New energy power prediction method based on improved wavelet transform and neural network

Abstract

The present application relates to a new energy power prediction method based on improved wavelet transform and neural network. The method comprises: obtaining multiple groups of meteorological data samples and historical power generation, determining the correlation between meteorological numerical samples and historical power generation under each variable type, obtaining the initial variable type corresponding to each correlation threshold, determining the target threshold and target variable type according to the first training accuracy corresponding to the initial variable type, clustering multiple groups of meteorological data samples, obtaining the first target category to which each group of meteorological data samples belongs, performing wavelet decomposition on the historical power generation, training multiple initial power prediction models using target meteorological samples and power signal samples under each signal frequency, and determining the power prediction model corresponding to each first target category. The present method can traverse multiple correlation thresholds, integrate and learn multiple initial power prediction models, and thus improve the accuracy of the power prediction model.

Description

Technical Field

[0001] The present application relates to the field of renewable energy power generation technology, and in particular to a renewable energy power prediction method, device, computer equipment, storage medium and computer program product based on improved wavelet transform and neural network. Background Art

[0002] Renewable energy power generation technology is a technology that converts new energy such as solar photovoltaic, wind energy, biogas, geothermal or tidal energy into electrical energy. However, a large number of studies have found that the use of new energy such as wind energy or solar photovoltaic for power generation is easily affected by meteorological factors, resulting in random and fluctuating changes in the power generation of new energy. Therefore, in order to ensure the energy security supply of the new energy power system, it is urgent to carry out high-precision prediction of the power generation of new energy.

[0003] In traditional technology, historical meteorological data and historical power data can be obtained and used to train the neural network model. Real-time meteorological data is input into the trained neural network model to generate predicted power data of renewable energy power generation in the future. However, with the further improvement of the prediction accuracy of renewable energy power, a more accurate prediction method for renewable energy power is urgently needed. Summary of the invention

[0004] Based on this, it is necessary to provide a new energy power prediction method, device, computer equipment, computer-readable storage medium and computer program product with high accuracy based on improved wavelet transform and neural network to address the above technical problems.

[0005] In a first aspect, the present application provides a new energy power prediction method based on improved wavelet transform and neural network.

[0006] The method comprises:

[0007] Acquire multiple groups of meteorological data samples and historical power generation corresponding to each group of meteorological data samples, wherein each group of meteorological data samples includes meteorological numerical samples under multiple variable types;

[0008] Determine the correlation between a plurality of meteorological value samples under each variable type and a plurality of historical power generation powers in a plurality of groups of meteorological data samples;

[0009] According to the correlation and multiple correlation thresholds, an initial variable type corresponding to each of the correlation thresholds is determined from the multiple variable types; according to a first training accuracy corresponding to the initial variable type, a target threshold and a target variable type corresponding to the target threshold are determined from the multiple correlation thresholds, wherein the first training accuracy is obtained by training an initial power prediction model using meteorological numerical samples under the initial variable type;

[0010] Performing clustering processing on the multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs;

[0011] For each of the meteorological data samples under the first target category, the following processing is performed:

[0012] Performing wavelet decomposition processing on the historical power generation corresponding to each group of meteorological data samples to obtain power signal samples at multiple signal frequencies;

[0013] The target meteorological samples under the target variable type in each group of the meteorological data samples and the power signal samples under each signal frequency are used to train multiple initial power prediction models, and from the multiple trained initial power prediction models, a power prediction model corresponding to the meteorological data samples under each of the first target categories is determined, and the power prediction model is subsequently used to generate the power generation corresponding to the meteorological data under the first target category.

[0014] In one embodiment, the target meteorological samples under the target variable type in each group of the meteorological data samples and the power signal samples under each signal frequency are used to train multiple initial power prediction models, and a power prediction model corresponding to the meteorological data samples under each first target category is determined from the multiple trained initial power prediction models, including:

[0015] Using the target meteorological samples in each group of meteorological data samples and the power signal samples at each signal frequency to train each of the initial power prediction models, to obtain a target power signal corresponding to each of the signal frequencies generated by each of the initial power prediction models;

[0016] Generating a second training accuracy corresponding to each of the initial power prediction models according to the target power signal;

[0017] According to the second training accuracy, the power prediction model corresponding to the meteorological data sample under each first target category is determined from the multiple trained initial power prediction models.

[0018] In one embodiment, determining a target threshold value from a plurality of correlation threshold values ​​according to the first training accuracy corresponding to the initial variable type, and a target variable type corresponding to the target threshold value, comprises:

[0019] The initial power prediction model is trained using meteorological numerical samples under the initial variable type corresponding to each of the correlation thresholds to obtain a first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under the initial variable type corresponding to each of the correlation thresholds;

[0020] According to the first training accuracy, the target threshold is determined from the multiple correlation thresholds, and the initial variable type corresponding to the target threshold is used as the target variable type.

[0021] In one embodiment, clustering the multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs includes:

[0022] Determining initial clustering features corresponding to each group of meteorological data samples;

[0023] Normalizing the initial clustering features to obtain target clustering features corresponding to each group of meteorological data samples;

[0024] The first target category to which each group of meteorological data samples belongs is determined according to the target clustering features and the category features corresponding to each category.

[0025] In one embodiment, determining the first target category to which each group of meteorological data samples belongs according to the target clustering feature and the category feature corresponding to each category includes:

[0026] Using multiple initial clustering numbers to perform clustering processing on multiple groups of meteorological data samples respectively, to obtain initial categories of the meteorological data samples under each of the initial clustering numbers;

[0027] Generating an evaluation parameter corresponding to the initial number of clusters according to the difference of target clustering characteristics between two meteorological data samples;

[0028] According to the evaluation parameters, a target cluster number is determined from the multiple initial cluster numbers, and the initial category corresponding to the meteorological data sample under the target cluster number is used as the first target category.

[0029] In one of the embodiments, during the application phase of the power prediction model, the method further includes:

[0030] Acquiring meteorological data to be predicted, wherein the meteorological data to be predicted includes meteorological values ​​under multiple variable types;

[0031] Performing clustering processing on the meteorological data to be predicted and the multiple groups of meteorological data samples to determine the second target category to which the meteorological data to be predicted belongs;

[0032] From the meteorological values ​​under the multiple variable types, determine the meteorological value under the target variable type corresponding to the target threshold, input the meteorological value under the target variable type into the power prediction model corresponding to the second target category to obtain the generated power.

[0033] In a second aspect, the present application also provides a new energy power prediction device based on improved wavelet transform and neural network. The device comprises:

[0034] A data sample acquisition module, used to acquire multiple groups of meteorological data samples and historical power generation corresponding to each group of meteorological data samples, each group of meteorological data samples including meteorological numerical samples under multiple variable types;

[0035] A correlation determination module, used to determine the correlation between a plurality of meteorological value samples under each of the variable types and a plurality of the historical power generation powers in a plurality of groups of the meteorological data samples;

[0036] a variable type determination module, configured to determine, according to the correlation and multiple correlation thresholds, an initial variable type corresponding to each of the correlation thresholds from a plurality of the variable types, and to determine, according to a first training accuracy corresponding to the initial variable type, a target threshold from the multiple correlation thresholds, and a target variable type corresponding to the target threshold, wherein the first training accuracy is obtained by training an initial power prediction model using meteorological numerical samples under the initial variable type;

[0037] A first category determination module, used for performing clustering processing on the multiple groups of meteorological data samples to obtain a first target category to which each group of meteorological data samples belongs;

[0038] A wavelet decomposition module is used to perform the following processing for each meteorological data sample under the first target category: performing wavelet decomposition processing on the historical power generation corresponding to each group of meteorological data samples to obtain power signal samples under multiple signal frequencies;

[0039] A model training module is used to train multiple initial power prediction models using the target meteorological samples under the target variable type in each group of the meteorological data samples and the power signal samples under each signal frequency, and to determine the power prediction model corresponding to the meteorological data samples under each of the first target categories from the trained multiple initial power prediction models. The power prediction model is subsequently used to generate the power generation corresponding to the meteorological data under the first target category.

[0040] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the new energy power prediction method based on improved wavelet transform and neural network as described in any embodiment of the first aspect above when executing the computer program.

[0041] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the new energy power prediction method based on improved wavelet transform and neural network as described in any one of the embodiments of the first aspect above.

[0042] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the new energy power prediction method based on improved wavelet transform and neural network described in any embodiment of the first aspect above is implemented.

[0043] The above-mentioned new energy power prediction method, device, computer equipment, storage medium and computer program product based on improved wavelet transform and neural network obtain multiple groups of meteorological data samples, and the historical power generation corresponding to each group of meteorological data samples, determine the correlation between multiple meteorological numerical samples under each variable type and multiple historical power generation in the multiple groups of meteorological data samples, and determine the initial variable type corresponding to each correlation threshold from the multiple variable types according to the correlation and multiple correlation thresholds, and determine the target threshold from the multiple correlation thresholds according to the first training accuracy corresponding to the initial variable type, and the target variable type corresponding to the target threshold, perform clustering processing on the multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs, and for each meteorological data sample under the first target category , perform the following processing: perform wavelet decomposition processing on the historical power generation corresponding to each group of meteorological data samples to obtain power signal samples under multiple signal frequencies; use the target meteorological samples under the target variable type in each group of meteorological data samples and the power signal samples under each signal frequency to train multiple initial power prediction models, and determine the power prediction model corresponding to the meteorological data samples under each first target category from the trained multiple initial power prediction models, and be able to traverse multiple correlation thresholds to obtain the initial variable type corresponding to each correlation threshold, and use the meteorological numerical samples under the initial variable type to train the initial power detection model, and determine the target threshold from multiple correlation thresholds according to the first training accuracy obtained by training, so as to improve the accuracy of the model input data, thereby improving the accuracy of the power prediction model. At the same time, using the new energy power prediction method based on improved wavelet transform and neural network provided by the present application, it is possible to use the target meteorological samples under the target variable type to train multiple initial power prediction models for each meteorological data sample under the first target category, and integrate the multiple initial power prediction models, and use the initial power prediction model with the highest prediction accuracy as the power prediction model to improve the prediction accuracy of the power prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of a new energy power prediction method based on improved wavelet transform and neural network in one embodiment;

[0045] Figure 2 A schematic diagram of a flow chart of a determination step of a power prediction model in one embodiment;

[0046] Figure 3 A schematic diagram of a process of generating clustering features in one embodiment;

[0047] Figure 4 A schematic diagram of a flow chart of a first target category determination step in one embodiment;

[0048] Figure 4a A schematic diagram of a first clustering evaluation parameter determination step in one embodiment;

[0049] Figure 4b A schematic diagram of a second cluster evaluation parameter determination step in one embodiment;

[0050] Figure 5 A schematic diagram of a flow chart of a power generation step in one embodiment;

[0051] Figure 6 It is a flow chart of a new energy power prediction method based on improved wavelet transform and neural network in another embodiment;

[0052] Figure 7 It is a structural block diagram of a new energy power prediction device based on improved wavelet transform and neural network in one embodiment;

[0053] Figure 8 is an internal structure diagram of a computer device in one embodiment;

[0054] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] In one embodiment, Figure 1 As shown, a new energy power prediction method based on improved wavelet transform and neural network is provided, and the method is applied to a computer device, which can be a server or a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be but not limited to various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices, and the Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented as an independent server or a server cluster consisting of multiple servers.

[0057] In this embodiment, the method includes the following steps:

[0058] Step S102, obtaining multiple groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples.

[0059] The meteorological data samples may be used to characterize the historical meteorological data collected in the target collection area. In one example, a set of meteorological data samples may include meteorological value samples obtained at multiple collection times within a collection period. The collection period may be, but is not limited to, any one of a time period such as a day, a week, or a month.

[0060] Each group of meteorological data samples may include meteorological numerical samples under multiple variable types. The variable type of the meteorological numerical samples may include, but is not limited to, any one of the types including wind speed type, irradiance type, temperature type, humidity type, and air pressure type. The meteorological numerical samples may include, but are not limited to, any one or more of the data samples including wind speed samples under the wind speed type, irradiance samples under the irradiance type, temperature samples under the temperature type, humidity samples under the humidity type, and air pressure samples under the air pressure type.

[0061] The historical power generation can be used to characterize the new energy power generation that has been collected in the target area. In one example, the historical power generation can be obtained by predicting the wind power generation of a wind farm. In another example, the historical power generation can be obtained by predicting the solar power generation of a photovoltaic power station. The historical power generation has a corresponding relationship with the collection time of the meteorological numerical sample. For example, for the target area, multiple meteorological numerical samples obtained at a collection time correspond to one historical power generation.

[0062] Specifically, the computer device obtains multiple groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples in response to the training request of the power prediction model. The training request of the power prediction model can be manually triggered by the user, for example, the user clicks the corresponding model training button on the interface to trigger the training request of the power prediction model; it can also be automatically triggered by the computer device, for example, when the computer device detects the existence of meteorological data samples, it automatically triggers the training request of the power prediction model. In an example, a group of meteorological data samples may include the first temperature sample, the first humidity sample, the first air pressure sample and the first irradiance sample collected at time T1 in the same day, the second temperature sample, the second humidity sample, the second air pressure sample and the second irradiance sample collected at time T2, and the third temperature sample, the third humidity sample, the third air pressure sample and the third irradiance sample collected at time T3. The power generation corresponding to the group of meteorological data samples may include the first historical power generation corresponding to time T1, the second historical power generation corresponding to time T2 and the third historical power generation corresponding to time T3.

[0063] Step S104, determining the correlation between a plurality of meteorological value samples under each variable type and a plurality of historical power generation powers in a plurality of groups of meteorological data samples.

[0064] Specifically, the computer device obtains multiple meteorological numerical samples corresponding to each variable type in multiple groups of meteorological data samples. The multiple meteorological numerical samples under each variable type and the historical power generation corresponding to the multiple meteorological numerical samples are processed by calculation to obtain the correlation between the meteorological numerical samples under each variable type and the historical power generation. In one example, the computer device obtains two groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples. Each group of meteorological data samples includes wind speed samples under the wind speed type and temperature samples under the temperature type. The wind speed samples in the two groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples are processed by calculation to obtain the first correlation between the wind speed samples under the wind speed type and the historical power generation. The temperature samples in the two groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples are processed by calculation to obtain the second correlation between the temperature samples under the temperature type and the historical power generation.

[0065] Step S106, determining an initial variable type corresponding to each correlation threshold from multiple variable types according to the correlation and multiple correlation thresholds, and determining a target threshold from multiple correlation thresholds according to a first training accuracy corresponding to the initial variable type.

[0066] The first training accuracy may be used to characterize the prediction accuracy of the initial power prediction model obtained after the initial power prediction model is trained using meteorological numerical samples under the initial variable type in multiple groups of meteorological data samples as input data.

[0067] Specifically, a plurality of correlation thresholds and threshold determination logics are pre-stored in the computer device. Each correlation threshold is used to judge the correlation corresponding to the meteorological numerical samples under multiple variable types, and the variable type with a correlation greater than the correlation threshold is used as the initial variable type corresponding to the correlation threshold, so as to obtain the initial variable type corresponding to each correlation threshold. For the initial variable type corresponding to each correlation threshold, the following operations are performed: the meteorological numerical samples under the initial variable type corresponding to the correlation threshold in multiple groups of meteorological data samples are used as input data to train the initial power prediction model, so as to obtain the first training accuracy corresponding to the meteorological numerical samples under the initial variable type corresponding to the correlation threshold. The correlation threshold whose first training accuracy meets the threshold determination logic is used as the target threshold, and the initial variable type corresponding to the target threshold is used as the target variable type. Among them, the threshold determination logic can be to compare the first training accuracy with the accuracy threshold, and to use the correlation threshold whose first training accuracy is greater than the accuracy threshold as the target threshold; or to compare the first training accuracy corresponding to multiple correlation thresholds, and to use the correlation threshold corresponding to the largest first training accuracy as the target threshold. In one example, the correlation between the humidity samples under the humidity type obtained by the computer device and the power generation is 0.5, the correlation between the temperature samples under the temperature type and the power generation is 0.7, and the correlation between the irradiance samples under the irradiance type and the power generation is 0.9. The irradiance type with the highest correlation is used as the target variable type. Subsequently, the irradiance samples under the irradiance type in each group of meteorological data samples are used as training input data to train the untrained initial power prediction model.

[0068] Step S108, clustering the multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs.

[0069] Specifically, clustering logic is pre-stored in the computer device. Clustering logic is used to perform clustering processing on multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs. Among them, the clustering logic can be an unsupervised clustering logic based on similar days, such as randomly selecting several groups of meteorological data samples as cluster centers, obtaining the distance between the remaining groups of meteorological data samples and each cluster center, and treating a group of meteorological data samples with a distance less than a distance threshold and the cluster center as data of the same category; or it can be treating several groups of meteorological data samples whose values ​​of meteorological data samples are higher than the first category threshold as data of the same category, and treating several groups of meteorological data samples whose values ​​of meteorological data samples are lower than the second category threshold as data of another category. In one example, when the meteorological data samples obtained by the computer device are wind speed samples, the clustering logic can adopt the unsupervised clustering logic of wind speed similar days. When the meteorological data samples obtained by the computer device are irradiance samples, the clustering logic can adopt the unsupervised clustering logic of irradiance similar days.

[0070] Step S110 , training multiple initial power prediction models using target meteorological samples and power signal samples at each signal frequency to obtain a power prediction model corresponding to each first target category.

[0071] The target meteorological sample can be used to characterize the meteorological numerical sample under the target variable type in each group of meteorological data samples under the first target category. In one example, the target variable type determined by the computer device is the wind speed type, and the meteorological numerical sample under the wind speed type in each group of meteorological data samples is used as the target meteorological sample.

[0072] The initial power prediction model may include but is not limited to at least one of a BP neural network model (Back Propagation), an LSTM model (Long Short-Term Memory), a BLSTM model (Bi-directional Long Short-Term Memory), an SVM model (Support Vector Machine), a GRU model (Gate Recurrent Unit), an MLP model (Multilayer Perceptron), an RF model (Random Forest), and the like.

[0073] Specifically, a variety of untrained initial power prediction models are pre-deployed in the computer device. For each meteorological data sample under the first target category, the following operations are performed: the power generation corresponding to each group of meteorological data samples is subjected to wavelet decomposition processing to obtain power signal samples under multiple signal frequencies. The result of each layer of decomposition is that the low-frequency signal obtained by the last decomposition is decomposed into two parts, low-frequency and high-frequency. In an example, after the computer device performs N layers of wavelet decomposition on the power generation, the power generation X is decomposed into: X = D1 + D2 + ... + DN + AN, where D1, D2, ..., DN are high-frequency signals obtained by decomposition of the first layer, the second layer to the same N layers, respectively, and AN is the low-frequency signal obtained by decomposition of the Nth layer.

[0074] The target meteorological samples under the target variable type in each group of meteorological data samples under the target category are used as training input data, and the power small signal samples under each signal frequency corresponding to each group of meteorological data samples are used as training labels. The computer device inputs the target meteorological samples into each untrained initial power prediction model to obtain the predicted power generation generated by each trained initial power prediction model. The loss function is used to calculate the loss value between the power generation and the predicted power generation generated by each initial power prediction model. The model parameters of the initial power model are adjusted in the direction where the loss value becomes smaller until the training stop condition is met, and a plurality of initial power prediction models corresponding to the meteorological data samples under the target category are obtained. According to the prediction accuracy of each initial power prediction model, the power prediction model corresponding to the meteorological data samples under the target category is determined from the multiple trained initial power prediction models. Among them, the training stop condition can be any one of the following: the number of training iterations reaches a preset number, or the loss value reaches a preset value.

[0075] In one example, the initial power prediction model deployed in the computer device includes a BP neural network model, an LSTM model, and an SVM model. After clustering processing, multiple groups of meteorological data samples a under category A and multiple groups of meteorological data samples b under category B are obtained. For multiple groups of meteorological data samples a under category A, the following operations are performed: the target meteorological samples under the target variable type in the multiple groups of meteorological data samples a are respectively input into the BP neural network model, the LSTM model, and the SVM model to obtain the first predicted power generation generated by the BP neural network model, the second predicted power generation generated by the LSTM model, and the third predicted power generation generated by the SVM model. The loss function in the BP neural network model is used to calculate the loss value between the first predicted power generation and the power generation corresponding to the target meteorological sample, and the first model parameter of the BP neural network model is adjusted in the direction of decreasing the loss value until the training stop condition is met, and the first prediction accuracy corresponding to the trained BP neural network model is obtained to be 0.8. Repeat the above model training operation to obtain the second prediction accuracy corresponding to the trained LSTM model to be 0.6, and the third prediction accuracy corresponding to the trained SVM model to be 0.7. Compare the first prediction accuracy, the second prediction accuracy, and the third prediction accuracy, and use the BP neural network model with the highest prediction accuracy as the power prediction model corresponding to the multiple groups of meteorological data samples a under category A. Repeat the above operation for multiple groups of meteorological data samples b under category B, and input the target meteorological samples under the target variable type in the multiple groups of meteorological data samples b into the BP neural network model, LSTM model, and SVM model respectively, and obtain the fourth predicted power generation of 0.55 corresponding to the BP neural network model trained with the target meteorological samples under the target variable type in the multiple groups of meteorological data samples b, the fifth predicted power generation of 0.7 corresponding to the LSTM model, and the sixth predicted power generation of 0.63 corresponding to the SVM model. Compare the fourth prediction accuracy, the fifth prediction accuracy, and the sixth prediction accuracy, and use the LSTM model with the highest prediction accuracy as the power prediction model corresponding to the multiple groups of meteorological data samples b under category B.

[0076] In the above-mentioned new energy power prediction method based on improved wavelet transform and neural network, by obtaining multiple groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples, the correlation between multiple meteorological numerical samples under each variable type and multiple historical power generation is determined in the multiple groups of meteorological data samples, and the initial variable type corresponding to each of the correlation thresholds is obtained. According to the first training accuracy corresponding to the initial variable type, the target threshold and the target variable type corresponding to the target threshold are determined from the multiple correlation thresholds, and the multiple groups of meteorological data samples are clustered to obtain the target category to which each group of meteorological data samples belongs. For each first Meteorological data samples under the target category, using the target meteorological samples under the target variable type in the meteorological data samples and the power signal samples under each signal frequency to train multiple initial power prediction models, determine the power prediction model corresponding to each first target category from the multiple initial power prediction models after training, traverse multiple correlation thresholds, obtain the initial variable type corresponding to each correlation threshold, use the meteorological numerical samples under the initial variable type to train the initial power detection model, and determine the target threshold from multiple correlation thresholds according to the first training accuracy obtained by training to improve the accuracy of the model input data, thereby improving the accuracy of the power prediction model. At the same time, using the new energy power prediction method based on improved wavelet transform and neural network provided in this embodiment, it is possible to use the target meteorological samples under the target variable type for each meteorological data sample under the target category to train multiple initial power prediction models, integrate and learn multiple initial power prediction models, and use the initial power prediction model with the highest prediction accuracy as the power prediction model to improve the prediction accuracy of the power prediction model.

[0077] In one embodiment, Figure 2 As shown, step S110, using target meteorological samples and power signal samples at each signal frequency to train multiple initial power prediction models, to obtain a power prediction model corresponding to each first target category, including:

[0078] Step S202: training each initial power prediction model using target meteorological samples and power signal samples to obtain a target power signal generated by each initial power prediction model.

[0079] Step S204: generating a second training accuracy corresponding to each initial power prediction model according to the target power signal.

[0080] Specifically, a plurality of untrained initial power prediction models are pre-deployed in the computer device. For each meteorological data sample under the first target category, the following processing is performed: the target meteorological sample in each group of meteorological data samples is used as the training input data, and the power signal sample at each signal frequency corresponding to each group of meteorological data samples is used as the training label corresponding to the training input data, and each initial power prediction model is trained to obtain the target power signal at each signal frequency output by each initial power prediction model. The target power signal at multiple signal frequencies is subjected to wavelet reconstruction processing to obtain the predicted power generation corresponding to the meteorological numerical sample. The predicted power generation and the power generation corresponding to each group of meteorological data samples are subjected to computational processing to obtain the second training accuracy of each initial power prediction model corresponding to the meteorological data sample under the target category.

[0081] Step S206: Determine, according to the second training accuracy, a power prediction model corresponding to each first target category from the trained multiple initial power prediction models.

[0082] Specifically, the power prediction model conditions are pre-stored in the computer device. For each meteorological data sample under the first target category, the following processing is performed: according to the second training accuracy of each initial power prediction model, the trained initial power prediction model that meets the power prediction model conditions is used as the power prediction model corresponding to the meteorological data sample under the target category. Repeat the above operation to obtain the power prediction model corresponding to the meteorological data sample under each target category. Among them, the power prediction model condition can be to determine that the first training accuracy exceeds the preset model accuracy threshold, or to compare the first training accuracy of each initial power prediction model to determine the highest first training accuracy.

[0083] In this embodiment, by training multiple initial power prediction models, based on the second training accuracy corresponding to the trained initial power prediction model, a power prediction model corresponding to the meteorological data sample under each first target category is determined from the multiple trained initial power prediction models. This can achieve integrated learning of multiple initial power prediction models, avoid the prediction error that exists when using a single initial power prediction model for training, and thereby improve the accuracy of the power prediction model.

[0084] In one embodiment, step S106, based on the correlation and multiple correlation thresholds, determines the initial variable type corresponding to each correlation threshold from multiple variable types, and determines the target threshold from multiple correlation thresholds based on the first training accuracy corresponding to the initial variable type, including: using the meteorological numerical samples under each initial variable type to train the initial power prediction model to obtain the first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under each initial variable type; and determining the target threshold from multiple correlation thresholds based on the first training accuracy.

[0085] Specifically, the computer device performs the following processing on the meteorological numerical samples under the initial variable type corresponding to each correlation threshold: the meteorological numerical samples under the initial variable type in the multiple groups of meteorological data samples under each first target category are used as training input data, the historical power generation corresponding to each group of meteorological data samples is used as a training label, and multiple untrained initial power prediction models are trained to obtain the first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under the initial variable type in the multiple groups of meteorological data samples under each first target category. The first training accuracy corresponding to each correlation threshold is compared, and the correlation threshold with the highest first training accuracy is determined from multiple correlation thresholds, and the correlation threshold with the highest first training accuracy is used as the target threshold corresponding to the meteorological data sample under the target category, and the initial variable type corresponding to the target threshold is used as the target variable type corresponding to the meteorological data sample under the target category. The specific method for generating the first training accuracy can be implemented by referring to the generation step of the second training accuracy provided in the above embodiment, and the specific method for training the initial power prediction model can be implemented by referring to the training step of the initial power prediction model provided in the above embodiment, which will not be specifically described here.

[0086] In this embodiment, by setting multiple correlation thresholds, the initial variable type corresponding to each correlation threshold is determined from multiple variable types, and the initial power model is trained using meteorological numerical samples under each initial variable type. According to the first training accuracy of the model, the target variable type corresponding to the meteorological data samples under each first target category is determined from multiple initial variable types, which can further improve the correlation between the training input data and the target category, so as to improve the accuracy of the power prediction model.

[0087] In one embodiment, Figure 3 As shown, step S108, clustering is performed on multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs, including:

[0088] Step S302: determining initial clustering features corresponding to each group of meteorological data samples.

[0089] Specifically, the computer device may use the following formula to process the meteorological numerical samples under multiple variable types in each group of meteorological data samples to generate initial clustering features corresponding to each group of meteorological data samples:

[0090] a=(a s1 ,a s2 ,…,a sH ,a smax ,a smin ,a smean ,a sstd )

[0091] Among them, a is the initial clustering feature, a s1 ,a s2 ,…,a sH is the meteorological numerical sample obtained at each collection time in each set of meteorological data samples, a smax is the largest meteorological numerical sample in each set of meteorological data samples, a smin is the smallest meteorological numerical sample in each set of meteorological data samples, a smean is the mean value of meteorological numerical samples in each group of meteorological data samples, a sstd is the standard deviation of the meteorological numerical samples in each group of meteorological data samples. In one example, wind speed samples can be used to generate initial clustering features.

[0092] In another example, irradiance samples can be used to generate initial clustering features. Unsupervised clustering logic for similar irradiance days In order to divide days with similar irradiance characteristics throughout the year into several categories, it is necessary to select physical quantities that can reflect the daily characteristics of irradiance, such as the trend of irradiance changes during the day (daily average irradiance samples), amplitude (i.e., irradiance value samples at each collection time of the day corresponding to each group of irradiance samples, daily maximum irradiance samples, daily minimum irradiance samples), volatility (daily irradiance standard deviation samples), etc. The selected physical quantities are processed using the above formula to obtain the initial clustering features corresponding to each group of irradiance samples. The initial clustering feature construction takes into account both the daily trend of irradiance and the magnitude of irradiance, which can improve the accuracy of subsequent clustering of meteorological data samples.

[0093] Step S304: normalize the initial cluster features to obtain target cluster features corresponding to each group of meteorological data samples.

[0094] Step S306, determining the first target category to which each group of meteorological data samples belongs according to the target clustering feature and the category feature corresponding to each category.

[0095] Among them, the category features can be used to characterize the target clustering features of the cluster centers corresponding to multiple groups of meteorological data samples in the category.

[0096] Specifically, since the initial clustering features include meteorological numerical samples under multiple variable types, the initial clustering features need to be normalized to eliminate the impact of dimensional differences between different physical quantities on the clustering results. The computer device can use the maximum and minimum normalization method to normalize the initial clustering features to obtain the target clustering features corresponding to each group of meteorological data samples. The target clustering features corresponding to each group of meteorological data samples and the category features corresponding to each category are processed to obtain the category difference between the target clustering features and each category feature. The category with a category difference less than the category threshold is taken as the first target category to which the group of meteorological data samples belongs.

[0097] In an example, the initial clustering features can be normalized according to the following formula to obtain the target clustering features:

[0098]

[0099] Among them, b ji ' is the i-th sample value of the j-th physical quantity in the target cluster feature, b ji is the i-th sample value of the j-th physical quantity in the initial clustering feature; b jmin is the minimum value of the jth physical quantity in the initial clustering feature; b jmax is the maximum value of the jth physical quantity in the initial clustering features.

[0100] In this embodiment, by determining the initial clustering features corresponding to each group of meteorological data samples, normalizing the initial clustering features, obtaining the target clustering features corresponding to each group of meteorological data samples, and determining the target category to which each group of meteorological data samples belongs based on the target clustering features, the accuracy of clustering can be improved, thereby improving the accuracy of the power prediction model corresponding to meteorological data samples in different categories.

[0101] In one embodiment, Figure 4 As shown, step S306, determining the first target category to which each group of meteorological data samples belongs according to the target clustering feature and the category feature corresponding to each category, includes:

[0102] Step S402, obtaining initial categories of meteorological data samples under multiple initial clustering quantities.

[0103] Specifically, a plurality of initial clustering numbers are stored in the computer device, and each initial clustering number is used to perform clustering processing on the normalized meteorological data samples. An initial clustering center corresponding to each initial clustering result is obtained, and a calculation process is performed on the target clustering features of the initial clustering center and multiple groups of meteorological data samples other than the initial clustering center, and meteorological data samples belonging to the same category as the initial clustering center are determined, and the initial category of the meteorological data samples corresponding to each initial clustering number is obtained.

[0104] Step S404: generating evaluation parameters corresponding to each initial cluster number according to the difference of target cluster characteristics between any two meteorological data samples.

[0105] Step S406: According to the evaluation parameters, a target number of clusters is determined from a plurality of initial numbers of clusters, and the initial category corresponding to the meteorological data sample under the target number of clusters is used as the first target category.

[0106] Specifically, the computer device performs computational processing on the target clustering features of multiple groups of meteorological data samples under the same category in the initial clustering results and the target clustering features corresponding to the initial clustering center to obtain a first clustering evaluation parameter corresponding to each initial clustering number. The target clustering features of multiple groups of meteorological data samples belonging to the same category in the initial clustering results are computed to determine the intra-cluster distance corresponding to the initial clustering results. The target clustering features of multiple groups of meteorological data samples belonging to different categories in the initial clustering results are computed to determine the inter-cluster distance corresponding to the initial clustering results. The intra-cluster distance and the inter-cluster distance are computed to determine the second clustering evaluation parameter corresponding to each initial clustering number. According to the first clustering evaluation parameter and the second clustering evaluation parameter, the target clustering number is determined from the multiple initial clustering numbers. The clustering result corresponding to the target clustering number is obtained to obtain the first target category to which each group of meteorological data samples belongs.

[0107] In one example, the computer device can randomly select several groups of meteorological data samples corresponding to the initial number of clusters as the initial cluster centers, and use the target clustering features of the initial cluster centers as the center clustering features. The center clustering features of each initial cluster center and the target clustering features of each group of meteorological data samples are processed and calculated to determine several groups of meteorological data samples whose distance from the initial distance center is less than the clustering threshold, obtain meteorological data samples belonging to the same category as each initial cluster center, and obtain the first clustering result after clustering here. Repeat the above operation to cluster the first clustering result after clustering here again until the cluster center in the first clustering result no longer changes, and obtain the final initial clustering result. In another example, the computer device can also randomly initialize the center point multiple times, and then select a clustering result with the best running result as the initial clustering result corresponding to the initial clustering data.

[0108] In another example, the computer device can use k-means (an unsupervised clustering method) for clustering, and determine the number of clusters according to the Elbow Method. The k-means clustering method takes minimizing the square error between samples and particles as the objective function, and the sum of the square distance errors between the particles of each cluster and the sample points in the cluster is called the degree of distortion, and the degree of distortion is used as the first clustering evaluation parameter. Therefore, for a cluster, the lower its degree of distortion, the closer the members in the cluster are, and the higher the degree of distortion, the looser the structure in the cluster. The degree of distortion will decrease as the number of categories increases, and then the degree of distortion can be used to determine the number of categories in the cluster. However, for data with a certain degree of discrimination, the degree of distortion will be greatly improved when a certain critical point is reached, and then slowly decrease. This critical point can be considered as a point with better clustering performance.

[0109] The degree of distortion corresponding to the initial clustering result can be obtained by referring to the following formula:

[0110]

[0111] Among them, Ci is the i-th category, p is the meteorological data sample of Ci, mi is the cluster center of Ci, SSE can represent the clustering error or distortion degree of all meteorological data samples, and k is the number of clusters.

[0112] In another example, for a clustering task, it is hoped that the clusters obtained are as close as possible within the clusters and as far apart as possible, so the computer device can use the silhouette coefficient as the second clustering evaluation parameter.

[0113] The silhouette coefficient can be obtained by referring to the following formula:

[0114]

[0115] Among them, a(i) = average (the distance from the i vector to all other points in the cluster to which it belongs, that is, the intra-cluster distance corresponding to the i vector), b(i) = min (the average distance from the i vector to all points in a cluster that does not contain it, that is, the inter-cluster distance corresponding to the i vector). S(i) is the silhouette coefficient, which ranges from [-1,1]. The larger the silhouette coefficient, the more reasonable the clustering. That is, if S(i) is close to 1, it means that the clustering of sample i is reasonable; if S(i) is close to -1, it means that sample i should be classified into another cluster; if S(i) is approximately 0, it means that sample i is on the boundary of two clusters. S(i) of all samples is calculated and processed, and the average value of S(i) is taken as the silhouette coefficient of the current clustering result, and the number of clusters corresponding to the larger silhouette coefficient is taken as the target number of clusters.

[0116] In an example, taking the data of a photovoltaic power station in Yunnan as an example, similar day clustering prediction is performed:

[0117] The training set data obtained by the computer equipment include: location: a photovoltaic power station; time: 2020 / 7 / 1 8:00:00-2021 / 8 / 7 7:45:00; 400 days of data, reported from 8:00 every day, 3 days of data each time, only the forecast data of the first day is taken as the training input data; each test time step: 15 minutes; data volume: 115,200 data.

[0118] The test set data obtained by the computer equipment includes: Location: a photovoltaic power station; Time: 2021 / 8 / 5 8:00:00-2021 / 12 / 15 7:45:00; 130 days of data. Reporting starts at 8:00 every day, 3 days of data each time, only the forecast data of the first day is taken as test data; Each test time step: 15 minutes; Data volume: 12480 data.

[0119] The computer device uses the elbow rule to process the number of multiple clusters (i.e., the initial number of clusters) to determine the sum of the squares of the intra-cluster deviations corresponding to the number of each cluster (i.e., the first cluster evaluation parameter), such as Figure 4a As shown. Obtain the silhouette coefficient (i.e., the second clustering evaluation parameter) corresponding to the number of each cluster, as Figure 4b As shown, the points in the circle are the corresponding silhouette coefficients when the number of clusters is 3. According to the sum of the squares of the deviations within the cluster and the silhouette coefficient, the optimal number of clusters is determined to be 3 (i.e., the target number of clusters). The target number of clusters is used to cluster the meteorological data samples collected from a photovoltaic power station in Yunnan for 400 days. The multiple groups of meteorological data samples collected on 125 days are classified into the first category; the multiple groups of meteorological data samples collected on 167 days are classified into the second category; and the multiple groups of meteorological data samples collected on 108 days are classified into the third category.

[0120] In this embodiment, by obtaining the initial categories of meteorological data samples under multiple initial clustering numbers, evaluation parameters corresponding to each initial clustering number are generated according to the difference in target clustering characteristics between any two meteorological data samples. Based on the evaluation parameters, the target clustering number is determined from the multiple initial clustering numbers, and the initial category corresponding to the meteorological data samples under the target clustering number is used as the first target category, which can further improve the accuracy of the first target category.

[0121] In one embodiment, step S104, determining the correlation between multiple meteorological value samples under each variable type and multiple historical power generation in multiple groups of meteorological data samples, includes: obtaining the correlation between multiple meteorological value samples under each variable type and multiple historical power generation by the following formula:

[0122]

[0123] Among them, R is the correlation, k is the collection time corresponding to the meteorological numerical samples under the variable type, n is the number of meteorological numerical samples under the variable type in multiple groups of meteorological data samples, T k is the meteorological numerical sample collected at time k; is the mean of the meteorological numerical samples under the variable type in multiple groups of meteorological data samples, P k is the historical power generation corresponding to the meteorological numerical sample collected at time k; It is the mean value of the historical power generation corresponding to the meteorological numerical samples under the variable type in multiple groups of meteorological data samples.

[0124] In this embodiment, the correlation determination formula provided is different from the traditional correlation determination method. Variable types with higher correlation values ​​(i.e., approaching 1) and lower correlation values ​​(i.e., approaching -1) with historical power generation are regarded as variable types with higher correlation with historical power generation, thereby improving the accuracy of subsequent determination of the target variable type based on the correlation, thereby improving the accuracy of the power prediction model.

[0125] In one embodiment, Figure 5 As shown, in the application stage of the power prediction model, the new energy power prediction method based on the improved wavelet transform and neural network also includes:

[0126] Step S502, obtaining the meteorological data to be predicted.

[0127] Step S504: determine the second target category to which the meteorological data to be predicted belongs.

[0128] Step S506, determining the meteorological value under the target variable type corresponding to the target threshold from the meteorological values ​​under multiple variable types, inputting the meteorological value under the target variable type into the power prediction model corresponding to the second target category to obtain the generated power.

[0129] The meteorological data to be predicted include meteorological values ​​under various variable types.

[0130] Specifically, the computer device obtains the meteorological data to be predicted in response to the prediction request of the generated power. The meteorological data to be predicted and multiple groups of meteorological data samples are clustered to obtain the second target category to which the meteorological data to be predicted belongs. From the meteorological values ​​under multiple variable types, the meteorological value under the target variable type corresponding to the target threshold is determined. The power prediction model corresponding to the second target category to which the meteorological data to be predicted belongs is obtained, and the meteorological value under the target variable type is input into the power prediction model corresponding to the second target category to obtain the generated power.

[0131] In this embodiment, by acquiring the meteorological data to be predicted, determining the second target category to which the meteorological data to be predicted belongs, and inputting the meteorological values ​​under the target variable type into the power prediction model corresponding to the second target category, the accuracy of power generation prediction can be improved.

[0132] In one embodiment, Figure 6 As shown, a new energy power prediction method based on improved wavelet transform and neural network is provided, including:

[0133] Step S602, obtaining multiple groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples, and determining the correlation between multiple meteorological value samples under each variable type in the multiple groups of meteorological data samples and multiple historical power generation.

[0134] Specifically, the computer device obtains multiple groups of meteorological data samples and the power generation corresponding to each group of meteorological data samples. The multiple meteorological numerical samples under each variable type and the power generation corresponding to the multiple meteorological numerical samples are processed to obtain the correlation between the meteorological numerical samples under each variable type and the power generation. The specific correlation calculation method can be implemented with reference to the method provided in the above embodiment, and will not be elaborated here.

[0135] Step S604: determining an initial variable type corresponding to each correlation threshold from multiple variable types according to the correlation and multiple correlation thresholds.

[0136] Step S606: train the initial power prediction model using the meteorological numerical samples under each initial variable type, and determine the target threshold value according to the first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under each initial variable type.

[0137] Specifically, the computer device compares the correlation with each correlation threshold respectively, and uses the variable type greater than the correlation threshold as the initial variable type corresponding to the correlation threshold. The meteorological numerical samples under each initial variable type are used as training input data, and the power generation corresponding to the meteorological numerical samples is used as a training label to train the initial power prediction model. The correlation threshold with the highest first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under each initial variable type is used as the target threshold, and the initial variable type corresponding to the target threshold is used as the target variable type. The specific method for determining the target threshold can be implemented with reference to the method provided in the above embodiment, and will not be elaborated here.

[0138] Step S608, determine the initial clustering features corresponding to each group of meteorological data samples, normalize the initial clustering features to obtain the target clustering features corresponding to each group of meteorological data samples, and determine the first target category to which each group of meteorological data samples belongs based on the target clustering features and the category features corresponding to each category.

[0139] Specifically, the computer device processes the meteorological numerical samples under multiple variable types in each group of meteorological data samples to generate initial clustering features corresponding to each group of meteorological data samples. The initial clustering features are normalized by maximum and minimum normalization to obtain the target clustering features corresponding to each group of meteorological data samples. According to the target clustering features of multiple groups of meteorological data samples and the category features corresponding to each category, the target category to which each group of meteorological data samples belongs is determined. The specific target category determination method can be implemented with reference to the method provided in the above embodiment, which will not be elaborated in detail here.

[0140] Step S610, performing wavelet decomposition processing on the historical power generation to obtain power signal samples at multiple signal frequencies, using target meteorological samples and power signal samples to train multiple initial power prediction models, and determining a power prediction model corresponding to the first target category from the trained multiple initial power prediction models.

[0141] Specifically, the computer device performs the following operations for the meteorological data samples under each target category: the generated power corresponding to each group of meteorological data samples is subjected to wavelet decomposition processing to obtain power signal samples under multiple signal frequencies. The target meteorological samples and the power signal samples under each signal frequency are used to perform the following operations for each initial power prediction model: the target meteorological samples are used as training input data, the power signal samples under each signal frequency are used as training labels, and multiple untrained initial power prediction models are trained to obtain trained initial power prediction models corresponding to each signal frequency. The predicted power signals under each signal frequency output by the multiple trained initial power prediction models are subjected to wavelet reconstruction processing to obtain the first power prediction model corresponding to the initial power prediction model. The second training accuracy of the first power prediction model corresponding to each initial power prediction model is obtained, and the second training accuracy of the multiple first power prediction models is compared to determine the power prediction model. The specific method for determining the power prediction model can be implemented with reference to the method provided in the above embodiment, which will not be elaborated in detail here.

[0142] In an example, taking the short-term power prediction of a photovoltaic power station based on wavelet transform and BP neural network as an example, the steps of generating a power prediction model are explained: obtaining the measured historical power generation data of the photovoltaic power station and the meteorological data samples at the corresponding time; performing wavelet decomposition on the historical power generation data to obtain power signals D1, D2...DN and AN at multiple signal frequencies; using multiple BP neural networks to train D1, D2...DN and AN respectively, wherein the BP neural network training function can be selected. The meteorological numerical samples under the target variable type in the multiple groups of meteorological data samples are divided into k parts without duplication, and k-1 parts are used as training sets, and the other part is used as validation sets. The data in the training set and the corresponding power signals D1, D2...DN and AN at multiple signal frequencies are input into the untrained initial power prediction model for training, so that the trained initial power prediction model outputs the target values ​​of D1, D2...DN and AN. Using the data in the validation set as input data, the predicted power signals D1, D2...DN and AN output by the trained initial power prediction model are obtained. Then, the predicted power signal is reconstructed by wavelet to obtain the prediction result of the historical power generation data. Repeat the above operation with different validation sets to cross-validate the initial power prediction model, that is, use the training set to train the initial power prediction model, use the test set to test the trained initial power prediction model, and save the evaluation index of the initial power prediction model. Repeat the above operation k times until each meteorological numerical sample is used as a test set to test the trained initial power prediction model. Calculate the average value of the k groups of test indicators as an estimate of the model accuracy and as the performance index of the model under the current k-fold cross validation. Calculate the performance index under different parameters for k values, and select the parameter with the highest performance index as the current model parameter. Under normal circumstances, the number of subsets k of the training set is generally 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 models trained will also increase; when the original training set is large, k can be smaller.

[0143] In one example, the computer device divides the meteorological numerical samples into ten parts. In the first iterative training, the last part is taken as the test set to obtain the test index E1 of the initial power prediction model at this time. In the second iterative training, the second to last part is taken as the test set to obtain the test index E2 of the initial power prediction model at this time. In the third iterative training, the third to last part is taken as the test set to obtain the test index E3 of the initial power prediction model at this time. Repeat the above operation until the tenth iterative training, take the first part as the test set to obtain the test index E10 of the initial power prediction model at this time.

[0144] The performance index of the initial power prediction model can be obtained by using the following formula when the number of subsets k of the training set is 10:

[0145]

[0146] Among them, E is the performance index of the initial power prediction model, i is the number of iterative training, and E i is the test index of the initial power prediction model obtained by the i-th iteration training.

[0147] In this embodiment, by obtaining multiple groups of meteorological data samples and historical power generation corresponding to each group of meteorological data samples, the correlation between multiple meteorological numerical samples under each variable type and multiple historical power generation powers in the multiple groups of meteorological data samples is determined, and the target variable type is determined from multiple variable types according to the correlation. The multiple groups of meteorological data samples are clustered to obtain the first target category to which each group of meteorological data samples belongs. For each meteorological data sample under the first target category, each initial power prediction model is trained using the target meteorological samples under the target variable type in the meteorological data samples and the power signal samples under each signal frequency to obtain a power prediction model corresponding to each first target category. It is possible to traverse multiple correlation thresholds and perform integrated learning on multiple initial power prediction models to improve the accuracy of the selected model training data, and the initial power prediction model is trained separately for the meteorological data samples under each target category, thereby improving the accuracy of the power prediction model.

[0148] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0149] Based on the same inventive concept, the embodiment of the present application also provides a new energy power prediction device based on improved wavelet transform and neural network for implementing the new energy power prediction method based on improved wavelet transform and neural network involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the new energy power prediction device based on improved wavelet transform and neural network provided below can refer to the limitations of the new energy power prediction method based on improved wavelet transform and neural network above, and will not be repeated here.

[0150] In one embodiment, Figure 7 As shown, a new energy power prediction device 700 based on improved wavelet transform and neural network is provided, comprising: a data sample acquisition module 702, a correlation determination module 704, a variable type determination module 706, a first category determination module 708, a wavelet decomposition module 710 and a model training module 712, wherein:

[0151] The data sample acquisition module 702 is used to acquire multiple groups of meteorological data samples and the historical power generation corresponding to each group of meteorological data samples. Each group of meteorological data samples includes meteorological numerical samples under multiple variable types.

[0152] The correlation determination module 704 is used to determine the correlation between a plurality of meteorological value samples under each variable type and a plurality of historical power generation powers in a plurality of groups of meteorological data samples.

[0153] The variable type determination module 706 is used to determine the initial variable type corresponding to each correlation threshold from multiple variable types according to the correlation and multiple correlation thresholds, and to determine the target threshold from multiple correlation thresholds according to the first training accuracy corresponding to the initial variable type, and the target variable type corresponding to the target threshold, wherein the first training accuracy is obtained by training the initial power prediction model using meteorological numerical samples under the initial variable type.

[0154] The first category determination module 708 is used to perform clustering processing on multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs.

[0155] The wavelet decomposition module 710 is used to perform the following processing for each meteorological data sample under the first target category: perform wavelet decomposition processing on the historical power generation corresponding to each group of meteorological data samples to obtain power signal samples under multiple signal frequencies.

[0156] The model training module 712 is used to train multiple initial power prediction models using the target meteorological samples under the target variable type in each group of meteorological data samples and the power signal samples under each signal frequency, and determine the power prediction model corresponding to each first target category from the trained multiple initial power prediction models. The power prediction model is subsequently used to generate the power generation corresponding to the meteorological data under the first target category.

[0157] In one embodiment, the model training module 712 includes: a signal generating unit, which is used to train each initial power prediction model using the target meteorological samples in each group of meteorological data samples and the power signal samples under each signal frequency to obtain the target power signal corresponding to each signal frequency generated by each initial power prediction model; an accuracy generating unit, which is used to generate a second training accuracy corresponding to each initial power prediction model based on the target power signal; and a model determining unit, which is used to determine, based on the second training accuracy, the power prediction model corresponding to the meteorological data samples under each first target category from the multiple initial power prediction models after training.

[0158] In one embodiment, the variable type determination module 806 includes: a first accuracy acquisition unit, which is used to train the initial power prediction model using meteorological numerical samples under the initial variable type corresponding to each correlation threshold, and obtain the first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under the initial variable type corresponding to each correlation threshold; a type determination unit, which is used to determine the target threshold from multiple correlation thresholds according to the first training accuracy, and use the initial variable type corresponding to the target threshold as the target variable type.

[0159] In one embodiment, the first category determination module 808 includes: an initial feature determination unit, used to determine the initial clustering features corresponding to each group of meteorological data samples; a target feature determination unit, used to normalize the initial clustering features to obtain the target clustering features corresponding to each group of meteorological data samples; a target category determination unit, used to determine the first target category to which each group of meteorological data samples belongs based on the target clustering features and the category features corresponding to each category.

[0160] In one embodiment, the target category determination unit includes: an initial category determination subunit, which is used to use multiple initial cluster numbers to cluster multiple groups of meteorological data samples respectively, and obtain the initial categories of the meteorological data samples under each initial cluster number; an evaluation parameter generation subunit, which is used to generate evaluation parameters corresponding to the initial cluster number according to the difference of target cluster characteristics between two meteorological data samples; a cluster number determination subunit, which is used to determine the target cluster number from multiple initial cluster numbers according to the evaluation parameters, and take the initial category corresponding to the meteorological data samples under the target cluster number as the first target category.

[0161] In one embodiment, during the application stage of the power prediction model, the new energy power prediction device 700 based on improved wavelet transform and neural network also includes: a data acquisition module, used to acquire meteorological data to be predicted, the meteorological data to be predicted includes meteorological values ​​under multiple variable types; a second category determination module, used to cluster the meteorological data to be predicted and multiple groups of meteorological data samples, and determine the second target category to which the meteorological data to be predicted belongs; a power generation module, used to determine the meteorological value under the target variable type corresponding to the target threshold from the meteorological values ​​under multiple variable types, and input the meteorological value under the target variable type into the power prediction model corresponding to the second target category to obtain the generated power.

[0162] Each module in the above-mentioned new energy power prediction device based on improved wavelet transform and neural network can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0163] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store an initial power prediction model. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a new energy power prediction method based on an improved wavelet transform and a neural network is implemented.

[0164] In one embodiment, another computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a new energy power prediction method based on improved wavelet transform and neural network is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse.

[0165] Those skilled in the art will understand that Figure 8 and Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0166] In one embodiment, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0167] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0168] 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 this application are all information and data authorized by the user or fully authorized by all parties.

[0169] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application 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), magnetoresistive 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. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0170] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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.

[0171] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A new energy power prediction method based on improved wavelet transform and neural network, characterized in that: The method comprises: Acquire multiple groups of meteorological data samples and historical power generation corresponding to each group of meteorological data samples, wherein each group of meteorological data samples includes meteorological numerical samples under multiple variable types; Determine the correlation between a plurality of meteorological value samples under each variable type and a plurality of historical power generation powers in a plurality of groups of meteorological data samples; According to the correlation and multiple correlation thresholds, an initial variable type corresponding to each of the correlation thresholds is determined from the multiple variable types; according to a first training accuracy corresponding to the initial variable type, a target threshold and a target variable type corresponding to the target threshold are determined from the multiple correlation thresholds, wherein the first training accuracy is obtained by training an initial power prediction model using meteorological numerical samples under the initial variable type; Performing clustering processing on the multiple groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs; For each of the meteorological data samples under the first target category, the following processing is performed: Performing wavelet decomposition processing on the historical power generation corresponding to each group of meteorological data samples to obtain power signal samples at multiple signal frequencies; Using the target meteorological samples under the target variable type in each group of the meteorological data samples and the power signal samples under each signal frequency to train a plurality of the initial power prediction models, and determining a power prediction model corresponding to the meteorological data samples under each of the first target categories from the trained plurality of the initial power prediction models, the power prediction model being subsequently used to generate power generation corresponding to the meteorological data under the first target category; The clustering process is performed on the plurality of groups of meteorological data samples to obtain the first target category to which each group of meteorological data samples belongs, including: Determining initial clustering features corresponding to each group of meteorological data samples; Normalizing the initial clustering features to obtain target clustering features corresponding to each group of meteorological data samples; Determining the first target category to which each group of meteorological data samples belongs according to the target clustering feature and the category feature corresponding to each category; The determining, according to the target clustering feature and the category feature corresponding to each category, the first target category to which each group of meteorological data samples belongs includes: Using multiple initial clustering numbers to perform clustering processing on multiple groups of meteorological data samples respectively, to obtain initial categories of the meteorological data samples under each of the initial clustering numbers; Generating an evaluation parameter corresponding to the initial number of clusters according to the difference of target clustering characteristics between two meteorological data samples; According to the evaluation parameters, a target cluster number is determined from the multiple initial cluster numbers, and the initial category corresponding to the meteorological data sample under the target cluster number is used as the first target category.

2. The method according to claim 1, characterized in that The method of training a plurality of the initial power prediction models by using the target meteorological samples under the target variable type in each group of the meteorological data samples and the power signal samples under each signal frequency, and determining a power prediction model corresponding to the meteorological data samples under each of the first target categories from the trained plurality of the initial power prediction models, comprises: Using the target meteorological samples in each group of meteorological data samples and the power signal samples at each signal frequency to train each of the initial power prediction models, to obtain a target power signal corresponding to each of the signal frequencies generated by each of the initial power prediction models; Generating a second training accuracy corresponding to each of the initial power prediction models according to the target power signal; According to the second training accuracy, the power prediction model corresponding to the meteorological data sample under each of the first target categories is determined from the multiple trained initial power prediction models.

3. The method according to claim 1, characterized in that The step of determining a target threshold value and a target variable type corresponding to the target threshold value from the plurality of correlation threshold values ​​according to the first training accuracy corresponding to the initial variable type includes: The initial power prediction model is trained using meteorological numerical samples under the initial variable type corresponding to each of the correlation thresholds to obtain a first training accuracy of the initial power prediction model corresponding to the meteorological numerical samples under the initial variable type corresponding to each of the correlation thresholds; According to the first training accuracy, the target threshold is determined from the multiple correlation thresholds, and the initial variable type corresponding to the target threshold is used as the target variable type.

4. The method according to claim 1, characterized in that: The variable type includes any one of a wind speed type, an irradiance type, a temperature type, a humidity type, and an air pressure type.

5. The method according to claim 1, characterized in that The initial power prediction model includes at least one of a BP neural network model, an LSTM model, a BLSTM model, a SVM model, a GRU model, an MLP model, and an RF model.

6. The method according to any one of claims 1 to 5, characterized in that In the application stage of the power prediction model, the method further includes: Acquiring meteorological data to be predicted, wherein the meteorological data to be predicted includes meteorological values ​​under multiple variable types; Performing clustering processing on the meteorological data to be predicted and the multiple groups of meteorological data samples to determine the second target category to which the meteorological data to be predicted belongs; From the meteorological values ​​under the multiple variable types, determine the meteorological value under the target variable type corresponding to the target threshold, input the meteorological value under the target variable type into the power prediction model corresponding to the second target category to obtain the generated power.

7. A new energy power prediction device based on improved wavelet transform and neural network, characterized in that: The device comprises: A data sample acquisition module, used to acquire multiple groups of meteorological data samples and historical power generation corresponding to each group of meteorological data samples, each group of meteorological data samples including meteorological numerical samples under multiple variable types; A correlation determination module, used to determine the correlation between a plurality of meteorological value samples under each of the variable types and a plurality of the historical power generation powers in a plurality of groups of the meteorological data samples; a variable type determination module, configured to determine, according to the correlation and multiple correlation thresholds, an initial variable type corresponding to each of the correlation thresholds from a plurality of the variable types, and to determine, according to a first training accuracy corresponding to the initial variable type, a target threshold from the multiple correlation thresholds, and a target variable type corresponding to the target threshold, wherein the first training accuracy is obtained by training an initial power prediction model using meteorological numerical samples under the initial variable type; A first category determination module is used to perform clustering processing on the multiple groups of meteorological data samples to obtain a first target category to which each group of meteorological data samples belongs; A wavelet decomposition module is used to perform the following processing for each meteorological data sample under the first target category: performing wavelet decomposition processing on the historical power generation corresponding to each group of meteorological data samples to obtain power signal samples under multiple signal frequencies; a model training module, configured to train a plurality of the initial power prediction models by using the target meteorological samples under the target variable type in each group of the meteorological data samples and the power signal samples under each signal frequency, and to determine a power prediction model corresponding to the meteorological data samples under each of the first target categories from the trained plurality of the initial power prediction models, wherein the power prediction model is subsequently used to generate power generation corresponding to the meteorological data under the first target category; The first category determination module includes: an initial feature determination unit, which is used to determine the initial clustering features corresponding to each group of meteorological data samples; a target feature determination unit, which is used to normalize the initial clustering features to obtain the target clustering features corresponding to each group of meteorological data samples; a target category determination unit, which is used to determine the first target category to which each group of meteorological data samples belongs based on the target clustering features and the category features corresponding to each category; The target category determination unit includes: an initial category determination subunit, which is used to use multiple initial clustering numbers to cluster multiple groups of meteorological data samples respectively, and obtain the initial categories of the meteorological data samples under each initial clustering number; an evaluation parameter generation subunit, which is used to generate evaluation parameters corresponding to the initial clustering number according to the difference of target clustering characteristics between two meteorological data samples; a cluster number determination subunit, which is used to determine the target clustering number from multiple initial clustering numbers according to the evaluation parameters, and use the initial category corresponding to the meteorological data samples under the target clustering number as the first target category.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images