Power prediction model construction method, power prediction method, power prediction device, power prediction equipment and medium
By using convolutional neural networks to select and cluster features in the historical data of photovoltaic power generation and wind power plant stations, an accurate power prediction model is constructed, which solves the problem of inaccurate meteorological data processing in the existing technology and achieves more efficient power generation prediction.
Patent Information
- Application Number
- CN202510252976.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-05
- Publication Date
- 2025-08-01
AI Technical Summary
Existing power prediction models are difficult to identify linear and nonlinear correlations and complex periodic relationships between meteorological data and power generation. Traditional clustering methods perform poorly when processing non-convex and unbalanced data, resulting in inaccurate prediction of photovoltaic power generation and wind power generation.
By acquiring the historical acquisition data of the target site, feature selection is performed based on the dependence between the sub-hierarchical acquisition data and historical power data, clustering is used to use convolutional neural networks to build a cluster cluster with different clustering centers, and inputting it into a preset power prediction model for training, optimizing model parameters to reduce prediction loss.
Deeply digging into the linear and nonlinear correlations between variables improves the accuracy of power generation prediction and the generalization ability of the model, and reduces the calculation cost.
Smart Images
Figure CN120408236A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of clean energy, and particularly to a method for constructing a power prediction model, a prediction method, a device, equipment and a medium. Background Art
[0002] Photovoltaic power generation and wind power generation are important components of clean energy, which are of great significance for reducing the dependence on fossil energy and reducing greenhouse gas emissions; accurately predicting the power generation can help grid companies to carry out reasonable scheduling, improve the utilization rate of new energy, and promote the transformation and upgrading of the energy structure.
[0003] For the power generation prediction of photovoltaic power generation and wind power generation, it is usually realized by inputting meteorological data into a pre-trained power prediction model to obtain a power prediction result. The power prediction model is obtained by fitting the corresponding relationship between meteorological data and power generation data during training; however, on the one hand, the existing power prediction models are difficult to identify the linear and nonlinear correlations between input variables, as well as the complex periodic relationships between variables; on the other hand, for different categories of meteorological data, the traditional clustering method has problems of being unsuitable for non-convex data and unbalanced data;
[0004] The above two problems lead to inaccurate power prediction. Summary of the Invention
[0005] In view of the above existing problems, the present invention is proposed.
[0006] Therefore, the present invention provides a method for constructing a power prediction model, a prediction method, a device, equipment and a medium to solve the problems that the existing power prediction models are difficult to identify the linear and nonlinear correlations and complex periodic relationships between meteorological data and power generation, and at the same time, the traditional clustering method performs poorly in dealing with non-convex and unbalanced meteorological data, resulting in inaccurate prediction of photovoltaic power generation and wind power generation.
[0007] To solve the above technical problems, the present invention provides the following technical solutions:
[0008] In a first aspect, the present invention provides a method for constructing a power prediction model, including:
[0009] Obtain historical acquisition data and historical power data of a target station, where the historical acquisition data includes at least two sub-historical acquisition data related to the historical power data;
[0010] Perform a first feature selection based on the dependence relationship between each sub-historical acquisition data and the historical power data to obtain a first feature selection result;
[0011] Input the first feature selection result into a pre-constructed convolutional neural network for second feature selection to obtain multiple clustering clusters with different clustering centers;
[0012] Input each of the clustering clusters into a pre-constructed preset power prediction model for model training, and obtain the preset power prediction model with an actual prediction loss less than the preset prediction loss during the training process as the power prediction model.
[0013] As a preferred solution of the power prediction model construction method of the present invention, wherein: the first feature selection based on the dependency relationship between each of the sub-historical acquisition data and the historical power data includes:
[0014] Calculate the first maximum information coefficient value of the sub-historical acquisition data relative to the historical power data;
[0015] Perform first feature selection on the sub-historical acquisition data based on the first maximum information coefficient value to obtain a first feature selection result; wherein, the first feature selection result includes multiple target sub-acquisition data with a first maximum information coefficient value greater than a first preset information coefficient value.
[0016] As a preferred solution of the power prediction model construction method of the present invention, wherein: the first feature selection based on the dependency relationship between each of the sub-historical acquisition data and the historical power data further includes:
[0017] Calculate the second maximum information coefficient value between each of the target sub-acquisition data;
[0018] In the first feature selection result, eliminate the target sub-acquisition data with a second maximum information coefficient value greater than a second preset information coefficient value.
[0019] As a preferred solution of the power prediction model construction method of the present invention, wherein: the inputting the first feature selection result into a pre-constructed convolutional neural network for second feature selection includes:
[0020] Input the first feature selection result into a pre-constructed convolutional neural network for clustering to obtain the clustering clusters;
[0021] The convolutional neural network includes at least two convolutional layers, and the convolutional neural network is used to perform multi-layer convolution based on the convolutional layers to determine the clustering center points of the first feature selection result; during the clustering process of the convolutional neural network, based on the average value of the nearest clustering center of each target sub-acquisition data as the loss function, optimize the network weight parameters of the convolutional neural network through backpropagation gradient descent, and update the position of the clustering center points through forward propagation.
[0022] As a preferred solution of the power prediction model construction method described in the present invention, wherein: the step of inputting each of the clustering clusters into a preset power prediction model constructed in advance for model training includes: <tmp
[0023] Dividing the clustering clusters to obtain training sets and validation sets corresponding to multiple clustering clusters;
[0024] Inputting the training sets corresponding to each of the clustering clusters into a preset power prediction model constructed in advance for multiple rounds of iteration. During each round of iteration, updating the parameters of the preset power prediction model constructed in advance until the model converges to obtain a preset power prediction model corresponding to each of the clustering clusters;
[0025] Inputting the validation sets of each of the clustering clusters into the preset power prediction model corresponding to the clustering cluster for verification to obtain verification results;
[0026] Based on the verification results, selecting the preset power prediction model with an actual prediction loss less than the preset prediction loss as the power prediction model.
[0027] As a preferred solution of the power prediction model construction method described in the present invention, wherein: the preset power prediction model further includes an algorithm optimization module. During each round of iteration, the algorithm optimization module is used to update the model parameters of the power prediction module.
[0028] In a second aspect, the present invention provides a power prediction method, including:
[0029] Obtaining a preset power prediction model, wherein the preset power prediction model is obtained by training based on preset clustering clusters, and the actual prediction loss of the preset clustering clusters during the training process is less than the preset prediction loss; the preset clustering clusters are obtained by performing second feature selection based on the clustering degree between the first feature selection results, and the first feature selection results include multiple target sub-collection data with an actual dependence degree greater than the preset dependence degree, and each of the target sub-collection data is obtained by performing first feature selection based on the dependence relationship between the pre-collected historical collection data and the historical power data;
[0030] Obtaining prediction data corresponding to the target sub-collection data;
[0031] Inputting the prediction data into the preset power prediction model for power prediction to obtain a power prediction result.
[0032] In a third aspect, the present invention provides a power prediction model construction device, including:
[0033] An acquisition module, configured to acquire historical acquisition data and historical power data of a target station, where the historical acquisition data includes at least two sub-historical acquisition data related to the historical power data;
[0034] A first feature selection module, configured to perform first feature selection based on the dependence relationship between each of the sub-historical acquisition data and the historical power data, to obtain a first feature selection result;
[0035] A second feature selection module, configured to input the first feature selection result into a pre-constructed convolutional neural network to perform second feature selection, to obtain a plurality of clustering clusters with different clustering centers;
[0036] A model training module, configured to input each of the clustering clusters into a pre-constructed preset power prediction model to perform model training, and use the preset power prediction model with an actual prediction loss less than a preset prediction loss during the training process as a power prediction model.
[0037] Fourthly, the present invention provides an electronic device, including:
[0038] A memory and a processor;
[0039] The memory is configured to store computer-executable instructions, and the processor is configured to execute the computer-executable instructions, and when the computer-executable instructions are executed by the processor, the steps of the above method are implemented.
[0040] Fifthly, the present invention provides a computer-readable storage medium, which stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the steps of the above method are implemented.
[0041] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a power prediction model construction method, prediction method, device, equipment and medium, which obtains the historical collection data and historical power data of the target site, performs a first feature selection based on the dependency relationship between each sub-historical collection data and the historical power data, and obtains a first feature selection result; inputs the target sub-collection data into a pre-built convolutional neural network for second feature selection, and obtains multiple clusters with different cluster centers; inputs each cluster into a pre-built preset power prediction model for model training, and obtains a preset power prediction model in which the actual prediction loss during the training process is less than the preset prediction loss as a power prediction model. By inputting the target sub-collection data into the pre-built convolutional neural network for second feature selection, it is possible to deeply explore the linear and nonlinear correlations between variables, including complex periodic relationships, so as to more accurately predict power generation; the convolutional neural network is used as a feature selection tool to identify the features with the highest correlation with the target variable, thereby reducing computational costs and improving the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0043] Figure 1 A schematic diagram of a flow chart of a method for constructing a power prediction model according to an embodiment of the present invention;
[0044] Figure 2 A schematic flow chart of a power prediction method according to an embodiment of the present invention;
[0045] Figure 3 This is a structural block diagram of a power prediction model construction device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0046] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.
[0047] Example 1
[0048] In this embodiment, a method for constructing a power prediction model is provided.Figure 1 The figure shows a flowchart of a method for constructing a power prediction model according to an embodiment of the present invention. The process includes the following steps:
[0049] S101: Obtain the historical acquisition data and historical power data of the target station, where the historical acquisition data includes at least two sub-historical acquisition data related to the historical power data;
[0050] In an embodiment of the present application, the target station may be a photovoltaic power station or a wind farm, and the historical acquisition data may include at least two sub-historical acquisition data related to meteorology;
[0051] Exemplarily, the sub-historical acquisition data may include temperature, humidity, wind speed, air pressure, etc. The sub-historical acquisition data may also include photovoltaic power influence parameters such as irradiance and aerosol concentration influence parameters according to the type of the target station.
[0052] It should be noted that the above step S101 provides a rich data basis for subsequent feature selection and model training by obtaining the historical acquisition data and historical power data of the target station, which helps to improve the accuracy and reliability of the prediction model.
[0053] S102: Perform a first feature selection based on the dependence relationship between each sub-historical acquisition data and the historical power data to obtain a first feature selection result; wherein, the first feature selection result includes a plurality of target sub-acquisition data with an actual dependence degree greater than a preset dependence degree;
[0054] It should be noted that after obtaining the sub-historical acquisition data and the historical power data, the historical sub-acquisition data is screened according to the dependence relationship between the sub-historical acquisition data and the historical power data to obtain target sub-acquisition data with an actual dependence degree greater than the preset dependence degree on the historical power data.
[0055] In an optional embodiment, a statistical method may be used to implement the extraction of the dependence relationship between the sub-historical acquisition data and the historical power data, and further perform a first feature selection according to the dependence relationship extraction result to obtain a first feature selection result.
[0056] Exemplarily, the statistical method may include Pearson correlation coefficient, mutual information, etc.
[0057] In an alternative embodiment, based on the obtained historical acquisition data and historical power data, the dependence relationship between the sub-historical acquisition data and the historical power data is extracted by calculating the Maximal Information Coefficient (MIC) value between the historical acquisition data and the historical power data; wherein, the MIC value is proportional to the dependence relationship between the sub-historical acquisition data and the historical power data; after obtaining the MIC value, the target sub-acquisition data with the MIC value greater than the preset MIC value is selected as the first feature selection result.
[0058] In an alternative embodiment, in addition to the above statistical method, there are various other techniques for the first feature selection operation to evaluate the dependence relationship between the sub-historical acquisition data and the historical power data and perform feature selection accordingly. For example, the principal component analysis method can be used to help reduce the number of input variables while retaining as much information as possible by identifying the main directions in the data; the recursive feature elimination method can also be used, starting from the complete feature set, removing the least important features in each iteration until the required number of features is reached.
[0059] In the embodiment of the present application, the steps of performing the first feature selection based on the dependence relationship between each sub-historical acquisition data and the historical power data to obtain the first feature selection result specifically include: calculating the first maximal information coefficient value of each sub-historical acquisition data relative to the historical power data; wherein, the first maximal information coefficient value is proportional to the degree of dependence of the sub-historical acquisition data on the historical power data; performing the first feature selection on the sub-historical acquisition data based on the first maximal information coefficient value to obtain the first feature selection result; wherein, the first feature selection result includes multiple target sub-acquisition data with the first maximal information coefficient value greater than the first preset information coefficient value.
[0060] In this embodiment, the dependence relationship between the sub-historical acquisition data and the historical power data is characterized by the maximal information coefficient value; as a possible implementation, the maximal information coefficient value can be calculated through a dedicated library; specifically, the library can include minepy.
[0061] As an exemplary embodiment, the power prediction model construction method further includes: calculating the second maximal information coefficient value between each target sub-acquisition data; in the first feature selection result, removing the target sub-acquisition data with the second maximal information coefficient value greater than the second preset information coefficient value.
[0062] It should be noted that the first preset information coefficient value and the second preset information coefficient value are key thresholds for determining which sub-historical acquisition data are selected as features, and can be determined according to specific application scenarios, characteristics of the data set, and requirements for model performance.
[0063] In the method for constructing a power prediction model in the related art, after performing a correlation analysis between variable features and power, all variable features related to power are usually used as influencing factors affecting power to construct a data set, and further, the constructed data set is used for model training. However, this method does not consider the redundancy between variable features related to power, and using highly correlated features during model training results in a poor prediction effect of the obtained power prediction model and inaccurate power prediction.
[0064] To solve this problem, in this embodiment, after obtaining the first feature selection result, the second maximum information coefficient value between each target sub-collected data is calculated, and further, the target sub-collected data with the second maximum information coefficient value greater than the second preset information coefficient value is removed from the first feature selection result, so as to consider the redundancy with other features at the same time and avoid selecting highly correlated features.
[0065] It should be noted that the above step S102 performs the first feature selection based on the dependence relationship between the sub-historical collected data and the historical power data, filters out the target sub-collected data with an actual dependence degree greater than the preset threshold, accurately locates the features that have a significant impact on power prediction, and improves the data quality and prediction accuracy of subsequent model training.
[0066] S103: Input the target sub-collected data into a pre-constructed convolutional neural network for the second feature selection to obtain multiple clustering clusters with different clustering centers;
[0067] It should be noted that when the power prediction model in the related art is trained, on the one hand, if the training data set is not clustered, although each data contained in the data set is related to the power generation data, however, the data set usually contains meteorological data in different situations and categories, and the finally trained model cannot distinguish according to the actual situation and category of the data set; if the training data set is clustered, due to the influence of seasonality, randomness, and volatility of wind power generation and photovoltaic power generation, the data set usually has imbalance and non-convexity, and using traditional clustering methods has problems of being unsuitable for non-convex data and unbalanced data; on the other hand, the existing power prediction model is difficult to identify the linear and non-linear correlations between input variables, as well as the complex periodic relationships between variables; the above two problems lead to inaccurate power prediction in the prior art.
[0068] In this embodiment, to solve the above problems, the target sub-collected data is input into a pre-constructed convolutional neural network for second feature selection to obtain multiple clustering clusters with different clustering centers. Among them, the convolutional neural network can learn the abstract feature representation of the target sub-collected data through convolutional operations, so that the same type of data gathers in the feature space. The method of using a pre-constructed convolutional neural network for clustering can utilize the characteristics of the convolutional neural network to obtain the optimal clustering center points through multiple layers of convolution. It is insensitive to data types, can also perform clustering well on imbalanced data and non-convex data, and is insensitive to the initial clustering center, with strong clustering stability.
[0069] In an alternative embodiment, the convolutional neural network includes at least two convolutional layers. The convolutional neural network is used to perform multiple layers of convolution based on the convolutional layers to determine the clustering center points of the first feature selection result. During the clustering process, the average value of the nearest clustering center of each target sub-collected data is used as the loss function, and the network weight parameters of the convolutional neural network are optimized through backpropagation gradient descent, and the positions of the clustering center points are updated through forward propagation.
[0070] As an exemplary embodiment, inputting the target sub-collected data into a pre-constructed convolutional neural network for second feature selection to obtain multiple clustering clusters with different clustering centers includes: inputting the first feature selection result into a pre-constructed convolutional neural network for clustering to obtain clustering clusters. Among them, the convolutional neural network includes at least two convolutional layers. The convolutional neural network is used to perform multiple layers of convolution based on the convolutional layers to determine the clustering center points of the first feature selection result. During the clustering process, the average value of the nearest clustering center of each target sub-collected data is used as the loss function, and the network weight parameters of the convolutional neural network are optimized through backpropagation gradient descent, and the positions of the clustering center points are updated through forward propagation.
[0071] In this embodiment, for the selected target sub-collected data, the method of convolutional clustering is used for clustering. During the clustering process, the characteristics of the Convolutional Neural Networks (CNN) are utilized to extract the optimal clustering center points through multiple layers of convolution. Among them, the CNN includes at least two convolutional layers. The CNN is used to perform multiple layers of convolution based on the convolutional layers to determine the clustering center points of the first feature selection result. During the clustering process, the average value of the Euclidean distances from each data point to the nearest clustering center is used as the loss function, and the network weight parameters are continuously optimized through backpropagation gradient descent, and the positions of the clustering center points are updated through forward propagation to obtain the optimal clustering center. The method of using the CNN for data clustering is insensitive to data types, can also perform clustering well on imbalanced data and non-convex data, and is insensitive to the initial clustering center, with strong clustering stability.
[0072] It should be noted that the above method of optimizing the network weight parameters of the convolutional neural network through backpropagation gradient descent, with the optimized convolutional neural network as a feature selection tool, can deeply explore the linear and non-linear correlations between variables, including complex periodic relationships, and can identify the features with the highest correlation with power data.
[0073] S104: Input each clustering cluster into a pre-constructed preset power prediction model for model training, and obtain the preset power prediction model with the actual prediction loss less than the preset prediction loss during the training process as the power prediction model.
[0074] Specifically, after obtaining multiple clustering clusters clustered by using a pre-constructed convolutional neural network, exemplarily, a corresponding preset power prediction model is constructed for each clustering cluster. Further, each clustering cluster is respectively input into the pre-constructed preset power prediction model for model training to obtain the power prediction model corresponding to each clustering cluster; finally, based on the prediction loss of the power prediction model, the preset power prediction model with the actual prediction loss less than the preset prediction loss is selected as the power prediction model.
[0075] Specifically, when inputting each clustering cluster into the pre-constructed preset power prediction model for model training, each clustering cluster is at least divided into a training set, a validation set, and a test set; after inputting the training set included in each clustering cluster into the correspondingly constructed preset power prediction model to obtain the training set prediction results corresponding to each clustering cluster, the validation set included in each clustering cluster is input into the trained power prediction model for validation, and based on the validation results, the preset power prediction model with the actual prediction loss less than the preset prediction loss is retained as the power prediction model.
[0076] In an alternative embodiment, after obtaining the power prediction model, the test sets of each clustering cluster are input into the power prediction model for testing to evaluate the performance of the model;
[0077] Specifically, when inputting the test set into the power prediction model, read the data line by line, determine the clustering cluster where it is located, and then make a prediction, output the prediction results and calculate a series of statistical metrics to evaluate the performance of the model; among them, the statistical metrics may include mean absolute percentage error, root mean square error, mean bias error, coefficient of determination, etc.
[0078] As an exemplary embodiment, each clustering cluster is input into a pre-constructed preset power prediction model for model training, including: dividing the clustering cluster to obtain training sets and validation sets corresponding to multiple clustering clusters; respectively inputting the validation sets corresponding to each clustering cluster into the power prediction module for multiple rounds of iteration, and during each round of iteration, updating the model parameters of the power prediction module until the model converges to obtain a power prediction model corresponding to each clustering cluster; inputting the validation sets of each clustering cluster into the power prediction model corresponding to the clustering cluster for verification to obtain verification results; based on the verification results, selecting a power prediction model with an actual prediction loss less than the preset prediction loss as the power prediction model.
[0079] In this embodiment, the pre-constructed preset power prediction model can be an Informer prediction model; when using the Informer prediction model for model training, the Informer prediction model can capture the global information of each input data time series, and using the Informer prediction model can improve the performance when processing complex time series data.
[0080] Exemplarily, divide the clustering cluster to obtain training sets and validation sets corresponding to multiple clustering clusters; respectively input the validation sets corresponding to each clustering cluster into the power prediction module for multiple rounds of iteration, and during each round of iteration, update the model parameters of the power prediction module until the model converges to obtain an Informer prediction model corresponding to each clustering cluster; input the validation sets of each clustering cluster into the Informer prediction model corresponding to the clustering cluster for verification to obtain verification results; based on the verification results, select an Informer prediction model with an actual prediction loss less than the preset prediction loss as the Informer prediction model.
[0081] As an exemplary embodiment, the preset power prediction model further includes an algorithm optimization module; the method for constructing the power prediction model further includes: during each round of iteration, using the algorithm optimization module to update the model parameters of the power prediction module.
[0082] In this embodiment, the algorithm optimization module is used to update the model parameters of the power prediction module during each round of iteration. As an exemplary embodiment, the algorithm optimization module can be constructed based on the Sparrow Search Algorithm (SSA). When the validation sets corresponding to each clustering cluster are respectively input into the power prediction module for multiple rounds of iteration, during each round of iteration, the hyperparameters of the Informer prediction model are updated through SSA until the model converges. Using SSA in combination with Informer can not only capture the global information of the time series, but also effectively adjust the hyperparameters of the Informer model, helping to find a better parameter combination, thereby improving the performance of the model in processing complex time series data.
[0083] It should be noted that in the above step S104, by inputting each clustering cluster into a preset power prediction model for training and screening out the model with an actual prediction loss less than a preset threshold, the efficiency and accuracy of the model during training are ensured, thereby improving the reliability and prediction performance of the final power prediction model and effectively meeting the actual application requirements.
[0084] As can be seen from the above embodiments, the present invention provides a method for constructing a power prediction model. By obtaining the historical acquisition data and historical power data of the target power station, based on the dependence relationship between each sub-historical acquisition data and the historical power data, the first feature selection is performed to obtain the first feature selection result. The target sub-acquisition data is input into a pre-constructed convolutional neural network for the second feature selection to obtain multiple clustering clusters with different clustering centers. Each clustering cluster is respectively input into a pre-constructed preset power prediction model for model training, and the preset power prediction model with an actual prediction loss less than the preset prediction loss during the training process is used as the power prediction model. By inputting the target sub-acquisition data into a pre-constructed convolutional neural network for the second feature selection, the linear and non-linear correlations between variables, including complex periodic relationships, can be deeply mined, so as to more accurately predict the power generation. The convolutional neural network, as a tool for feature selection, is used to identify the features with the highest correlation with the target variable, thereby reducing the computational cost and improving the generalization ability of the model.
[0085] In this embodiment, a power prediction method is also provided, as Figure 2 shown in the flowchart of the power prediction method according to an embodiment of the present invention. The process includes the following steps:
[0086] S201: Obtain a preset power prediction model, where the preset power prediction model is obtained by training a model based on a preset clustering cluster, and the actual prediction loss of the preset clustering cluster during the training process is less than the preset prediction loss; the preset clustering cluster is obtained by performing second feature selection based on the clustering degree among the first feature selection results, and the first feature selection results include multiple target sub-collection data with an actual dependence degree greater than the preset dependence degree, and each target sub-collection data is obtained by performing first feature selection based on the dependence relationship between the pre-collected historical collection data and the historical power data.
[0087] In the embodiment of the present application, the preset power prediction model is obtained by training a model based on a preset clustering cluster, and the actual prediction loss of the preset clustering cluster during the training process is less than the preset prediction loss; the preset clustering cluster is obtained by performing second feature selection based on the clustering degree among the first feature selection results, and the first feature selection results include multiple target sub-collection data with an actual dependence degree greater than the preset dependence degree, and each target sub-collection data is obtained by performing first feature selection based on the dependence relationship between the pre-collected historical collection data and the historical power data; by using the prediction power prediction model trained based on the preset clustering cluster to perform power prediction, the linear and non-linear correlations between variables in the input data, including complex periodic relationships, can be deeply mined, so as to more accurately predict the power generation; the convolutional neural network is used as a tool for feature selection to identify the features with the highest correlation with the target variable, thereby reducing the calculation cost and improving the generalization ability of the model.
[0088] S202: Obtain prediction data corresponding to the target sub-collection data;
[0089] S203: Input the prediction data into the preset power prediction model for power prediction to obtain a power prediction result.
[0090] It should be noted that for the above power prediction method in this embodiment, by using the prediction power prediction model trained based on the preset clustering cluster to perform power prediction, the linear and non-linear correlations and complex periodic relationships between variables in the input data can be deeply mined, so as to more accurately predict the power generation.
[0091] Embodiment 2
[0092] In this embodiment, a power prediction model construction device is provided, as Figure 3 shown, including:
[0093] An acquisition module 301, configured to acquire historical collection data and historical power data of a target power station, where the historical collection data includes at least two sub-historical collection data related to the historical power data;
[0094] The first feature selection module 302 is configured to perform first feature selection based on the dependency relationships between the respective sub-historical acquisition data and the historical power data, so as to obtain a first feature selection result;
[0095] The second feature selection module 303 is configured to input the first feature selection result into a pre-constructed convolutional neural network for second feature selection, so as to obtain multiple clustering clusters with different clustering centers;
[0096] The model training module 304 is configured to input each clustering cluster into a pre-constructed preset power prediction model for model training, and obtain a preset power prediction model with an actual prediction loss less than a preset prediction loss during the training process as the power prediction model.
[0097] The above-mentioned unit modules can be embedded in a processor in a computer device in a hardware form or be independent of the processor, or can be stored in a memory in the computer device in a software form, so as to facilitate the processor to call and execute the operations corresponding to the above respective modules.
[0098] This embodiment also provides an electronic device, which 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 implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. 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 covering the display screen, or can be a button, a trackball, or a touchpad provided on the housing of the computer device, or can also be an external keyboard, a touchpad, or a mouse, etc.
[0099] This embodiment also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the method proposed in the above embodiment.
[0100] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.
[0101] From the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software and necessary general-purpose hardware. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disc of a computer, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of the embodiments of the present invention.
[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.
[0103] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages.
[0104] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.
[0105] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means embodying the functionality specified in the flowchart(s) Figure 1 or flowcharts and / or block(s) Figure 1 or blocks.
[0106] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing steps for implementing the functionality specified in the flowchart(s) Figure 1 or flowcharts and / or block(s) Figure 1 or blocks.
[0107] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.
[0108] It is apparent that those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and modifications.
Claims
1. A method for constructing a power prediction model, characterized in that, Including: Obtain the historical acquisition data and historical power data of the target station, where the historical acquisition data includes at least two sub-historical acquisition data related to the historical power data; Perform first feature selection based on the dependency relationship between each sub-historical acquisition data and the historical power data to obtain a first feature selection result; Input the first feature selection result into a pre-constructed convolutional neural network for second feature selection to obtain multiple clustering clusters with different clustering centers; Input each clustering cluster into a pre-constructed preset power prediction model for model training, and obtain the preset power prediction model with an actual prediction loss less than the preset prediction loss during the training process as the power prediction model.
2. The method for constructing a power prediction model according to claim 1, wherein The performing first feature selection based on the dependency relationship between each sub-historical acquisition data and the historical power data includes: Calculate the first maximum information coefficient value of the sub-historical acquisition data relative to the historical power data; Perform first feature selection on the sub-historical acquisition data based on the first maximum information coefficient value to obtain a first feature selection result; wherein, the first feature selection result includes multiple target sub-acquisition data with a first maximum information coefficient value greater than a first preset information coefficient value.
3. The power prediction model construction method according to claim 2, characterized in that, The performing first feature selection based on the dependency relationship between each sub-historical acquisition data and the historical power data further includes: Calculate the second maximum information coefficient value between each target sub-acquisition data; In the first feature selection result, eliminate the target sub-acquisition data with a second maximum information coefficient value greater than a second preset information coefficient value.
4. The method for constructing a power prediction model according to claim 3, wherein The inputting the first feature selection result into a pre-constructed convolutional neural network for second feature selection includes: Input the first feature selection result into a pre-constructed convolutional neural network for clustering to obtain the clustering clusters; The convolutional neural network includes at least two convolutional layers, and the convolutional neural network is used to perform multi-layer convolution based on the convolutional layers to determine the clustering center points of the first feature selection result; during the clustering process, the average value of the nearest clustering center of each target sub-acquisition data is used as the loss function, and the network weight parameters of the convolutional neural network are optimized through backpropagation gradient descent, and the positions of the clustering center points are updated through forward propagation.
5. The method for constructing a power prediction model according to claim 4, wherein The inputting each clustering cluster into a pre-constructed preset power prediction model for model training includes: Divide the clustering clusters to obtain training sets and validation sets corresponding to multiple clustering clusters; Input the training sets corresponding to each clustering cluster into a pre-constructed preset power prediction model for multiple rounds of iteration. During each round of iteration, update the parameters of the pre-constructed preset power prediction model until the model converges to obtain the preset power prediction models corresponding to each clustering cluster; Input the validation sets of each clustering cluster into the preset power prediction model corresponding to the clustering cluster for validation to obtain a validation result; Based on the verification result, select the preset power prediction model with an actual prediction loss less than the preset prediction loss as the power prediction model.
6. The power prediction model construction method according to claim 5, wherein, The preset power prediction model further includes an algorithm optimization module, and during each round of iteration, the algorithm optimization module is used to update the model parameters of the power prediction module.
7. A power prediction method, characterized in that, Including: Obtain a preset power prediction model, where the preset power prediction model is obtained by training a model based on a preset clustering cluster, and the actual prediction loss of the preset clustering cluster during the training process is less than the preset prediction loss; the preset clustering cluster is obtained by performing second feature selection based on the clustering degree among the first feature selection results, and the first feature selection results include multiple target sub-collection data with an actual dependence degree greater than the preset dependence degree, and each of the target sub-collection data is obtained by performing first feature selection based on the dependence relationship between the pre-collected historical collection data and the historical power data; Obtain prediction data corresponding to the target sub-collection data; Input the prediction data into the preset power prediction model for power prediction to obtain a power prediction result.
8. A power prediction model construction device, characterized in that, Including: An acquisition module, configured to acquire historical collection data and historical power data of a target station, where the historical collection data includes at least two sub-historical collection data related to the historical power data; A first feature selection module, configured to perform first feature selection based on the dependence relationship between each of the sub-historical collection data and the historical power data to obtain a first feature selection result; A second feature selection module, configured to input the first feature selection result into a pre-constructed convolutional neural network for second feature selection to obtain multiple clustering clusters with different clustering centers; A model training module, configured to input each of the clustering clusters into a pre-constructed preset power prediction model for model training, and obtain the preset power prediction model with an actual prediction loss less than the preset prediction loss during the training process as the power prediction model.
9. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the power prediction model construction method according to any one of claims 1 to 6 or the power prediction method according to claim 7 are implemented.
10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer-executable instructions are executed by the processor, the steps of the power prediction model construction method according to any one of claims 1 to 6 or the power prediction method according to claim 7 are implemented.