New energy svm power prediction method based on massive meteorological data feature optimization

Abstract

The present disclosure relates to a new energy SVM power prediction method and device based on massive meteorological data feature optimization, computer equipment, storage medium and computer program product. The method comprises: obtaining meteorological feature data with a correlation degree greater than a preset boundary value with power generation; inputting the meteorological feature data into a new energy power generation prediction model, and outputting the predicted power generation through the new energy power generation prediction model, wherein the new energy power generation prediction model is set to be obtained by training according to the corresponding relationship between sample meteorological feature data and power generation, and the new energy power generation prediction model comprises an SVM model. The method can improve the accuracy of the new energy power generation prediction model, and is beneficial to the formulation and adjustment of the subsequent power system scheduling plan.

Description

Technical Field

[0001] This disclosure relates to the field of new energy power generation technology, and in particular to a new energy SVM power prediction method, device, computer equipment and storage medium based on feature optimization of massive meteorological data. Background Technology

[0002] New energy is a type of renewable energy source, including solar energy, biomass energy, wind energy, geothermal energy, wave energy, ocean current energy, and tidal energy. New energy power generation is the process of generating electricity using these new energy sources.

[0003] Since the renewable energy sources upon which renewable energy generation relies are not stable and continuous, the process of renewable energy generation is random and volatile. Therefore, when using renewable energy generation technology to supply electricity, it is necessary to predict the power generation capacity, which helps the power system to formulate dispatch plans and make timely adjustments to the plans to ensure the stability of power supply.

[0004] However, existing technologies obtain prediction results by directly inputting historical feature data into the power prediction model, resulting in low prediction accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide a new energy SVM power prediction method, device, computer equipment, and storage medium based on feature optimization of massive meteorological data that can improve prediction accuracy in response to the above-mentioned technical problems.

[0006] In a first aspect, embodiments of this disclosure provide a new energy SVM power prediction method based on feature optimization of massive meteorological data. The method includes:

[0007] Acquire meteorological characteristic data whose correlation with power generation is greater than a preset boundary value;

[0008] The meteorological characteristic data is input into the new energy power generation prediction model, and the predicted power generation is output by the new energy power generation prediction model. The new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological characteristic data and power generation. The new energy power generation prediction model includes an SVM model.

[0009] In one embodiment, the method for determining the preset boundary value includes:

[0010] Obtain the relevance value range;

[0011] Multiple candidate boundary values ​​are obtained by taking values ​​at preset step sizes from the value range;

[0012] Multiple candidate new energy power generation prediction models were obtained by training new energy power generation prediction models using meteorological feature data that had a correlation greater than the candidate boundary value with power generation.

[0013] The boundary value corresponding to the model with the highest accuracy score among the multiple candidate new energy power generation prediction models is selected as the preset boundary value.

[0014] In one embodiment, the method for determining the hyperparameters of the new energy power generation prediction model includes:

[0015] A set of sample meteorological feature data, a first value range of a first hyperparameter, and a second value range of a second hyperparameter are obtained. The set includes multiple sample meteorological feature data labeled with power.

[0016] Construct multiple power prediction models, wherein the first hyperparameter and / or the second hyperparameter are different in the multiple power prediction models;

[0017] The accuracy of the multiple power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores.

[0018] The first and second hyperparameters of the power prediction model with the highest prediction accuracy score among the multiple power prediction models are determined as the hyperparameters of the new energy power generation prediction model.

[0019] In one embodiment, the method for determining the first value range includes:

[0020] Multiple first power prediction models are constructed, wherein the first hyperparameters corresponding to the multiple first power prediction models are multiple parameter values ​​within a third value range, and the second hyperparameters are preset second hyperparameters;

[0021] The accuracy of the multiple first power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores.

[0022] The preset range of the first hyperparameter of the first power prediction model with the highest prediction accuracy score among the plurality of first power prediction models is determined as the first value range.

[0023] In one embodiment, the method for determining the second value range includes:

[0024] Multiple second power prediction models are constructed. The first hyperparameter of the multiple second power prediction models is the first hyperparameter of the first power prediction model with the highest prediction accuracy, and the second hyperparameter is multiple parameter values ​​within the fourth value range.

[0025] The accuracy of the multiple second power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores;

[0026] The preset range of the second hyperparameter of the second power prediction model with the highest prediction accuracy score among the multiple second power prediction models is determined as the second value range.

[0027] In one embodiment, the cross-validation method includes:

[0028] Divide the set into multiple subsets;

[0029] By sequentially determining one subset of the plurality of subsets as the test set and the remaining subsets as the training set, multiple sets are obtained;

[0030] The training set from the set group is input into the power prediction model for training to obtain the intermediate process model;

[0031] The test set in the set is input into the intermediate process model to obtain the prediction power corresponding to the test set in the set, and the prediction accuracy corresponding to the set is calculated.

[0032] The average of the multiple prediction accuracies corresponding to the multiple sets is determined as the prediction accuracy score of the power prediction model.

[0033] In one embodiment, the method for determining the kernel function of the new energy power generation prediction model includes:

[0034] Multiple candidate power prediction models are obtained, wherein the kernel functions in the multiple candidate power prediction models are of different types;

[0035] The accuracy of the multiple candidate power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores.

[0036] The kernel function of the candidate power prediction model with the highest prediction accuracy score among the multiple candidate power prediction models is selected as the kernel function of the new energy power generation prediction model.

[0037] In one embodiment, the new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological characteristic data and power generation, including:

[0038] A set of sample meteorological feature data is obtained, the set including multiple sample meteorological feature data labeled with power.

[0039] An initial power prediction model is constructed, wherein training parameters are set in the initial power prediction model;

[0040] The sample meteorological characteristic data is input into the initial power prediction model to obtain the output result;

[0041] Based on the difference between the output result and the labeled power, the initial power prediction model is iteratively adjusted until the difference meets the preset requirements, thus obtaining the new energy power generation prediction model.

[0042] In one embodiment, after acquiring meteorological feature data whose correlation with power generation is greater than a preset boundary value, the method further includes:

[0043] Meteorological feature data are divided into multiple category groups by unsupervised clustering, and the correlation between the meteorological feature data and power generation is greater than a preset boundary value.

[0044] The meteorological feature data is divided according to the multiple category groups.

[0045] Secondly, this disclosure also provides a new energy SVM power prediction device based on feature optimization of massive meteorological data. The device includes:

[0046] The acquisition module is used to acquire meteorological feature data whose correlation with power generation is greater than a preset boundary value;

[0047] The prediction module is used to input the meteorological feature data into the new energy power generation prediction model, and output the predicted power generation through the new energy power generation prediction model. The new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological feature data and power generation, and the new energy power generation prediction model includes an SVM model.

[0048] Thirdly, embodiments of this disclosure also provide a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in any one of the embodiments of this disclosure.

[0049] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in any one of the embodiments of this disclosure.

[0050] Fifthly, embodiments of this disclosure also provide a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in any one of the embodiments of this disclosure.

[0051] In this embodiment, meteorological feature data with a correlation greater than a preset boundary value with power generation is acquired. This acquired meteorological feature data is then input into a new energy power generation prediction model. The new energy power generation prediction model outputs a predicted power generation. This model is trained using the correspondence between sample meteorological feature data and power generation. By setting a boundary value for the correlation with power generation, meteorological data can be filtered. The filtered meteorological feature data is then used to train the prediction model, thereby improving the prediction accuracy of the new energy power generation prediction model. This helps the power system formulate and adjust scheduling plans based on the prediction results, thus improving the stability of power supply. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a method for predicting power generation in one embodiment;

[0053] Figure 2 This is a flowchart illustrating how a preset boundary value is determined in one embodiment;

[0054] Figure 3 This is a graph showing the prediction accuracy of cross-validation in one embodiment.

[0055] Figure 4 This is a graph showing the prediction accuracy of cross-validation in one embodiment.

[0056] Figure 5 This is a graph showing the prediction accuracy of cross-validation in one embodiment.

[0057] Figure 6 This is a flowchart illustrating a method for predicting power generation in one embodiment;

[0058] Figure 7 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0059] Figure 8 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0060] Figure 9 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0061] Figure 10 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0062] Figure 11 This is a graph showing the prediction accuracy of cross-validation in one embodiment.

[0063] Figure 12 This is a graph showing the prediction accuracy of cross-validation in one embodiment.

[0064] Figure 13 This is a graph showing the prediction accuracy of cross-validation in one embodiment.

[0065] Figure 14 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0066] Figure 15 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0067] Figure 16 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0068] Figure 17 This is a schematic diagram of the prediction results of a prediction model in one embodiment;

[0069] Figure 18 This is a structural block diagram of a power generation prediction device in one embodiment;

[0070] Figure 19 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the embodiments of this disclosure will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of this disclosure and are not intended to limit the embodiments of this disclosure.

[0072] In one embodiment, such as Figure 1 As shown, a new energy SVM power prediction method based on feature optimization of massive meteorological data is provided. This embodiment illustrates the application of this method to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0073] Step S110: Obtain meteorological characteristic data whose correlation with power generation is greater than a preset boundary value;

[0074] In the process of generating new energy, the power generation is affected by a variety of factors due to the characteristics of new energy itself. In some scenarios, such as wind power and photovoltaic power generation, the power generation is greatly affected by the weather. Therefore, the power generation is usually predicted by the relationship between the power generation and meteorological data.

[0075] In this embodiment, meteorological characteristic data and historical power generation at different historical moments are first acquired. The meteorological characteristics at different moments may include, but are not limited to, total irradiance, diffuse irradiance, ambient temperature, and wind volume. In one example, historical power generation data and corresponding historical numerical weather prediction data can be acquired. Numerical weather prediction refers to a method that, based on actual atmospheric conditions and under certain initial and boundary conditions, uses a large computer to perform numerical calculations to solve the fluid dynamics and thermodynamics equations describing the weather evolution process, predicting the atmospheric motion state and weather phenomena over a certain period in the future. After acquiring the meteorological characteristic data and power generation at different moments, the absolute value of the correlation between the meteorological characteristics and power generation is calculated using formula (1), where R represents the absolute value of the correlation between a certain meteorological characteristic and power generation within the statistical period; T k This represents the meteorological value of a certain meteorological characteristic at time k. P represents the average value of meteorological characteristic quantities within the statistical period; k This represents the power generation value of new energy sources at time k. This represents the average power generation during the statistical period, where n is the number of moments in the power generation period.

[0076]

[0077] After calculating the absolute value of the correlation between meteorological features and power generation, a boundary value is set. Meteorological features with an absolute correlation value greater than the preset boundary value are selected as the meteorological features needed to predict power generation, while meteorological features with an absolute correlation value less than the preset boundary value are discarded. When power generation prediction is required, meteorological feature data with an absolute correlation value greater than the preset boundary value associated with the power generation period to be predicted are acquired. In one example, the preset boundary value can be calculated based on the relationship between the boundary value and the prediction accuracy score, thereby maximizing the accuracy of the prediction model trained on the meteorological feature data corresponding to the preset boundary value. Meteorological features may include, but are not limited to, temperature, humidity, air pressure, irradiance, and wind speed.

[0078] Step S120: Input the meteorological feature data into the new energy power generation prediction model, and output the predicted power generation through the new energy power generation prediction model. The new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological feature data and power generation. The new energy power generation prediction model includes an SVM model.

[0079] In this embodiment, after obtaining the required meteorological feature data, the meteorological feature data is input into the new energy power generation prediction model, and the predicted power generation is output by the new energy power generation prediction model. The new energy power generation prediction model is obtained by training on the correspondence between sample meteorological feature data and power generation, wherein the sample meteorological feature data is the meteorological feature data corresponding to meteorological features whose absolute correlation value is greater than a preset boundary value. In one example, different prediction period lengths correspond to different prediction models and meteorological feature data; for example, short-term prediction and ultra-short-term prediction correspond to different prediction models and required meteorological feature data. In another example, the new energy power generation prediction model can be an SVM model.

[0080] In this embodiment, meteorological feature data with a correlation greater than a preset boundary value with power generation is acquired. This acquired meteorological feature data is then input into a new energy power generation prediction model. The new energy power generation prediction model outputs a predicted power generation. This model is trained using the correspondence between sample meteorological feature data and power generation. By setting a boundary value for the correlation with power generation, meteorological data can be filtered. The filtered meteorological feature data is then used to train the prediction model, thereby improving the prediction accuracy of the new energy power generation prediction model. This helps the power system formulate and adjust scheduling plans based on the prediction results, thus improving the stability of power supply.

[0081] In one embodiment, such as Figure 2 As shown, the method for determining the preset boundary value includes:

[0082] Step S210: Obtain the relevance value range;

[0083] Step S220: Take values ​​from the value range at preset step intervals to obtain multiple candidate boundary values;

[0084] Step S230: Train new energy power generation prediction models using meteorological feature data that have a correlation with power generation greater than the candidate boundary value, and obtain multiple candidate new energy power generation prediction models.

[0085] Step S240: Select the boundary value corresponding to the model with the highest accuracy score from the multiple candidate new energy power generation prediction models as the preset boundary value.

[0086] In this embodiment of the disclosure, when determining the preset boundary value of the correlation, the value range of the correlation is first determined. Since the correlation is expressed in absolute value in this embodiment, the value range of the correlation is 0 to 1. Multiple candidate boundary values ​​are obtained within this value range by taking values ​​according to a preset step size. The preset step size can be a suitable step size value selected based on the actual scenario. These multiple candidate boundary values ​​are used as boundary values ​​to determine multiple sets of meteorological feature data. A power prediction model is trained based on the relationship between the multiple sets of meteorological feature data and power generation, resulting in multiple trained candidate power generation prediction models. The accuracy of the multiple trained candidate power generation prediction models is verified, and the boundary value corresponding to the model with the highest accuracy score is determined as the preset boundary value. Generally, the higher the accuracy score, the higher the accuracy of the model. In one example, the accuracy verification method can be verification through a validation set or cross-validation to obtain the model accuracy score of the prediction model and perform model accuracy testing. In one example, the preset step size can be set to 0.01.

[0087] In this embodiment, multiple candidate boundary values ​​are determined within the range of correlation values ​​according to a preset step size. Based on the multiple candidate boundary values, corresponding meteorological feature data is selected for model training. The boundary value corresponding to the training model with the highest accuracy score is taken as the preset boundary value. In this way, the optimal boundary value can be found, the best meteorological features can be selected, the accuracy of the power prediction model is improved, and it is beneficial to the subsequent prediction of power generation and the formulation and adjustment of power system dispatch plans.

[0088] In one embodiment, the method for determining the hyperparameters of the new energy power generation prediction model includes:

[0089] A set of sample meteorological feature data, a first value range of a first hyperparameter, and a second value range of a second hyperparameter are obtained. The set includes multiple sample meteorological feature data labeled with power.

[0090] Construct multiple power prediction models, wherein the first hyperparameter and / or the second hyperparameter are different in the multiple power prediction models;

[0091] The accuracy of the multiple power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores.

[0092] The first and second hyperparameters of the power prediction model with the highest prediction accuracy score among the multiple power prediction models are determined as the hyperparameters of the new energy power generation prediction model.

[0093] In this embodiment of the disclosure, when constructing a new energy power generation prediction model, it is first necessary to determine the hyperparameters of the power generation model. The hyperparameters of the power generation model can be one or multiple. A set of sample meteorological feature data is obtained, including multiple sample meteorological feature data labeled with power. A first value range corresponding to a first hyperparameter and a second value range corresponding to a second hyperparameter are obtained. The setting methods for the first and second value ranges can include, but are not limited to, setting them based on experience according to actual scenarios or setting them based on the accuracy of the prediction model under different hyperparameter values. Multiple power prediction models are determined based on the first and second value ranges. The first and / or second hyperparameters of the multiple power prediction models are different, where the first hyperparameter is a parameter value obtained from the first value range, and the second hyperparameter is a parameter value obtained from the second value range. In one example, multiple candidate first hyperparameters are determined within a first value range according to a preset step size, and multiple candidate second hyperparameters are determined within a second value range according to a preset step size. Multiple power prediction models are constructed by non-repeating permutations and combinations of the multiple candidate first and second hyperparameters, and the first and / or second hyperparameters of the multiple power prediction models are different. After obtaining multiple power prediction models, multiple prediction accuracy scores are obtained for the multiple power prediction models using cross-validation based on a pre-acquired set of sample meteorological feature data. The first and second hyperparameters corresponding to the power prediction model with the highest prediction accuracy score are determined as the hyperparameters of the new energy power generation prediction model. The prediction accuracy is typically the prediction accuracy obtained based on the prediction results and actual results. In one example, the higher the model accuracy score, the higher the model's prediction accuracy. In one example, the power prediction model can be an SVM prediction model, using a Gaussian kernel function as the penalty function. The first hyperparameter can be the penalty factor, and the second hyperparameter can be the kernel coefficient of the kernel function. Through the embodiments of this disclosure, the penalty factor and the value of the kernel function corresponding to the prediction model with the highest accuracy under the Gaussian kernel function can be determined.

[0094] In this embodiment, by determining the prediction accuracy of the model with first and second hyperparameters under different values, the prediction performance of the prediction model under different hyperparameters is obtained. The first and second hyperparameters of the prediction model with the highest prediction accuracy score are determined as the parameters of the prediction model, thereby improving the accuracy of the prediction results of the final prediction model and ensuring the accuracy of the prediction results.

[0095] Figure 3 This is a schematic diagram illustrating the prediction accuracy curve of hyperparameter-optimized cross-validation according to an exemplary embodiment, with reference to... Figure 3As shown, the horizontal axis represents the kernel coefficients of the kernel function, and the vertical axis represents the model accuracy score. Each curve represents the relationship between the model accuracy score and the kernel coefficients under different penalty factors C. Since the local optimum of the penalty factor is 3 and the local optimum of the kernel coefficient is 0.0013, the optimization range of the penalty factor is set to [2, 5] with a step size of 0.1; the optimization range of the kernel coefficient is set to [0.0001, 0.002] with a step size of 0.0001. The final maximum accuracy is 86.45%, corresponding to a penalty factor value of 4.9 and a kernel coefficient value of 0.0003, which are the optimal values ​​of the penalty factor and kernel coefficient. In this example, the optimization range of the penalty factor and kernel coefficient is obtained through cross-validation.

[0096] In one embodiment, the method for determining the first value range includes:

[0097] Multiple first power prediction models are constructed, wherein the first hyperparameters corresponding to the multiple first power prediction models are multiple parameter values ​​within a third value range, and the second hyperparameters are preset second hyperparameters;

[0098] The accuracy of the multiple first power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores.

[0099] The preset range of the first hyperparameter of the first power prediction model with the highest prediction accuracy score among the plurality of first power prediction models is determined as the first value range.

[0100] In this embodiment of the disclosure, when determining the first value range of the first hyperparameter, multiple first power prediction models are first constructed. The first hyperparameters corresponding to these multiple first power prediction models are multiple parameter values ​​within a third value range, and the corresponding second hyperparameter is a pre-set fixed parameter value. The second hyperparameter can be a parameter value selected based on the actual scenario. In one example, the third value range can be a suitable value range selected based on the actual scenario and requirements, and the multiple parameter values ​​can be multiple first hyperparameter values ​​obtained by taking values ​​within the third value range at preset step intervals. After obtaining multiple first power prediction models, multiple prediction accuracy scores corresponding to the multiple first power prediction models are obtained using cross-validation based on a set of pre-acquired sample meteorological feature data. The preset range of the first hyperparameter corresponding to the first power prediction model with the highest prediction accuracy score is determined as the first value range. The minimum value of the preset range can be the corresponding first hyperparameter minus a preset value, and the maximum value can be the corresponding first hyperparameter plus a preset value. The prediction accuracy is typically the prediction accuracy obtained based on the prediction result and the actual result. In one example, the higher the model accuracy score, the higher the model's prediction accuracy.

[0101] In this embodiment, the value of the second hyperparameter is first fixed, and multiple first power prediction models are determined according to the values ​​of multiple first hyperparameters within a third value range. The accuracy scores corresponding to the multiple first power prediction models are determined by cross-validation. The preset range of the first hyperparameter corresponding to the model with the highest accuracy score is determined as the first value range. This enables the determination of the optimal value of the first hyperparameter, thereby allowing the optimization of the remaining hyperparameters based on the optimal value of the first hyperparameter. This helps to subsequently jointly seek the global optimal value and improve the accuracy of the prediction model.

[0102] Figure 4 This is a schematic diagram illustrating a hyperparameter-optimized prediction accuracy curve according to an exemplary embodiment, with reference to... Figure 4 As shown, the horizontal axis represents the value of the penalty factor, and the vertical axis represents the model accuracy score. Based on the actual scenario requirements, the penalty factor parameter optimization range is set to [0.1, 4], with a step size of 0.1; the kernel coefficient is fixed at 0.001, and the kernel function is a Gaussian kernel function. The penalty factor value corresponding to the maximum accuracy is 3, i.e., the local optimal value of the penalty factor is 3.

[0103] In one embodiment, the method for determining the second value range includes:

[0104] Multiple second power prediction models are constructed. The first hyperparameter of the multiple second power prediction models is the first hyperparameter of the first power prediction model with the highest prediction accuracy, and the second hyperparameter is multiple parameter values ​​within the fourth value range.

[0105] The accuracy of the multiple second power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores;

[0106] The preset range of the second hyperparameter of the second power prediction model with the highest prediction accuracy score among the multiple second power prediction models is determined as the second value range.

[0107] In this embodiment of the disclosure, when determining the second value range of the second hyperparameter, multiple second power prediction models are first constructed. The first hyperparameter corresponding to each of the multiple second power prediction models is the first hyperparameter of the first power prediction model with the highest prediction accuracy, and the corresponding second hyperparameter is multiple parameter values ​​within a fourth value range. In one example, the fourth value range can be a suitable value range selected based on the actual scenario and requirements, and the multiple parameter values ​​can be multiple second hyperparameter values ​​obtained by taking values ​​within the fourth value range at preset step intervals. After obtaining multiple second power prediction models, multiple prediction accuracy scores corresponding to the multiple second power prediction models are obtained by cross-validation based on a set of pre-acquired sample meteorological feature data. The preset range of the second hyperparameter corresponding to the second power prediction model with the highest prediction accuracy score is determined as the second value range. The minimum value of the preset range can be the corresponding second hyperparameter minus a preset value, and the maximum value can be the corresponding second hyperparameter plus a preset value. The prediction accuracy is typically the prediction accuracy obtained based on the prediction result and the actual result. In one example, the higher the model accuracy score, the higher the model's prediction accuracy.

[0108] In this embodiment, the first hyperparameter is fixed as the optimal first hyperparameter obtained through multiple first power prediction models. Multiple second power prediction models are determined according to the values ​​of multiple second hyperparameters within a fourth value range. Accuracy scores corresponding to multiple second power prediction models are obtained by cross-validation. The preset range of the second hyperparameter corresponding to the model with the highest accuracy score is determined as the second value range. This can optimize the second hyperparameter based on the optimal first hyperparameter, which helps to jointly seek the global optimal value and improve the accuracy of the prediction model.

[0109] Figure 5 This is a schematic diagram illustrating a hyperparameter-optimized prediction accuracy curve according to an exemplary embodiment, with reference to... Figure 5 As shown, the horizontal axis represents the values ​​of the kernel coefficients, and the vertical axis represents the model accuracy score. The penalty factor is set to a local optimum of 3. Based on the actual scenario requirements, the kernel coefficient values ​​are set to (0.0005, 0.0025), with a step size of 0.0001 and a Gaussian kernel function. The kernel coefficient value corresponding to the maximum accuracy is 0.0013, that is, the local optimum of the kernel coefficient is 0.0013.

[0110] In one embodiment, the cross-validation method includes:

[0111] Divide the set into multiple subsets;

[0112] By sequentially determining one subset of the plurality of subsets as the test set and the remaining subsets as the training set, multiple sets are obtained;

[0113] The training set from the set group is input into the power prediction model for training to obtain the intermediate process model;

[0114] The test set in the set is input into the intermediate process model to obtain the prediction power corresponding to the test set in the set, and the prediction accuracy corresponding to the set is calculated.

[0115] The average of the multiple prediction accuracies corresponding to the multiple sets is determined as the prediction accuracy score of the power prediction model.

[0116] In this embodiment, when cross-validating the prediction model, the acquired sample meteorological feature data set is first divided into multiple subsets. One subset is used as a test set to test the prediction accuracy of the prediction model, and the remaining subsets are used as training sets to train the prediction model. This process results in multiple sets equal in number to the number of subsets. The training sets from these multiple sets are sequentially input into the constructed power prediction model. The power prediction model is iteratively adjusted based on the output results until the difference between the output results and the power labels in the sets meets the requirements, resulting in multiple intermediate process models corresponding to the multiple set groups. The test sets from each set group are then input into the corresponding intermediate process model to obtain the prediction accuracy corresponding to the test sets in the multiple set groups. The average of these multiple prediction accuracies is taken as the prediction accuracy score of the power prediction model. In one example, the basic idea of ​​the k-fold cross-validation method adopted in this embodiment is as follows: First, the original training set is randomly divided into k parts without repetition; second, one part is selected as the validation set, and the remaining k-1 parts are used as the training set for model training. After training on the training set, a model is obtained, and this model is tested on the validation set, and the evaluation metric of the model is saved; third, the second step is repeated k times (ensuring that each subset has one chance to be used as the validation set); fourth, the average of the k test metrics is calculated as an estimate of the model accuracy and used as the performance metric of the model under the current k-fold cross-validation. Typically, k is usually taken as 10; when the original training set is small, k can be larger so that the proportion of the training set to the whole is not too small, but the number of trained models also increases accordingly; when the original training set is large, k can be smaller.

[0117] In this embodiment of the disclosure, the prediction accuracy score of the prediction model can be determined based on the sample set through cross-validation, thereby effectively determining the performance of the prediction model, which helps to determine the various parameters of the prediction model, select the most accurate prediction model, and improve the accuracy of subsequent prediction models in predicting power generation.

[0118] In one embodiment, the method for determining the kernel function of the new energy power generation prediction model includes:

[0119] Multiple candidate power prediction models are obtained, wherein the kernel functions in the multiple candidate power prediction models are of different types;

[0120] The accuracy of the multiple candidate power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores.

[0121] The kernel function of the candidate power prediction model with the highest prediction accuracy score among the multiple candidate power prediction models is selected as the kernel function of the new energy power generation prediction model.

[0122] In this embodiment, the new energy power generation prediction model can be a prediction model including a kernel function. In this case, the performance of prediction models constructed with different kernel functions is not the same; that is, the prediction accuracy of prediction models corresponding to different kernel functions is also different. Therefore, it is necessary to determine the optimal kernel function. By using different kernel functions as the kernel functions of the prediction model, multiple candidate power prediction models are obtained. Based on a set of pre-acquired sample meteorological feature data, multiple prediction accuracy scores are obtained for the multiple candidate power prediction models using cross-validation. The kernel function corresponding to the candidate power prediction model with the highest prediction accuracy score is determined as the kernel function of the new energy power generation prediction model. In one example, each kernel function can also correspond to multiple prediction models with different boundary values ​​and hyperparameters, thereby determining the prediction accuracy score corresponding to the optimal hyperparameters and optimal boundary values ​​under each kernel function. The kernel function corresponding to the highest accuracy score is determined as the kernel function of the final prediction model. The kernel function may include, but is not limited to, linear kernel functions, Gaussian radial basis function kernel functions, polynomial kernel functions, and nonlinear interaction kernel functions of neurons.

[0123] In this embodiment, multiple different prediction models are obtained by setting different kernel functions, and the accuracy scores of multiple candidate power prediction models are obtained by cross-validation. The kernel function corresponding to the model with the highest accuracy score is determined as the kernel function of the new energy power generation prediction model. This can determine the optimal kernel function and improve the accuracy of the prediction model.

[0124] In one embodiment, the new energy power generation prediction model is trained based on the correspondence between sample meteorological characteristic data and power generation, including:

[0125] A set of sample meteorological feature data is obtained, the set including multiple sample meteorological feature data labeled with power.

[0126] An initial power prediction model is constructed, wherein training parameters are set in the initial power prediction model;

[0127] The sample meteorological characteristic data is input into the initial power prediction model to obtain the output result;

[0128] Based on the difference between the output result and the labeled power, the initial power prediction model is iteratively adjusted until the difference meets the preset requirements, thus obtaining the new energy power generation prediction model.

[0129] In this embodiment, when determining the new energy power generation prediction model, a set of sample meteorological feature data is first obtained. This set is typically a pre-prepared collection containing sample meteorological features, each labeled with a corresponding power tag, which is usually the actual power generation during operation. An initial power generation prediction model is constructed, with initial training parameters set. After inputting the sample meteorological feature data into the initial model, an initial result is output. The initial result is compared with the power generation tag corresponding to the sample meteorological feature data, and the initial training parameters of the initial model are adjusted based on the difference between the initial result and the power generation tag. This process is repeated iteratively until the difference between the power generation output by the adjusted prediction model and the power generation tag meets a preset requirement. At this point, the prediction model is determined to be a new energy power generation prediction model. The preset requirement can be a small range within the power generation tag set according to the actual scenario. When the prediction result falls within a small range of the power generation tag, the prediction result is considered relatively accurate.

[0130] In this embodiment, an initial new energy power generation prediction model is constructed. The initial new energy power generation prediction model is trained using sample meteorological feature data and power generation labels, thereby obtaining a new energy power generation prediction model, which helps the power system to formulate and adjust dispatch plans based on the prediction results.

[0131] In one embodiment, after acquiring meteorological characteristic data whose correlation with power generation is greater than a preset boundary value, the method further includes:

[0132] Meteorological feature data are divided into multiple category groups by unsupervised clustering, and the correlation between the meteorological feature data and power generation is greater than a preset boundary value.

[0133] The meteorological feature data is divided according to the multiple category groups.

[0134] In this embodiment, meteorological features are divided into multiple groups using unsupervised clustering, each group representing a category of meteorological features. The acquired meteorological feature data is then divided into multiple category groups according to these groups. In one example, the number of clusters is first selected (similar to K-Means), and the Gaussian distribution parameters (mean and variance) for each cluster are randomly initialized. Alternatively, a relatively accurate mean and variance can be obtained by observing the data. Given the Gaussian distribution for each cluster, the probability of each data point belonging to each cluster is calculated; a point closer to the center of the Gaussian distribution is more likely to belong to that cluster. Based on these probabilities, Gaussian distribution parameters are calculated to maximize the probability of the data points. These new parameters can be calculated using a weighted average of the data point probabilities, where the weights are the probabilities of the data points belonging to the cluster. This process is repeated iteratively until the changes during iteration are minimal. In this embodiment, meteorological features are used as features in the unsupervised clustering, for example: S = (TEM_meter, WD), where S is a cluster after unsupervised clustering, TEM_meter represents the temperature at 10 meters and 100 meters, and WD represents the wind direction.

[0135] In this embodiment, meteorological features are divided into multiple categories using unsupervised clustering. After clustering, the meteorological features can be divided into multiple categories, and the attributes of each category are basically the same. When training the prediction model, training on meteorological features of the same category can improve the prediction accuracy of the model. At the same time, during the prediction process, prediction based on meteorological features of the same category can also improve the prediction accuracy of the prediction model.

[0136] Figure 6 This is a schematic flowchart illustrating a method for predicting power generation according to an exemplary embodiment, with reference to... Figure 6As shown, after obtaining historical operating power data of new energy sources and historical numerical weather forecast (NWP) data of the target area, the correlation coefficient between meteorological characteristics and power at different times is calculated using formula (1) to study the temporal correlation between meteorological characteristics and new energy power generation. The boundary value r of the correlation coefficient between meteorological characteristics and power is set sequentially, and the feature quantity with a feature value greater than r calculated in formula (1) is selected as the input meteorological feature quantity for prediction. After classifying the meteorological feature quantity by unsupervised clustering, the kernel function of the SVM training model is selected. The kernel function can include, but is not limited to, the linear kernel function LINEAR, the Gaussian radial basis function RBF, the polynomial kernel function POLY, and the nonlinear action kernel function of the neuron Sigmoid. The above kernel functions are selected sequentially, and the SVM model parameters are optimized by using the K-CV (K-fold cross-validation) method based on the clustering results of historical numerical weather forecast data and power data, including the penalty factor C and the kernel coefficient gamma of the kernel function. Using historical numerical weather prediction (NWP) data and renewable energy power generation data as examples, the power generation prediction results and prediction accuracy scores were calculated. Prediction accuracy scores for different kernel functions and model parameters were obtained under the correlation coefficient boundary values. All values ​​of the established meteorological characteristics and power correlation boundary were iterated, and the selection of kernel functions and determination of hyperparameters were repeated. The SVM prediction model with the highest prediction accuracy score and the numerical weather prediction feature corresponding to the correlation coefficient boundary value were used as the prediction model and numerical weather prediction feature for this renewable energy power station. The prediction results and prediction accuracy of the final prediction model were calculated to obtain the prediction accuracy of the renewable energy power generation prediction model for this renewable energy power station.

[0137] In one example, in the short-term prediction of the power generation of a wind power plant, before parameter optimization, the penalty coefficient C is 0.5, the penalty function is 'rbf', and the bandwidth of the penalty function gamma = 0.001. The training set data is as follows: the correlation between the NWP feature and the power is calculated separately. Within the preset range of the correlation, multiple boundary values ​​are selected sequentially to calculate the prediction accuracy score of the corresponding model. The boundary value of the correlation coefficient corresponding to the highest prediction accuracy score of the model is 0.2. Therefore, the boundary value of the correlation coefficient is set to 0.2. Features with a correlation greater than 0.2 are selected as: TEM_meter and WD, a total of 2. Location: A wind farm in City A; Time: 2020 / 7 / 1 8:00:00 - 2021 / 8 / 7 7:45:00, a total of 400 days of data; Time step of each test: 15min; Data volume: 115200 time-point data for training. The test set data and training set meteorological features are consistent. Location: A wind farm in City A; Time: 8:00:00 on August 5, 2021 – 7:45:00 on December 15, 2021, totaling 130 days of data; Time step for each test: 15 minutes; Data volume: 37,440 time points for training. Prediction results are as follows. Figure 7 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the calculated prediction accuracy is 83.37%.

[0138] After parameter optimization, the penalty coefficient C is 4.9, the penalty function is 'rbf', and the penalty function bandwidth gamma = 0.0003. The training and test sets remain unchanged, and the prediction results are as follows: Figure 8 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 83.63%.

[0139] In one example, in the ultra-short-term prediction of the power generation of a wind power plant, before parameter optimization, the penalty coefficient C is 0.5, the penalty function is 'rbf', and the bandwidth of the penalty function gamma = 0.001. The training set data is as follows: the correlation between the NWP feature and the power is calculated separately. Within the preset range of the correlation, multiple boundary values ​​are selected sequentially to calculate the prediction accuracy score of the corresponding model. The boundary value of the correlation coefficient corresponding to the highest prediction accuracy score of the model is 0.2. The correlation coefficient boundary value is set to 0.2. Features with a correlation greater than 0.2 are selected as: TEM_meter and WD, a total of 2. Location: A wind farm in City A; Time: 2020 / 7 / 1 8:00:00—2021 / 8 / 7 7:45:00, a total of 400 days of data; Time step of each test: 15min; Data volume: 115200 time-point data for training. The meteorological features selected for the test set data are consistent with those for the training set. Features with a correlation greater than 0.2 are selected as features: TEM_meter and WD, a total of 2. Location: A wind farm in City A; Time: 2021 / 8 / 5 8:00:00—2021 / 12 / 15 7:45:00; 130 days of data; Time step for each test: 15 minutes; Data volume: 37440 time-stamped data points for training. Prediction results are as follows. Figure 9 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 91.64%. The prediction result after parameter optimization is shown below. Figure 10 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 91.75%.

[0140] In one example, during the parameter optimization process of a prediction model for a photovoltaic power plant, such as... Figure 11 As shown, the penalty factor was optimized, and the local optimum penalty factor was obtained as 13.3. Based on the penalty factor, the kernel coefficient was further optimized, as follows: Figure 12 As shown, the local optimum of the kernel coefficients is 0.00002. Finally, joint optimization is performed, as follows: Figure 13 As shown, the kernel coefficient corresponding to the maximum model accuracy score is 0.00003, and the penalty factor is 13.6, which is the optimal hyperparameter.

[0141] In the short-term forecasting of a photovoltaic power plant, before parameter optimization, the penalty coefficient C was 0.5, the penalty function was 'rbf', and the bandwidth of the penalty function gamma = 0.001. The training set data consisted of multiple boundary values ​​selected sequentially within a preset correlation range to calculate the corresponding model prediction accuracy score. The boundary value of the correlation coefficient corresponding to the highest prediction accuracy score was 0.3, and the boundary was set to 0.3. The NWP data features obtained were: RHU_hpa_200 (200Pa relative humidity), RHU_meter (100m relative humidity), SR (total radiation), and SWDDIF (scattered radiation), a total of 4 features; location: a photovoltaic power station in City A; time: 2020 / 7 / 1 8:00:00-2021 / 8 / 7 7:45:00; a total of 400 days of data. Reports were submitted daily starting at 8:00 AM, with 3 days of data per report; only the forecast data from the first day was used for training; the time step for each test was 15 minutes; the data volume was 115,200 records. The test set data consisted of the same meteorological features as the training set; location: a photovoltaic power station in City A; time: 8:00:00 on August 5, 2021 to 7:45:00 on December 15, 2021, totaling 130 days of data. Data was reported starting at 8:00 AM daily, with each test covering 3 days of data; the time step for each test was 15 minutes; and the data volume was 12,480 records. The test results are as follows: Figure 14 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 87.77%. The prediction result after parameter optimization is shown below. Figure 15 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 87.85%.

[0142] In the ultra-short-term forecasting of a photovoltaic power plant, before parameter optimization, the penalty coefficient C was 0.5, the penalty function was 'rbf', and the bandwidth of the penalty function gamma = 0.001. The training set data was prepared by sequentially selecting multiple boundary values ​​within a preset correlation range to calculate the corresponding model prediction accuracy score. The boundary value of the correlation coefficient corresponding to the highest prediction accuracy score was 0.3, and this boundary value was set to 0.3. The NWP data features obtained were: RHU_hpa_200 (200Pa relative humidity), RHU_meter (100m relative humidity), SR (total radiation), and SWDDIF (scattered radiation), totaling four features. Location: A photovoltaic power plant in City A; Time: 2020 / 7 / 1 8:00:00 - 2021 / 8 / 7 7:45:00; a total of 400 days of data. Reports are submitted daily starting at 8:00 AM, with each batch containing 3 days of data. Only the first day's forecast data is used for training. Each test session has a time step of 15 minutes and contains 115,200 data points. The test set data consists of the same meteorological features as the training set, located at a photovoltaic power station in City A, from 8:00:00 AM on August 5, 2021 to 7:45:00 AM on December 15, 2021, totaling 130 days of data. Reports are submitted daily starting at 8:00 AM, with each batch containing 3 days of data. Each test session has a time step of 15 minutes and contains 12,480 data points. Test results are as follows. Figure 16 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 91.56%. The prediction result after parameter optimization is shown below. Figure 17 As shown, the light-colored curve represents the actual power generation, and the dark-colored curve represents the predicted result. The vertical axis represents power in kW, and the prediction accuracy is 91.73%.

[0143] In this embodiment of the disclosure, when the prediction model is an SVM model, after joint optimization of the model's penalty factor and kernel coefficients through cross-validation, the prediction accuracy of the optimized model is greater than that of the unoptimized model. By optimizing the model's parameters through cross-validation, the prediction accuracy of the prediction model is improved, which is helpful for the formulation and adjustment of power system scheduling plans.

[0144] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple steps or stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0145] Based on the same inventive concept, this disclosure also provides a power generation prediction device for implementing the power generation prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more power generation prediction device embodiments provided below can be found in the limitations of the power generation prediction method described above, and will not be repeated here.

[0146] In one embodiment, such as Figure 18 As shown, a new energy SVM power prediction method based on feature optimization of massive meteorological data is provided, including: an acquisition module, used to acquire meteorological feature data with a correlation greater than a preset boundary value with power generation;

[0147] The prediction module is used to input the meteorological feature data into the new energy power generation prediction model, and output the predicted power generation through the new energy power generation prediction model. The new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological feature data and power generation, and the new energy power generation prediction model includes an SVM model.

[0148] In one embodiment, the module for determining the preset boundary value includes:

[0149] The acquisition module is used to obtain the relevance value range;

[0150] The value retrieval module is used to retrieve values ​​from the value range at preset step intervals to obtain multiple candidate boundary values;

[0151] The training module is used to train new energy power generation prediction models using meteorological feature data that are more correlated with power generation than the candidate boundary value, and obtain multiple candidate new energy power generation prediction models.

[0152] The selection module is used to select the boundary value corresponding to the model with the highest accuracy score from the multiple candidate new energy power generation prediction models as the preset boundary value.

[0153] In one embodiment, the module for determining the hyperparameters of the new energy power generation prediction model includes:

[0154] The acquisition module is used to acquire a set of sample meteorological feature data, a first value range of a first hyperparameter, and a second value range of a second hyperparameter. The set includes multiple sample meteorological feature data labeled with power tags.

[0155] A building module is used to build multiple power prediction models, wherein the first hyperparameter and / or the second hyperparameter are different in the multiple power prediction models;

[0156] The testing module is used to test the accuracy of the multiple power prediction models using cross-validation to obtain multiple prediction accuracy scores.

[0157] The determination module is used to determine the first and second hyperparameters of the power prediction model with the highest prediction accuracy score from the multiple power prediction models as the hyperparameters of the new energy power generation prediction model.

[0158] In one embodiment, the module for determining the first value range includes:

[0159] The construction module is used to construct multiple first power prediction models, wherein the first hyperparameters corresponding to the multiple first power prediction models are multiple parameter values ​​within a third value range, and the second hyperparameters are preset second hyperparameters;

[0160] The testing module is used to test the accuracy of the multiple first power prediction models using cross-validation to obtain multiple prediction accuracy scores.

[0161] The determining module is used to determine, from the plurality of first power prediction models, the preset range of the first hyperparameter of the first power prediction model with the highest prediction accuracy score as the first value range.

[0162] In one embodiment, the module for determining the second value range includes:

[0163] The construction module is used to construct multiple second power prediction models. The first hyperparameter of the multiple second power prediction models is the first hyperparameter of the first power prediction model with the highest prediction accuracy, and the second hyperparameter is multiple parameter values ​​within the fourth value range.

[0164] The testing module is used to test the accuracy of the multiple second power prediction models using cross-validation to obtain multiple prediction accuracy scores.

[0165] The determination module is used to determine the preset range of the second hyperparameter of the second power prediction model with the highest prediction accuracy score from the plurality of second power prediction models as the second value range.

[0166] In one embodiment, the cross-validation module includes:

[0167] A partitioning module is used to divide the set into multiple subsets;

[0168] The first determining module is used to sequentially determine one subset of the plurality of subsets as the test set and the remaining subsets as the training set, thereby obtaining a plurality of set groups;

[0169] The first input module is used to input the training set from the set group into the power prediction model for training, so as to obtain the intermediate process model.

[0170] The second input module is used to input the test set in the set group into the intermediate process model, obtain the prediction power corresponding to the test set in the set group, and calculate the prediction accuracy score corresponding to the set group.

[0171] The second determining module is used to determine the average value of the multiple prediction accuracies corresponding to the multiple set groups as the prediction accuracy score of the power prediction model.

[0172] In one embodiment, the module for determining the kernel function of the new energy power generation prediction model includes:

[0173] An acquisition module is used to acquire multiple candidate power prediction models, wherein the kernel functions in the multiple candidate power prediction models are of different types;

[0174] The testing module is used to test the accuracy of the multiple candidate power prediction models using cross-validation to obtain multiple prediction accuracy scores.

[0175] The determination module is used to determine the kernel function of the candidate power prediction model with the highest prediction accuracy score from the multiple candidate power prediction models as the kernel function of the new energy power generation prediction model.

[0176] In one embodiment, the prediction module includes:

[0177] The acquisition module is used to acquire a set of sample meteorological feature data, the set including multiple sample meteorological feature data labeled with power tags;

[0178] A construction module is used to build an initial power prediction model, which includes training parameters.

[0179] The input module is used to input the sample meteorological characteristic data into the initial power prediction model and obtain the output result;

[0180] The adjustment module is used to iteratively adjust the initial power prediction model based on the difference between the output result and the labeled power tag, until the difference meets the preset requirements, thereby obtaining the new energy power generation prediction model.

[0181] In one embodiment, the acquisition module is followed by:

[0182] The clustering module is used to divide meteorological feature data into multiple category groups through unsupervised clustering, wherein the correlation between the meteorological feature data and power generation is greater than a preset boundary value.

[0183] The segmentation module is used to segment the meteorological feature data according to the multiple category groups.

[0184] Each module in the aforementioned power generation prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0185] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 19 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data such as meteorological characteristics and power generation. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a method for predicting power generation.

[0186] Those skilled in the art will understand that Figure 19 The structures shown are merely block diagrams of some structures related to the embodiments of this disclosure and do not constitute a limitation on the computer devices on which the embodiments of this disclosure are applied. Specific computer devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0187] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0188] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0189] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0190] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0191] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this disclosure may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this disclosure may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0192] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0193] The above-described embodiments are merely illustrative of several implementation methods of the present disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent for the embodiments of the present disclosure. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of the present disclosure, and these all fall within the protection scope of the embodiments of the present disclosure. Therefore, the protection scope of the embodiments of the present disclosure should be determined by the appended claims.

Claims

1. A new energy SVM power prediction method based on feature optimization of massive meteorological data, characterized in that, The method includes: Meteorological feature data with a correlation greater than a preset boundary value is acquired; wherein, the preset boundary value is determined by: acquiring a correlation value range; taking values ​​from the value range at preset step intervals to obtain multiple candidate boundary values; training new energy power generation prediction models using meteorological feature data with a correlation greater than the candidate boundary values ​​to obtain multiple candidate new energy power generation prediction models; and selecting the boundary value corresponding to the model with the highest accuracy score from the multiple candidate new energy power generation prediction models as the preset boundary value. The meteorological feature data is input into a new energy power generation prediction model, which outputs the predicted power generation. The new energy power generation prediction model is trained based on the correspondence between sample meteorological feature data and power generation. The new energy power generation prediction model has hyperparameters, which are determined by: acquiring a set of sample meteorological feature data, a first value range for a first hyperparameter, and a second value range for a second hyperparameter. The set includes multiple sample meteorological feature data labeled with power. Multiple power prediction models are constructed, with different first and / or second hyperparameters in each model. The accuracy of the multiple power prediction models is tested using cross-validation to obtain multiple... The prediction accuracy score is calculated; the first hyperparameter and second hyperparameter of the power prediction model with the highest prediction accuracy score are determined from the multiple power prediction models as the hyperparameters of the new energy power generation prediction model, which includes an SVM model; the method for determining the first value range includes: constructing multiple first power prediction models, wherein the first hyperparameters of the multiple first power prediction models are multiple parameter values ​​within a third value range, and the second hyperparameter is a preset second hyperparameter; the accuracy of the multiple first power prediction models is tested using cross-validation to obtain multiple prediction accuracy scores; and the preset range of the first hyperparameter of the first power prediction model with the highest prediction accuracy score is determined from the multiple first power prediction models as the first value range.

2. The method according to claim 1, characterized in that, The method for determining the second value range includes: Multiple second power prediction models are constructed. The first hyperparameter of the multiple second power prediction models is the first hyperparameter of the first power prediction model with the highest prediction accuracy, and the second hyperparameter is multiple parameter values ​​within the fourth value range. The accuracy of the multiple second power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores; The preset range of the second hyperparameter of the second power prediction model with the highest prediction accuracy score among the multiple second power prediction models is determined as the second value range.

3. The method according to any one of claims 1 to 2, characterized in that, The cross-validation methods include: Divide the set into multiple subsets; By sequentially determining one subset of the plurality of subsets as the test set and the remaining subsets as the training set, multiple sets are obtained; The training set from the set group is input into the power prediction model for training to obtain the intermediate process model; The test set in the set is input into the intermediate process model to obtain the prediction power corresponding to the test set in the set, and the prediction accuracy corresponding to the set is calculated. The average of the multiple prediction accuracies corresponding to the multiple sets is determined as the prediction accuracy score of the power prediction model.

4. The method according to claim 1, characterized in that, The method for determining the kernel function of the new energy power generation prediction model includes: Multiple candidate power prediction models are obtained, wherein the kernel functions in the multiple candidate power prediction models are of different types; The accuracy of the multiple candidate power prediction models was tested using cross-validation to obtain multiple prediction accuracy scores. The kernel function of the candidate power prediction model with the highest prediction accuracy score among the multiple candidate power prediction models is selected as the kernel function of the new energy power generation prediction model.

5. The method according to claim 1, characterized in that, The new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological characteristic data and power generation, including: A set of sample meteorological feature data is obtained, the set including multiple sample meteorological feature data labeled with power. An initial power prediction model is constructed, wherein training parameters are set in the initial power prediction model; The sample meteorological characteristic data is input into the initial power prediction model to obtain the output result; Based on the difference between the output result and the labeled power, the initial power prediction model is iteratively adjusted until the difference meets the preset requirements, thus obtaining the new energy power generation prediction model.

6. The method according to claim 1, characterized in that, After acquiring meteorological characteristic data whose correlation with power generation is greater than a preset boundary value, the process further includes: Meteorological feature data are divided into multiple category groups by unsupervised clustering, and the correlation between the meteorological feature data and power generation is greater than a preset boundary value. The meteorological feature data is divided according to the multiple category groups.

7. A new energy SVM power prediction device based on feature optimization of massive meteorological data, characterized in that, The device includes: The acquisition module is used to acquire meteorological feature data whose correlation with power generation is greater than a preset boundary value; wherein, the preset boundary value is determined by: acquiring a correlation value range; taking values ​​from the value range at preset step intervals to obtain multiple candidate boundary values; training new energy power generation prediction models using meteorological feature data whose correlation with power generation is greater than the candidate boundary values ​​to obtain multiple candidate new energy power generation prediction models; and selecting the boundary value corresponding to the model with the highest accuracy score from the multiple candidate new energy power generation prediction models as the preset boundary value; The prediction module is used to input the meteorological characteristic data into the new energy power generation prediction model, and output the predicted power generation through the new energy power generation prediction model. The new energy power generation prediction model is set to be trained based on the correspondence between sample meteorological characteristic data and power generation. The new energy power generation prediction model has corresponding hyperparameters. The hyperparameter determination module includes: an acquisition module for acquiring a set of sample meteorological feature data, a first value range of a first hyperparameter, and a second value range of a second hyperparameter, wherein the set includes multiple sample meteorological feature data labeled with power; a construction module for constructing multiple power prediction models, wherein the first hyperparameter and / or the second hyperparameter are different in the multiple power prediction models; a testing module for testing the accuracy of the multiple power prediction models using cross-validation to obtain multiple prediction accuracy scores; and a determination module for determining the first hyperparameter and second hyperparameter of the power prediction model with the highest prediction accuracy score from the multiple power prediction models as the hyperparameters of the new energy power generation prediction model, wherein the new energy power generation prediction model includes an SVM model. The module for determining the first value range includes: a construction module for constructing multiple first power prediction models, wherein the first hyperparameters corresponding to the multiple first power prediction models are multiple parameter values ​​within a third value range, and the second hyperparameters are preset second hyperparameters; a testing module for testing the accuracy of the multiple first power prediction models using cross-validation to obtain multiple prediction accuracy scores; and a determination module for determining the preset range of the first hyperparameters of the first power prediction model with the highest prediction accuracy score from the multiple first power prediction models as the first value range.

8. The apparatus according to claim 7, characterized in that, The module for determining the second value range includes: The construction module is used to construct multiple second power prediction models. The first hyperparameter of the multiple second power prediction models is the first hyperparameter of the first power prediction model with the highest prediction accuracy, and the second hyperparameter is multiple parameter values ​​within the fourth value range. The testing module is used to test the accuracy of the multiple second power prediction models using cross-validation to obtain multiple prediction accuracy scores. The determining module is used to determine, from the plurality of second power prediction models, the preset range of the second hyperparameter of the second power prediction model with the highest prediction accuracy score as the second value range.

9. The apparatus according to any one of claims 7 to 8, characterized in that, The cross-validation module includes: A partitioning module is used to divide the set into multiple subsets; The first determining module is used to sequentially determine one subset of the plurality of subsets as the test set and the remaining subsets as the training set, thereby obtaining a plurality of set groups; The first input module is used to input the training set from the set group into the power prediction model for training, so as to obtain the intermediate process model. The second input module is used to input the test set in the set group into the intermediate process model, obtain the prediction power corresponding to the test set in the set group, and calculate the prediction accuracy score corresponding to the set group. The second determining module is used to determine the average value of the multiple prediction accuracies corresponding to the multiple set groups as the prediction accuracy score of the power prediction model.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

Images