distinguish weather types. The meteorological data, including temperature, humidity, wind speed, etc., is extracted from the multi-source dataset. To adapt to the input of the abnormal weather classifier, these data need to be standardized or normalized so that data from different data sources have a unified scale.

[0023] Input the preprocessed meteorological data into the abnormal weather classifier, which is usually a trained machine learning model (such as support vector machine, decision tree, neural network, etc.). It has learned from historical meteorological data how to distinguish and identify weather types. The output of the classifier is the weather type annotation. By annotating abnormal weather in the meteorological data, it ensures that the prediction model can identify the change patterns under different weather conditions, thus making a more accurate prediction of new energy output.

[0024] The data update module 13 is used to update the multi-modal fusion data set based on the weather annotation information to obtain an updated fusion data set.

[0025] Specifically, according to the weather annotation information, adjust and update the original multi-modal fusion data set so that it can better reflect the impact of different weather conditions on new energy output. The weather annotation information identifies the weather type at the current moment (such as normal weather, stormy weather, extreme weather, etc.) through the abnormal weather classifier and associates these weather types with the corresponding multi-modal fusion data set. According to different weather conditions, select different feature data from the multi-modal data set. The updated multi-modal fusion data set will be more in line with the characteristics of different weather conditions. For example, if the current weather is heavy rain or strong wind weather, enhance the attention to features such as wind speed and humidity, or add the annotation information corresponding to extreme weather to optimize the data set to adapt to specific weather conditions. Assign different weights to different features according to different weather conditions, making the prediction of specific meteorological features more accurate. By dynamically adjusting the data set according to weather conditions, the model can better learn the data features under different weather conditions, only retain the data related to the current weather, avoid unnecessary interference during model training, thus improving the calculation efficiency and the prediction accuracy.

[0026] The output prediction module 14 is used to match corresponding multiple output prediction channels in the new energy output prediction model according to the weather annotation information, and predict the updated fusion data set to obtain the new energy output prediction result.

[0027] Specifically, the current weather conditions are identified according to the weather annotation information, and the output prediction channel suitable for the current weather type is matched in the new energy output prediction model. The updated fusion dataset contains different types of data (meteorological data, equipment data, environmental data, etc.) and has been adaptively adjusted according to the weather annotation information. The updated fusion dataset is input into the selected prediction channel for prediction, and the data is processed in the selected output prediction channel to obtain a new energy output prediction result, representing the new energy output value corresponding to the current specific weather conditions (such as heavy rain, normal weather, etc.). By selecting different prediction channels according to the specific weather conditions, each channel focuses on the characteristics of a specific weather type, thereby improving the prediction accuracy and reducing the prediction error caused by weather changes.

[0028] Furthermore, the data acquisition module 11 in the new energy output prediction device under multi-source data fusion analysis is further configured to:

[0029] Align the multi-source dataset according to the acquisition timestamp to obtain an aligned dataset; perform data standardization processing and data normalization processing on the aligned dataset to obtain the multi-modal fusion dataset.

[0030] Specifically, meteorological data is collected in the target area through different types of sensors (such as wind speed sensors, temperature sensors, humidity sensors, etc.). Each data point will contain a timestamp to identify the data acquisition time. Since the sampling frequencies and time steps of different sensors may be different, all data needs to be aligned according to the timestamp. If the timestamps of the data sources are not synchronized, the data is aligned through interpolation (such as linear interpolation, spline interpolation, etc.). For example, if the wind speed data is updated every minute and the temperature data is updated every five minutes, the temperature data can be interpolated by filling with the average value to align its timestamp with the wind speed data. If some data is missing (such as no temperature data at some time points), methods such as filling with the previous value, filling with the next value, or linear interpolation can be used for supplementation.

[0031] Normalize the data of each sensor to eliminate the dimensional differences in the data. The normalization process refers to adjusting the data to a distribution with the same mean and variance, usually by subtracting the mean and dividing by the standard deviation (Z-score normalization). The purpose of normalization is to make the data of different features have the same scale and avoid certain features having too much impact on model training due to scale differences. To ensure that all input features are within the same range, the data of each sensor can be normalized, usually normalizing the data to the interval [0, 1]. Combine the normalized and standardized data sets to obtain a comprehensive multi-modal data set that contains all the information from different sensors. By aligning, normalizing, and standardizing multi-source data, the impacts caused by time differences and different dimensions between different sensors can be effectively eliminated.

[0032] Furthermore, the data annotation module 12 in the new energy output prediction device under the multi-source data fusion analysis is further configured to:

[0033] Interactively obtain a historical meteorological data set and its corresponding historical annotation information set; perform data augmentation on the historical meteorological data set and the historical annotation information set to obtain an augmented meteorological data set and an augmented annotation information set; use the augmented meteorological data set as input data and the augmented annotation information set as labels to perform supervised training on the abnormal weather classifier until the accuracy rate of the abnormal weather classifier reaches the preset requirement to obtain the abnormal weather classifier.

[0034] Specifically, by establishing a connection with meteorological sensors (such as temperature, humidity, air pressure, wind speed, etc. sensors) or meteorological data centers, collect a historical meteorological data set, including multiple meteorological elements such as temperature, humidity, wind speed, and air pressure. Each meteorological data has a timestamp, recording the meteorological conditions at different time points. Manually or automatically annotate the historical meteorological data set, including the weather types (such as sunny, heavy rain, thunderstorm, etc.) within a specific time period. The historical annotation information set is the label information corresponding one-to-one to the historical meteorological data set. The annotation information is generally generated by experts or automated classification systems, indicating the weather type (such as sunny, heavy rain, thunderstorm, etc.) to which each meteorological data belongs, or indicating a specific abnormal weather condition.

[0035] Data augmentation is a technique for generating new training samples from the original data, which is used to improve the diversity of the dataset and enhance the generalization ability of the model. Since the actual samples of extreme weather may be scarce, based on the historical meteorological dataset, the data augmentation method is adopted for the dataset with a small amount of meteorological data samples to increase the sample size and avoid affecting the accuracy of model training. By interpolating between adjacent samples in the meteorological dataset to generate new samples and adding noise perturbations to simulate the changes in meteorological data under different conditions, new synthetic data points are created to increase data diversity. After data augmentation, the generated meteorological data matches the original annotation information. The annotation information set remains unchanged, that is, the augmented data still corresponds to the original label.

[0036] Taking the augmented meteorological dataset as the input data and the augmented annotation information set as the label, a supervised learning method is used to train the abnormal weather classifier. During the training process, the model continuously performs backpropagation and adjustment on the error between the input data and the actual label until the accuracy of the model reaches the preset requirement (such as more than 90%). During training, the model learns how to judge the weather type based on meteorological data (such as temperature, humidity, wind speed, etc.), especially in the case of weather types with scarce data (such as extreme weather), the augmented data helps to improve the recognition ability and generalization ability of the model. The model will be repeatedly trained until its prediction accuracy reaches the preset standard, and its performance can usually be verified by methods such as cross-validation. If the accuracy reaches the requirement, the finally obtained abnormal weather classifier can accurately predict future meteorological data and identify weather abnormal events. By augmenting the dataset, especially for scarce weather types (such as typhoons, heavy rains, etc.), the model can better learn the characteristics of these weathers, thereby enhancing the generalization ability of the model.

[0037] Furthermore, the data annotation module 12 in the new energy output prediction device under the multi-source data fusion analysis is also used for:

[0038] Extract a first historical meteorological dataset with the data volume less than or equal to a predetermined quantity from the historical meteorological dataset, and extract a first annotation information set corresponding to the first historical meteorological dataset from the historical annotation information set; randomly select first meteorological data from the first historical meteorological dataset, and determine first neighboring data of the first meteorological data therefrom; randomly generate synthetic meteorological data between the first meteorological data and the first neighboring data; annotate the synthetic meteorological data based on the annotation information of the first meteorological data and the first neighboring data to obtain synthetic annotation data; add the synthetic meteorological data to the first historical meteorological dataset to obtain a first balanced meteorological dataset; add the synthetic annotation data to the first annotation information set to obtain a first balanced annotation information set; and so on, until the data volume of the first balanced meteorological dataset is greater than the predetermined quantity, add the first balanced meteorological dataset to the enhanced meteorological dataset, and add the first balanced annotation information set to the enhanced annotation information set.

[0039] Specifically, filter out multiple subsets with the data volume less than or equal to a predetermined quantity from the historical meteorological dataset, and extract any one of the subsets as the first historical meteorological dataset. At the same time, extract the annotation information corresponding to the first historical meteorological dataset from the historical annotation information set to obtain the first annotation information set. The predetermined quantity refers to the minimum or expected sample quantity that the target dataset should reach in the data augmentation process to achieve balance or meet the model training requirements, that is, the expected size of the dataset in the data augmentation. If the data of some weather types are less (such as thunderstorms or snow disasters, etc.), set a target quantity so that the data of such weather types are rich enough in the final dataset, thereby improving the model's recognition ability for this weather type.

[0040] Randomly select a piece of data from the first historical meteorological dataset, which is called the first meteorological data. Then, select the meteorological data adjacent to the first meteorological data as the first neighboring data. The time or space of these two pieces of data is close, and they usually have similar meteorological characteristics. Interpolate between the selected first meteorological data and the first neighboring data to obtain synthetic meteorological data.

[0041] Since the first historical meteorological data set is of the same type of meteorological data and its annotation information is the same, the synthetic meteorological data is directly annotated according to the annotation information of the first meteorological data and the first neighboring data to obtain synthetic annotation data. The label information corresponding to the synthetic annotation data and the synthetic meteorological data is usually inferred or directly generated according to the annotation of the meteorological data from which the synthetic data is sourced. The newly generated synthetic meteorological data and synthetic annotation data are respectively added to the first historical meteorological data set and the corresponding first annotation information set to obtain a new first balanced meteorological data set and a first balanced annotation data set. The above steps of data synthesis are continuously repeated until the data volume of the first historical meteorological data set is greater than a predetermined quantity, and then the first balanced meteorological data set and the first balanced annotation data set at this time are respectively added to the enhanced meteorological data set and the enhanced annotation information set. When the total number of samples in the enhanced data set exceeds or equals the predetermined quantity, the data enhancement process stops.

[0042] For other subsets in the historical meteorological data set with a data volume less than or equal to the predetermined quantity, the above data enhancement is performed, and finally an enhanced meteorological data set and an enhanced annotation information set are obtained. Through the data enhancement method, the training effect of the model is improved in the case of unbalanced data samples, and the prediction accuracy of the model caused by too little data is avoided.

[0043] Furthermore, the data annotation module 12 in the new energy output prediction device under the multi-source data fusion analysis is further configured to:

[0044] Calculate the difference between the first meteorological data and the first neighboring data; based on the product of a preset random factor and the difference, and add the product to the first meteorological data to obtain the synthetic meteorological data.

[0045] Specifically, the first meteorological data refers to a minority-class sample randomly selected from the first historical meteorological dataset. The first neighboring data refers to the nearest neighbor of the first meteorological data found in the minority-class sample set (i.e., the first historical meteorological dataset), usually calculated based on the Euclidean distance. For each feature, calculate the difference in features between the selected minority-class sample (the first meteorological data) and its nearest neighbor sample (the first neighboring data), then multiply it by a random number between 0 and 1, and finally add this result to the features of the original minority-class sample. When calculating the interpolation, a random factor is introduced, namely the preset random factor, which is a random number between 0 and 1 and is used to control the variation range in the data generation process. It may be randomly generated within a certain range or be a fixed value. Multiply the calculated difference by the random factor to obtain an adjustment factor. This adjustment factor is added to the first meteorological data to generate a new data sample, called the synthetic meteorological data. By appropriate perturbation, new data samples are added to balance certain scarce categories in the dataset. The generated synthetic data has a certain perturbation, enabling the model to learn more diverse features during the training process, reducing overfitting, and enhancing the model's prediction ability on unknown data.

[0046] Furthermore, the output prediction module 14 in the new energy output prediction device under the multi-source data fusion analysis is further configured to:

[0047] Collect the historical output dataset of the target area within the preset historical time period; perform time serialization processing on the historical output dataset to obtain the historical sequence dataset; construct the multiple output prediction channels with the historical sequence dataset; and connect the multiple output prediction channels in parallel to generate the new energy output prediction model.

[0048] Specifically, obtain the historical output dataset of the target area within the preset historical time period, that is, the new energy data collected in the past period (such as several months or years), such as meteorological data (temperature, humidity, wind speed, etc.), equipment operation data (power, efficiency, equipment status, etc.), environmental data (geography, obstacles, etc.), satellite remote sensing data, historical weather data, extreme weather warning information, etc., and the corresponding new energy output values. Perform time serialization processing on the collected historical output dataset, that is, sort and organize the historical output dataset according to the time stamps to ensure that each data point is consistent with the corresponding time stamp and can accurately represent the time dependence relationship, and obtain the historical sequence dataset. The historical sequence dataset refers to the dataset in which the historical output data is arranged in chronological order and includes multiple data sources (such as meteorological data, equipment data, etc.).

[0049] Construct multiple output prediction channels based on the historical sequence dataset. Train a corresponding output prediction channel for each weather type, and each channel is trained and predicted independently so that the model can make different predictions according to the actual weather conditions. After completing the training of each output prediction channel, connect multiple output prediction channels in parallel. Each channel runs independently and selects the corresponding channel according to different input data to generate prediction results. By designing multiple output prediction channels, independent optimization can be carried out for different factors such as weather conditions and time spans, thereby improving the overall prediction accuracy.

[0050] Further, the output prediction module 14 in the new energy output prediction device under multi-source data fusion analysis is further configured to:

[0051] Traverse the historical sequence dataset to extract the first historical sequence, where the first historical sequence has the first weather annotation information; extract the first time node and the second time node from the first historical sequence according to a preset time step, where the second time node is the next time node of the first time node; use the first historical sequence dataset corresponding to the first time node as input data and the true output value corresponding to the second time node as output data to perform supervised training on the first output prediction channel until the first output prediction channel reaches the convergence condition, obtain the first output prediction channel, and add the first output prediction channel to multiple output prediction channels.

[0052] Specifically, traverse the historical sequence dataset and arbitrarily extract a sequence of data as the first historical sequence, which includes the new energy output value corresponding to the first weather annotation information and multi-source data, that is, the historical data of a specific weather type. Train a corresponding output prediction channel for each weather type, and each channel is trained and predicted independently so that the model can make different predictions according to the actual weather conditions.

[0053] According to a preset time step, a first time node and a second time node are extracted from the first historical sequence. The preset time step is usually 1 hour or 1 day, that is, in each prediction, consecutive time nodes are selected for training. The first time node is one of the time points in the first historical sequence, and the second time node is the next time point immediately following the first time node. The first historical sequence data set corresponding to the first time node is used as input data, and the true output value corresponding to the second time node is used as output data to train the first output prediction channel. The goal is to predict the new energy output value for the next hour based on historical data (such as wind speed, temperature, output value, etc. in the previous hour). During the training process, the goal of supervised training is to minimize the error between the predicted value and the true output value. Usually, loss functions such as mean squared error (MSE) or cross-entropy are used for optimization until the error of the first output prediction channel is reduced to a sufficiently small value or the training no longer improves significantly and meets the convergence condition, then the training is stopped. At this time, the first output prediction channel can predict the new energy output for the next moment based on the meteorological data at the current moment.

[0054] After completing the training of the first output prediction channel, it is added to multiple output prediction channels in the model. Each output prediction channel is trained for different weather types (such as sunny, stormy, rainy, etc.) so that the model can make different predictions according to the actual weather conditions. With the training of each channel, the final model will be able to combine the prediction results of all channels for comprehensive prediction. Through the parallel operation of multiple output prediction channels, the model becomes more flexible, can handle multiple types of meteorological inputs, and finally combines the prediction results of multiple channels to obtain the most accurate output prediction value.

[0055] In summary, the new energy output prediction device under multi-source data fusion analysis provided by this application has the following technical effects:

[0056] A data acquisition module is used to collect a multi-source data set of a target area through multi-source sensors, perform data fusion on the multi-source data set to obtain a multi-modal fusion data set; a data annotation module is used to annotate the meteorological data in the multi-source data set through an abnormal weather classifier to obtain weather annotation information; a data update module is used to update the multi-modal fusion data set based on the weather annotation information to obtain an updated fusion data set; an output prediction module is used to match corresponding multiple output prediction channels in the new energy output prediction model according to the weather annotation information, predict the updated fusion data set to obtain a new energy output prediction result. That is to say, by collecting multi-source data and performing fusion, identifying and annotating abnormal weather, and selecting a suitable model for prediction according to different weather conditions, the error amplification of a single model under specific weather conditions is avoided, and the accuracy of new energy output prediction is improved.

[0057] Embodiment 2. Based on the same inventive concept as the new energy output prediction device under multi-source data fusion analysis in the foregoing Embodiment 1, the present application also provides a new energy output prediction method under multi-source data fusion analysis. Please refer to the attached Figure 2 , the new energy output prediction method under multi-source data fusion analysis includes:

[0058] S100: Collect a multi-source data set of a target area through multi-source sensors, perform data fusion on the multi-source data set to obtain a multi-modal fusion data set; S200: Label the meteorological data in the multi-source data set through an abnormal weather classifier to obtain weather annotation information; S300: Based on the weather annotation information, update the multi-modal fusion data set to obtain an updated fusion data set; S400: According to the weather annotation information, match corresponding multiple output prediction channels in the new energy output prediction model, and predict the updated fusion data set to obtain a new energy output prediction result.

[0059] Further, the step of collecting a multi-source data set of a target area through multi-source sensors, performing data fusion on the multi-source data set to obtain a multi-modal fusion data set includes:

[0060] Align the multi-source data set according to the acquisition timestamp to obtain an aligned data set; perform data standardization processing and data normalization processing on the aligned data set to obtain the multi-modal fusion data set.

[0061] Further, the construction steps of the abnormal weather classifier include:

[0062] Interactively obtain a historical meteorological data set and its corresponding historical annotation information set; perform data augmentation on the historical meteorological data set and the historical annotation information set to obtain an augmented meteorological data set and an augmented annotation information set; use the augmented meteorological data set as input data and the augmented annotation information set as labels to perform supervised training on the abnormal weather classifier until the accuracy rate of the abnormal weather classifier reaches a preset requirement to obtain the abnormal weather classifier.

[0063] Further, the step of performing data augmentation on the historical meteorological data set and the historical annotation information set to obtain an augmented meteorological data set and an augmented annotation information set includes:

[0064] Extract a first historical meteorological data set with a data volume less than or equal to a predetermined quantity from the historical meteorological data set, and extract a first annotation information set corresponding to the first historical meteorological data set from the historical annotation information set; randomly select first meteorological data from the first historical meteorological data set, and determine first neighboring data of the first meteorological data therefrom; randomly generate synthetic meteorological data from the first meteorological data and the first neighboring data; annotate the synthetic meteorological data based on the annotation information of the first meteorological data and the first neighboring data to obtain synthetic annotation data; add the synthetic meteorological data to the first historical meteorological data set to obtain a first balanced meteorological data set; add the synthetic annotation data to the first annotation information set to obtain a first balanced annotation information set; and so on, until the data volume of the first balanced meteorological data set is greater than the predetermined quantity, add the first balanced meteorological data set to the enhanced meteorological data set, and add the first balanced annotation information set to the enhanced annotation information set.

[0065] Further, the randomly generating synthetic meteorological data from the first meteorological data and the first neighboring data includes:

[0066] Calculate the difference between the first meteorological data and the first neighboring data; and add the product of a preset random factor and the difference to the first meteorological data to obtain the synthetic meteorological data.

[0067] Further, the steps for constructing the new energy output prediction model include:

[0068] Collect a historical output data set of the target area in a preset historical time period; perform time series processing on the historical output data set to obtain a historical sequence data set; construct the plurality of output prediction channels with the historical sequence data set; and connect the plurality of output prediction channels in parallel to generate the new energy output prediction model.

[0069] Further, the constructing the plurality of output prediction channels with the historical sequence data set includes:

[0070] Traverse the historical sequence data set to extract a first historical sequence, where the first historical sequence has first weather annotation information; extract a first time node and a second time node from the first historical sequence according to a preset time step, where the second time node is the next time node of the first time node; use the first historical sequence data set corresponding to the first time node as input data, and use the true output value corresponding to the second time node as output data to perform supervised training on a first output prediction channel until the first output prediction channel reaches a convergence condition to obtain the first output prediction channel, and add the first output prediction channel to the plurality of output prediction channels.

[0071] The various embodiments in this specification are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The Figure 1 new energy output prediction device and specific examples in the first embodiment under multi-source data fusion analysis are equally applicable to the new energy output prediction method under multi-source data fusion analysis in this embodiment. Through the detailed description of the new energy output prediction device under multi-source data fusion analysis above, those skilled in the art can clearly understand the new energy output prediction method under multi-source data fusion analysis in this embodiment. Therefore, for the sake of brevity of the specification, it will not be elaborated here. For the method disclosed in the embodiment, since it corresponds to the device disclosed in the embodiment, the description is relatively simple. For related parts, reference can be made to the description of the device part.

[0072] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0073] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application also intends to include these changes and modifications.

Claims

1. A new energy output prediction device based on multi-source data fusion analysis, characterized in that: include: A data acquisition module is used to collect a multi-source data set of a target area through a multi-source sensor, and perform data fusion on the multi-source data set to obtain a multi-modal fusion data set; A data annotation module, which annotates the meteorological data in the multi-source data set through an abnormal weather classifier to obtain weather annotation information; A data updating module, used for updating the multimodal fusion data set based on the weather annotation information to obtain an updated fusion data set; The output prediction module is used to match the corresponding multiple output prediction channels in the new energy output prediction model according to the weather annotation information, predict the updated fusion data set, and obtain the new energy output prediction result.

2. The new energy output prediction device under multi-source data fusion analysis according to claim 1 is characterized in that: The data acquisition module is specifically used for: Aligning the multi-source data sets according to acquisition timestamps to obtain an aligned data set; The aligned data set is subjected to data standardization and data normalization processing to obtain the multimodal fusion data set.

3. The new energy output prediction device under multi-source data fusion analysis according to claim 1 is characterized in that: The data annotation module is specifically used for: Interactively obtain historical meteorological data sets and their corresponding historical annotation information sets; Performing data enhancement on the historical meteorological data set and the historical annotation information set to obtain an enhanced meteorological data set and an enhanced annotation information set; The enhanced meteorological data set is used as input data, and the enhanced annotation information set is used as a label, and supervised training is performed on the abnormal weather classifier until the accuracy of the abnormal weather classifier reaches a preset requirement, thereby obtaining the abnormal weather classifier.

4. The new energy output prediction device under multi-source data fusion analysis according to claim 3 is characterized in that: The data annotation module is specifically used for: Extracting a first historical meteorological data set whose data volume is less than or equal to a predetermined amount from the historical meteorological data set, and extracting a first annotation information set corresponding to the first historical meteorological data set from the historical annotation information set; Randomly select first meteorological data from a first historical meteorological data set, and determine first neighboring data of the first meteorological data; randomly generating synthetic meteorological data from the first meteorological data and the first neighboring data; Annotating the synthetic meteorological data based on the annotation information of the first meteorological data and the first neighboring data to obtain synthetic annotated data; Adding the synthetic meteorological data to the first historical meteorological data set to obtain a first balanced meteorological data set; Adding the synthesized annotation data to the first annotation information set to obtain a first balanced annotation data set; And so on, until the data volume of the first balanced meteorological dataset is greater than the predetermined amount, the first balanced meteorological dataset is added to the enhanced meteorological dataset, and the first balanced annotation dataset is added to the enhanced annotation information set.

5. The new energy output prediction device under multi-source data fusion analysis according to claim 4 is characterized in that: The data annotation module is specifically used for: Calculating the difference between the first meteorological data and the first adjacent data; Based on the product of a preset random factor and the difference, the product is added to the first meteorological data to obtain the synthetic meteorological data.

6. The new energy output prediction device under multi-source data fusion analysis according to claim 1 is characterized in that: The output prediction module is specifically used for: Collecting a historical output data set of the target area within a preset historical time zone; Performing time series processing on the historical output data set to obtain a historical series data set; constructing the plurality of output prediction channels using the historical sequence data set; The multiple output prediction channels are connected in parallel to generate the new energy output prediction model.

7. The new energy output prediction device under multi-source data fusion analysis according to claim 6 is characterized in that: The output prediction module is specifically used for: Traversing the historical sequence data set to extract a first historical sequence, wherein the first historical sequence has first weather annotation information; Extracting a first time node and a second time node from the first historical sequence according to a preset time step, wherein the second time node is a next time node of the first time node; Taking the first historical sequence data set corresponding to the first time node as input data and the actual output value corresponding to the second time node as output data, supervised training is performed on the first output prediction channel until the first output prediction channel reaches the convergence condition, the first output prediction channel is obtained, and the first output prediction channel is added to multiple output prediction channels.

8. A new energy output prediction method based on multi-source data fusion analysis is characterized in that: The method for predicting new energy output under multi-source data fusion analysis is performed by the device for predicting new energy output under multi-source data fusion analysis according to any one of claims 1 to 7, and the method for predicting new energy output under multi-source data fusion analysis comprises: Collecting a multi-source data set of a target area through a multi-source sensor, and performing data fusion on the multi-source data set to obtain a multi-modal fusion data set; Annotating the meteorological data in the multi-source data set by using an abnormal weather classifier to obtain weather annotation information; Based on the weather annotation information, the multimodal fusion dataset is updated to obtain an updated fusion dataset; According to the weather annotation information, the corresponding multiple output prediction channels are matched in the new energy output prediction model, and the updated fusion data set is predicted to obtain the new energy output prediction result.

Images