Wind power prediction method and system based on data enhancement and hidden layer incremental cascading

Abstract

The present invention provides a wind power prediction method and system based on data enhancement and hidden layer incremental cascade. The method comprises: collecting an initial sample set and performing preprocessing, and dividing the initial sample set into a training set, a validation set and a test set; based on a traditional RVFL model, introducing a hidden layer node addition mechanism, and constructing an HLICRVFL model; performing preliminary training on the HLICRVFL model using the training set; adding Student's t distribution noise to the training set, generating a new training set, and training the newly added nodes, while keeping the parameters of the original nodes unchanged, continuously performing training until the model converges, and finally using the optimal model to predict wind power; the present invention adopts a hidden layer incremental training mode, does not need to retrain all parameters, only needs to train the newly added nodes, and significantly reduces the calculation cost; in addition, the present invention performs data enhancement based on Student's t distribution, can generate higher quality training samples, and further improves the prediction accuracy of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and more specifically, to a wind power prediction method and system based on data enhancement and hidden layer incremental cascading. Background Art

[0002] Wind power generation is intermittent and volatile, so accurate prediction of wind power is crucial for grid dispatch and the economics of wind power generation. Traditional wind power prediction methods mostly rely on a large amount of historical data, such as statistical analysis, time series models, regression analysis, etc. However, in practical applications, due to factors such as abnormal weather or newly built wind turbines, it is often difficult to obtain sufficient high-quality historical data, which makes small sample learning a worthy research direction.

[0003] In the field of small-sample wind power forecasting, researchers have made important contributions to solving the problem of data scarcity through a variety of innovative methods. First, meta-learning is widely used in wind power forecasting. Researchers use meta-learning methods to enable models to quickly adapt to new forecasting tasks with a small number of training samples. Meta-Learning has demonstrated its strong generalization ability on multiple wind farm tasks and can effectively overcome the problem of insufficient data for a single wind farm. Secondly, transfer learning has become another key method. Researchers have improved the prediction accuracy by transferring the knowledge learned from the source task to the target task, especially when performing collaborative forecasting between wind farms. For example, based on historical data of similar wind farms, the model can reduce the need for a large amount of data for the target wind farm. In addition, the application of generative models such as generative adversarial networks (GANs) and variational autoencoders (VAEs) also provides new solutions for small-sample forecasting. By generating virtual data to expand the training set, the prediction performance of the model is improved.

[0004] Although existing data augmentation methods can effectively expand the training data set, they still have some limitations. First, the generated data may deviate from the real data, which in turn interferes with the model training process. Second, the spatiotemporal data characteristics in wind power prediction are relatively complex, and simple augmentation methods often fail to capture the dynamic changes and nonlinear relationships of wind data. In addition, designing a reasonable data augmentation strategy not only requires rich domain knowledge, but also comes with high computational overhead, which increases the difficulty of implementation. Summary of the invention

[0005] In order to overcome the defects of low wind power prediction accuracy and high calculation cost of small sample in the above-mentioned prior art, the present invention provides a wind power prediction method and system based on data enhancement and hidden layer incremental cascade, and designs an adaptive hidden layer incremental cascade RVFL (Random vector functional link network) model (denoted as HLICRVFL) for data enhancement. The model can automatically select a suitable training sample size, thereby reducing the underfitting or overfitting problem of the model; secondly, the hidden layer incremental form adopted by HLICRVFL does not need to retrain all parameters, but only needs to train the newly added hidden layer nodes, which significantly reduces the calculation cost; in addition, the small sample generation method based on Student's t-distribution of the present invention can generate higher quality training samples and further improve the prediction accuracy of the model.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0007] A wind power prediction method based on data enhancement and hidden layer incremental cascading includes the following steps:

[0008] S1: Collect historical power generation data of the wind power cluster for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set, and a test set;

[0009] S2: Based on the traditional RVFL neural network model, a hidden layer node addition mechanism is introduced to construct the HLICRVFL neural network model;

[0010] S3: Performing preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training;

[0011] S4: adding Student's t distribution noise to the training set to generate a new training set;

[0012] S5: adding several nodes to the hidden layer of the HLICRVFL neural network model after the initial training, and training the newly added nodes using the new training set, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after the secondary training;

[0013] S6: Determine whether the HLICRVFL neural network model after the secondary training converges on the verification set. If so, complete the training, obtain the optimal HLICRVFL neural network model, and execute step S7; otherwise, re-execute steps S4-S5;

[0014] S7: Input the test set into the optimal HLICRVFL neural network model for wind power forecasting and evaluate the prediction effect of the model.

[0015] Preferably, in step S1, interpolation processing is performed on the abnormal values ​​in the initial sample set, and data normalization is performed to complete preprocessing.

[0016] Preferably, in step S2, the HLICRVFL neural network model comprises an input layer, a hidden layer and an output layer connected in sequence; the hidden layer comprises a plurality of original nodes;

[0017] The hidden layer node adding mechanism includes: adding a number of nodes to the hidden layer of the HLICRVFL neural network model each time a new training set is generated.

[0018] Preferably, the step S3 comprises:

[0019] The weights and biases of the original nodes in the hidden layer of the HLICRVFL neural network model are randomly generated, and the output matrix of all the original nodes in the hidden layer is calculated. The calculation process is:

[0020]

[0021]

[0022] Among them, Z is the output matrix of all original nodes in the hidden layer; X is the training set, L is the number of samples in the training set, and D is the number of features in the training set; f is the activation function; W is the weight matrix of the randomly generated original nodes in the hidden layer; B is the bias matrix of the randomly generated original nodes in the hidden layer; m is the number of original nodes in the hidden layer; is the weight corresponding to the mth original node and the Dth feature; is the Dth feature of the Lth sample; is the bias corresponding to the mth original node;

[0023] Calculate the output of the output layer of the HLICRVFL neural network model. The calculation process is:

[0024]

[0025] Among them, Y is the output of the output layer of the HLICRVFL neural network model during initial training; is the weight of the output layer of the HLICRVFL neural network model after preliminary training;

[0026] The weights and biases of the original nodes in the hidden layer are continuously adjusted until the HLICRVFL neural network model converges to obtain the HLICRVFL neural network model after preliminary training.

[0027] Preferably, in step S4, a new training set is obtained according to the following formula: :

[0028]

[0029] in, represents the training set; Represents the noise amplification ratio coefficient; Represents the probability density function of the Student's t distribution.

[0030] Preferably, in step S5, training the newly added nodes using the new training set includes:

[0031]

[0032]

[0033] in, is the output matrix of the hidden layer node added for the i-th time; is the randomly generated weight matrix of the i-th newly added node; is the bias matrix of the randomly generated i-th newly added node; is the new training set generated for the i-th time; n is the number of nodes added each time; is the weight corresponding to the nth newly added node and the Dth feature; is the bias corresponding to the nth newly added node.

[0034] Preferably, in step S5, the original nodes and the newly added nodes after training are densely linked using a fully connected layer to ensure that the model does not forget historical information, thereby obtaining a HLICRVFL neural network model after secondary training.

[0035] Preferably, in the HLICRVFL neural network model after the secondary training, all nodes in the hidden layer are densely connected, and the input of the i-th newly added node is composed of Together with the output of the i-1th newly added node;

[0036] In the hidden layer of the HLICRVFL neural network model after secondary training, the output matrix of the i-th newly added node is expressed as:

[0037]

[0038] in, is the output matrix of the i-th newly added node; is the activation function of the i-th newly added node.

[0039] Preferably, the output of the output layer of the HLICRVFL neural network model after secondary training is calculated according to the following formula:

[0040]

[0041] in, It is the output of the output layer of the HLICRVFL neural network model after secondary training; is the weight of the output layer of the HLICRVFL neural network model after secondary training.

[0042] The present invention also provides a wind power prediction system based on data enhancement and hidden layer incremental cascading, which applies the above-mentioned wind power prediction method based on data enhancement and hidden layer incremental cascading, including:

[0043] Preprocessing unit: used to collect historical power generation data of wind power clusters for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set, and a test set;

[0044] Model building unit: used to introduce a hidden layer node adding mechanism based on the traditional RVFL neural network model to build a HLICRVFL neural network model;

[0045] A preliminary training unit: used to perform preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training;

[0046] Data enhancement unit: used to add Student's t distribution noise to the training set to generate a new training set;

[0047] Secondary training unit: used to add several nodes to the hidden layer of the HLICRVFL neural network model after preliminary training, and use the new training set to train the newly added nodes, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after secondary training;

[0048] Verification unit: used to determine whether the HLICRVFL neural network model after secondary training converges on the verification set. If so, the training is completed, the optimal HLICRVFL neural network model is obtained, and the model testing unit is executed; otherwise, the data enhancement unit and the secondary training unit are re-executed;

[0049] Model testing unit: used to input the test set into the optimal HLICRVFL neural network model for wind power forecasting and evaluate the prediction effect of the model.

[0050] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0051] The invention provides a wind power prediction method and system based on data enhancement and hidden layer incremental cascade. Firstly, historical power generation data of a target wind power cluster is collected, an initial data set is constructed and preprocessed, and then a training set, a verification set and a test set are divided according to a certain ratio; a HLICRVFL is preliminarily trained using the training set to obtain a preliminary result; then, the training set is data enhanced using the Student t distribution to synthesize a new training set; in the face of newly added training samples, the HLICRVFL model can automatically adjust its structure, adapt to the new data distribution by adding nodes of the hidden layer, while the parameters of the original nodes remain unchanged, thereby achieving rapid learning and adaptation to the new data while ensuring the stability of the model; after the training of the newly added nodes is completed, the newly added nodes are linked to the original nodes obtained by the preliminary training to ensure that the model does not forget the historical data; finally, when the HLICRVFL model converges on the verification set, the iteration is stopped, and the test set is predicted using the optimal model, and the prediction effect is evaluated at the same time; the invention can effectively improve the prediction accuracy of small sample wind power.

[0052] The HLICRVFL model in the present invention can automatically select an appropriate training sample size when facing new training samples, thereby reducing the underfitting or overfitting problems of the model; more importantly, the hidden layer incremental form adopted by HLICRVFL does not require retraining of all parameters, but only requires training of newly added hidden layer nodes, which significantly reduces the computational cost; in addition, the present invention performs data enhancement based on Student's t distribution, which can generate higher quality training samples and further improve the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a wind power prediction method based on data enhancement and hidden layer incremental cascading provided in Example 1.

[0054] Figure 2 This is a flow chart of a wind power prediction method based on data enhancement and hidden layer incremental cascading provided in Example 2.

[0055] Figure 3 This is a structural diagram of the HLICRVFL neural network model provided in Example 2.

[0056] Figure 4 This is a structural diagram of a wind power prediction system based on data enhancement and hidden layer incremental cascading provided in Example 3. DETAILED DESCRIPTION

[0057] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0058] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0059] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0060] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0061] Example 1

[0062] like Figure 1 As shown, this embodiment provides a wind power prediction method based on data enhancement and hidden layer incremental cascading, comprising the following steps:

[0063] S1: Collect historical power generation data of the wind power cluster for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set, and a test set;

[0064] S2: Based on the traditional RVFL neural network model, a hidden layer node addition mechanism is introduced to construct the HLICRVFL neural network model;

[0065] S3: Performing preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training;

[0066] S4: adding Student's t distribution noise to the training set to generate a new training set;

[0067] S5: adding several nodes to the hidden layer of the HLICRVFL neural network model after the initial training, and training the newly added nodes using the new training set, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after the secondary training;

[0068] S6: Determine whether the HLICRVFL neural network model after the secondary training converges on the verification set. If so, complete the training, obtain the optimal HLICRVFL neural network model, and execute step S7; otherwise, re-execute steps S4-S5;

[0069] S7: Input the test set into the optimal HLICRVFL neural network model for wind power forecasting and evaluate the prediction effect of the model.

[0070] In the specific implementation process, the historical power generation data of the target wind power cluster is first collected, the initial data set is constructed and preprocessed, and then the training set, validation set and test set are divided according to a certain ratio;

[0071] In this embodiment, the collected historical data is small sample data. Small sample refers to the situation where the training sample size is small. There is no exact standard for how many samples are considered small. Generally speaking, the definition of small samples is affected by specific problems and research fields. When the sample size is so small that it cannot meet the assumptions such as normal distribution and central limit theorem, it can be regarded as a small sample. For the field of wind power prediction involved in this embodiment, most researchers usually take data sets ranging from 1 to 12 months.

[0072] The training set is used to perform preliminary training on HLICRVFL to obtain preliminary results. Then, the training set is augmented using Student's t distribution to synthesize a new training set.

[0073] Student's t distribution is a probability distribution, which is often used in statistics for hypothesis testing and confidence interval estimation of sample means, and is often used to process statistical inferences in small sample situations. It is worth mentioning that the noise in this embodiment can also be replaced by noise of other distributions, but Student's t distribution has a better modeling effect for small sample situations. When the sample size is small, the effect is usually better than other noises. Therefore, this embodiment introduces noise based on Student's t distribution for small sample prediction scenarios.

[0074] In the face of new training samples, the HLICRVFL model can automatically adjust its structure. Every time new data is generated, a batch of new nodes are generated, and the nodes in the hidden layer are added to adapt to the new data distribution, while the parameters of the original nodes remain unchanged, thus ensuring the stability of the model while achieving rapid learning and adaptation to new data;

[0075] After the training of the newly added nodes is completed, they are linked to the original nodes obtained through the preliminary training to ensure that the model does not forget the historical data and obtain the HLICRVFL neural network model after the secondary training;

[0076] Finally, when the HLICRVFL model converges on the validation set, the iteration is stopped, and the optimal model is used to predict the test set and the prediction effect is evaluated;

[0077] The HLICRVFL model in this method can automatically select the appropriate training sample size when facing new training samples, thereby reducing the underfitting or overfitting problems of the model; secondly, the hidden layer incremental form adopted by HLICRVFL does not need to retrain all parameters, but only needs to train the newly added hidden layer nodes, which significantly reduces the computational cost; in addition, this method performs data enhancement based on Student's t distribution, which can generate higher quality training samples and further improve the prediction accuracy of the model.

[0078] Example 2

[0079] like Figure 2As shown, this embodiment provides a wind power prediction method based on data enhancement and hidden layer incremental cascading, comprising the following steps:

[0080] S1: Collect historical power generation data of the wind power cluster for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set, and a test set;

[0081] S2: Based on the traditional RVFL neural network model, a hidden layer node addition mechanism is introduced to construct the HLICRVFL neural network model;

[0082] S3: Performing preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training;

[0083] S4: adding Student's t distribution noise to the training set to generate a new training set;

[0084] S5: adding several nodes to the hidden layer of the HLICRVFL neural network model after the initial training, and training the newly added nodes using the new training set, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after the secondary training;

[0085] S6: Determine whether the HLICRVFL neural network model after the secondary training converges on the verification set. If so, complete the training, obtain the optimal HLICRVFL neural network model, and execute step S7; otherwise, re-execute steps S4-S5;

[0086] S7: Input the test set into the optimal HLICRVFL neural network model to predict wind power and evaluate the prediction effect of the model;

[0087] In step S1, interpolation processing is performed on the outliers in the initial sample set, and data normalization is performed to complete preprocessing;

[0088] In step S2, the HLICRVFL neural network model includes an input layer, a hidden layer and an output layer connected in sequence; the hidden layer includes a plurality of original nodes;

[0089] The hidden layer node adding mechanism includes: adding a number of nodes to the hidden layer of the HLICRVFL neural network model each time a new training set is generated;

[0090] The step S3 comprises:

[0091] The weights and biases of the original nodes in the hidden layer of the HLICRVFL neural network model are randomly generated, and the output matrix of all the original nodes in the hidden layer is calculated. The calculation process is:

[0092]

[0093]

[0094] Among them, Z is the output matrix of all original nodes in the hidden layer; X is the training set, L is the number of samples in the training set, and D is the number of features in the training set; f is the activation function; W is the weight matrix of the randomly generated original nodes in the hidden layer; B is the bias matrix of the randomly generated original nodes in the hidden layer; m is the number of original nodes in the hidden layer; is the weight corresponding to the mth original node and the Dth feature; is the Dth feature of the Lth sample; is the bias corresponding to the mth original node;

[0095] Calculate the output of the output layer of the HLICRVFL neural network model. The calculation process is:

[0096]

[0097] Among them, Y is the output of the output layer of the HLICRVFL neural network model during initial training; is the weight of the output layer of the HLICRVFL neural network model after preliminary training;

[0098] The weights and biases of the original nodes of the hidden layer are continuously adjusted until the HLICRVFL neural network model converges, and the HLICRVFL neural network model after preliminary training is obtained;

[0099] In step S4, a new training set is obtained according to the following formula: :

[0100]

[0101] in, represents the training set; Represents the noise amplification ratio coefficient; represents the probability density function of the Student's t distribution;

[0102] In step S5, training the newly added nodes using the new training set includes:

[0103]

[0104]

[0105] in, is the output matrix of the hidden layer node added for the i-th time; is the randomly generated weight matrix of the i-th newly added node; is the bias matrix of the randomly generated i-th newly added node; is the new training set generated for the i-th time; n is the number of nodes added each time; is the weight corresponding to the nth newly added node and the Dth feature; is the bias corresponding to the nth newly added node;

[0106] In step S5, the original nodes and the newly added nodes after training are densely linked using a fully connected layer to ensure that the model does not forget historical information, and obtain the HLICRVFL neural network model after secondary training;

[0107] In the HLICRVFL neural network model after the secondary training, all nodes in the hidden layer are densely connected, and the input of the i-th newly added node is composed of Together with the output of the i-1th newly added node;

[0108] In the hidden layer of the HLICRVFL neural network model after secondary training, the output matrix of the i-th newly added node is expressed as:

[0109]

[0110] in, is the output matrix of the i-th newly added node; is the activation function of the newly added node for the i-th time;

[0111] The output of the output layer of the HLICRVFL neural network model after secondary training is calculated according to the following formula:

[0112]

[0113] in, It is the output of the output layer of the HLICRVFL neural network model after secondary training; is the weight of the output layer of the HLICRVFL neural network model after secondary training.

[0114] In the specific implementation process, we first obtain the power generation data of the wind farm within 10 days (i.e., small sample data), interpolate the outliers, and then normalize the data. Then, we divide the collected wind power data into training set, validation set, and test set in a ratio of 5:2:3.

[0115] Then, based on the traditional RVFL neural network model, the hidden layer node addition mechanism is introduced to construct the HLICRVFL neural network model; Figure 3As shown, the HLICRVFL neural network model includes an input layer, a hidden layer and an output layer connected in sequence; the hidden layer includes a number of original nodes; the mechanism for adding nodes in the hidden layer includes: adding a number of nodes in the hidden layer of the HLICRVFL neural network model each time a new training set is generated;

[0116] Then, the training set was used to conduct preliminary training on the hidden layer incremental cascade RVFL neural network (HLICRVFL) model;

[0117] The initial training is to train the original nodes. Specifically, the weights and biases of the original nodes in the hidden layer of the HLICRVFL neural network model are randomly generated, and the output matrix of all the original nodes in the hidden layer is calculated. The calculation process is:

[0118]

[0119]

[0120] Among them, Z is the output matrix of all original nodes in the hidden layer; X is the training set, L is the number of samples in the training set, and D is the number of features in the training set; f is the activation function; W is the weight matrix of the randomly generated original nodes in the hidden layer; B is the bias matrix of the randomly generated original nodes in the hidden layer; m is the number of original nodes in the hidden layer; is the weight corresponding to the mth original node and the Dth feature; is the Dth feature of the Lth sample; is the bias corresponding to the mth original node;

[0121] This embodiment uses the least squares method to calculate the output layer weights, and further calculates the output of the output layer of the HLICRVFL neural network model. The calculation process is:

[0122]

[0123] Among them, Y is the output of the output layer of the HLICRVFL neural network model during initial training; is the weight of the output layer of the HLICRVFL neural network model after preliminary training;

[0124] The weights and biases of the original nodes of the hidden layer are continuously adjusted until the HLICRVFL neural network model converges, and the HLICRVFL neural network model after preliminary training is obtained;

[0125] Then add Student t distribution noise to the training set to generate a new training set :

[0126]

[0127] in, represents the training set; Represents the noise amplification ratio coefficient; represents the probability density function of the Student's t distribution;

[0128] Student's t distribution is a statistical inference suitable for small sample data. The probability density function calculation formula of this method is as follows:

[0129]

[0130] Then, several nodes are added to the hidden layer of the HLICRVFL neural network model after preliminary training, and the newly added nodes are trained using the new training set, while the parameters of the original nodes remain unchanged. This embodiment uses the new training set to train the newly added nodes, including:

[0131]

[0132]

[0133] in, is the output matrix of the hidden layer node added for the i-th time; is the randomly generated weight matrix of the i-th newly added node; is the bias matrix of the randomly generated i-th newly added node; is the new training set generated for the i-th time; n is the number of nodes added each time; is the weight corresponding to the nth newly added node and the Dth feature; is the bias corresponding to the nth newly added node;

[0134] After the training of the newly added nodes is completed, the fully connected layer is used to densely link the original nodes and the newly added nodes after training to ensure that the model does not forget historical information and obtain the HLICRVFL neural network model after secondary training;

[0135] In the HLICRVFL neural network model after secondary training, all nodes in the hidden layer are densely connected, and the input of the i-th newly added node is composed of Together with the output of the i-1th newly added node;

[0136] In the hidden layer of the HLICRVFL neural network model after secondary training, the output matrix of the i-th newly added node is expressed as:

[0137]

[0138] in, is the output matrix of the i-th newly added node; is the activation function of the newly added node for the i-th time;

[0139] Similarly, based on the least squares method, the output of the output layer of the HLICRVFL neural network model after secondary training is calculated according to the following formula:

[0140]

[0141] in, It is the output of the output layer of the HLICRVFL neural network model after secondary training; is the weight of the output layer of the HLICRVFL neural network model after secondary training;

[0142] In this embodiment, each time the model is iterated, a new node is added; Figure 3 As shown in the figure, the bottom rectangle in the hidden layer is the i-th newly added node, and the number of nodes is n; the ellipsis above is the i-1-th newly added node, and the first matrix up is the original node; at the same time, each newly added node is also connected, as shown in Figure 3 The form in the upper left corner (the original node is connected to the first newly added node, the first newly added node is connected to the second newly added node, and so on until the i-th time, shown in the figure as connected from top to bottom);

[0143] Then, it is determined whether the HLICRVFL neural network model after the second training converges on the validation set. If it converges, the training is completed to obtain the optimal HLICRVFL neural network model. The test set is input into the optimal HLICRVFL neural network model for wind power prediction and the prediction effect of the model is evaluated. If it does not converge, a new training set is regenerated, and nodes are added and training is continued.

[0144] Finally, the HLICRVFL model found the most suitable network structure and model parameters, built an adaptive hidden layer incremental cascade RVFL neural network for small sample data enhancement, and used the optimal model to predict the power time series of the target wind farm at future times, which can effectively improve the wind power prediction accuracy.

[0145] The HLICRVFL model in this method can automatically select the appropriate training sample size when facing new training samples, thereby reducing the underfitting or overfitting problems of the model; secondly, the hidden layer incremental form adopted by HLICRVFL does not need to retrain all parameters, but only needs to train the newly added hidden layer nodes, which significantly reduces the computational cost. At the same time, while ensuring the stability of the model, it achieves rapid learning and adaptation to new data; in addition, this method performs data enhancement based on Student's t distribution, which can generate higher quality training samples and further improve the prediction accuracy of the model.

[0146] Example 3

[0147] like Figure 4 As shown, this embodiment provides a wind power prediction system based on data enhancement and hidden layer incremental cascading, and applies the wind power prediction method based on data enhancement and hidden layer incremental cascading described in Embodiment 1 or 2, including:

[0148] Preprocessing unit 301: used to collect historical power generation data of the wind power cluster for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set and a test set;

[0149] Model building unit 302: used to introduce a hidden layer node adding mechanism based on the traditional RVFL neural network model to build a HLICRVFL neural network model;

[0150] A preliminary training unit 303 is used to perform preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training;

[0151] Data enhancement unit 304: used to add Student's t distribution noise to the training set to generate a new training set;

[0152] Secondary training unit 305: used to add several nodes to the hidden layer of the HLICRVFL neural network model after the initial training, and use the new training set to train the newly added nodes, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after the secondary training;

[0153] Verification unit 306: used to determine whether the HLICRVFL neural network model after secondary training converges on the verification set. If so, the training is completed to obtain the optimal HLICRVFL neural network model and execute the model testing unit 307; otherwise, the data enhancement unit 304 and the secondary training unit 305 are re-executed;

[0154] Model testing unit 307: used for inputting the test set into the optimal HLICRVFL neural network model to perform wind power prediction and evaluate the prediction effect of the model.

[0155] In the specific implementation process, firstly, the preprocessing unit 301 collects the historical power generation data of the target wind power cluster, constructs the initial data set and performs preprocessing, and then divides the training set, the validation set and the test set according to a certain ratio;

[0156] The model building unit 302 introduces a hidden layer node adding mechanism based on the traditional RVFL neural network model to build a HLICRVFL neural network model;

[0157] Then the preliminary training unit 303 performs preliminary training on HLICRVFL using the training set to obtain preliminary results; then the data enhancement unit 304 performs data enhancement on the training set using Student's t distribution to synthesize a new training set;

[0158] In the secondary training unit 305, in the face of newly added training samples, the HLICRVFL model can automatically adjust its structure to adapt to the new data distribution by adding nodes in the hidden layer, while the parameters of the original nodes remain unchanged, thereby achieving rapid learning and adaptation to new data while ensuring the stability of the model;

[0159] After the training of the newly added nodes is completed, they are linked to the original nodes obtained through the preliminary training to ensure that the model does not forget the historical data and obtain the HLICRVFL neural network model after the secondary training;

[0160] Finally, the verification unit 306 is used to verify whether the model after the second training has converged. When the HLICRVFL model converges on the verification set, the iteration stops. The model testing unit 307 uses the optimal model to predict the test set and evaluates the prediction effect.

[0161] The HLICRVFL model in this system can automatically select the appropriate training sample size when faced with new training samples, thereby reducing the underfitting or overfitting problems of the model; secondly, the hidden layer incremental form adopted by HLICRVFL does not require retraining of all parameters, but only requires training of newly added hidden layer nodes, which significantly reduces the computational cost; in addition, this system performs data enhancement based on Student's t distribution, which can generate higher quality training samples and further improve the prediction accuracy of the model.

[0162] The same or similar reference numerals correspond to the same or similar components;

[0163] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0164] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A wind power prediction method based on data enhancement and hidden layer incremental cascading, characterized in that: The following steps are involved: S1: Collect historical power generation data of the wind power cluster for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set, and a test set; S2: Based on the traditional RVFL neural network model, a hidden layer node addition mechanism is introduced to construct the HLICRVFL neural network model; The HLICRVFL neural network model includes an input layer, a hidden layer and an output layer connected in sequence; the hidden layer contains a number of original nodes; The hidden layer node adding mechanism includes: adding a number of nodes to the hidden layer of the HLICRVFL neural network model each time a new training set is generated; S3: Performing preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training; S4: adding Student's t distribution noise to the training set to generate a new training set; S5: adding several nodes to the hidden layer of the HLICRVFL neural network model after the initial training, and training the newly added nodes using the new training set, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after the secondary training; S6: Determine whether the HLICRVFL neural network model after the secondary training converges on the verification set. If so, complete the training, obtain the optimal HLICRVFL neural network model, and execute step S7; otherwise, re-execute steps S4-S5; S7: Input the test set into the optimal HLICRVFL neural network model for wind power forecasting and evaluate the prediction effect of the model.

2. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 1, characterized in that: In step S1, interpolation processing is performed on the abnormal values ​​in the initial sample set, and data normalization is performed to complete preprocessing.

3. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 1, characterized in that: The step S3 comprises: Randomly generate the weights and biases of the original nodes in the hidden layer of the HLICRVFL neural network model, and calculate the output matrix of all the original nodes in the hidden layer. The calculation process is: Among them, Z is the output matrix of all original nodes in the hidden layer; X is the training set, L is the number of samples in the training set, and D is the number of features in the training set; f is the activation function; W is the weight matrix of the randomly generated original nodes in the hidden layer; B is the bias matrix of the randomly generated original nodes in the hidden layer; m is the number of original nodes in the hidden layer; is the weight corresponding to the mth original node and the Dth feature; is the Dth feature of the Lth sample; is the bias corresponding to the mth original node; Calculate the output of the output layer of the HLICRVFL neural network model. The calculation process is: Among them, Y is the output of the output layer of the HLICRVFL neural network model during initial training; is the weight of the output layer of the HLICRVFL neural network model after preliminary training; The weights and biases of the original nodes in the hidden layer are continuously adjusted until the HLICRVFL neural network model converges to obtain the HLICRVFL neural network model after preliminary training.

4. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 3, characterized in that: In step S4, a new training set is obtained according to the following formula: : in, represents the training set; Represents the noise amplification ratio coefficient; Represents the probability density function of the Student's t distribution.

5. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 4, characterized in that: In step S5, training the newly added nodes using the new training set includes: in, is the output matrix of the hidden layer node added for the i-th time; is the randomly generated weight matrix of the i-th newly added node; is the bias matrix of the randomly generated i-th newly added node; is the new training set generated for the i-th time; n is the number of nodes added each time; is the weight corresponding to the nth newly added node and the Dth feature; is the bias corresponding to the nth newly added node.

6. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 5, characterized in that: In step S5, the original nodes and the newly added nodes after training are densely linked using a fully connected layer to ensure that the model does not forget historical information, and the HLICRVFL neural network model after secondary training is obtained.

7. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 6, characterized in that: In the HLICRVFL neural network model after the secondary training, all nodes in the hidden layer are densely connected, and the input of the i-th newly added node is composed of Together with the output of the i-1th newly added node; In the hidden layer of the HLICRVFL neural network model after secondary training, the output matrix of the i-th newly added node is expressed as: in, is the output matrix of the i-th newly added node; is the activation function of the i-th newly added node.

8. A wind power prediction method based on data enhancement and hidden layer incremental cascading according to claim 7, characterized in that: The output of the output layer of the HLICRVFL neural network model after secondary training is calculated according to the following formula: in, It is the output of the output layer of the HLICRVFL neural network model after secondary training; is the weight of the output layer of the HLICRVFL neural network model after secondary training.

9. A wind power prediction system based on data enhancement and hidden layer incremental cascading, applying the wind power prediction method based on data enhancement and hidden layer incremental cascading as described in any one of claims 1 to 8, characterized in that: include: Preprocessing unit: used to collect historical power generation data of wind power clusters for several days, construct an initial sample set and perform preprocessing, and divide the preprocessed initial sample set into a training set, a validation set, and a test set; Model building unit: used to introduce a hidden layer node adding mechanism based on the traditional RVFL neural network model to build a HLICRVFL neural network model; A preliminary training unit: used to perform preliminary training on the HLICRVFL neural network model using the training set to obtain the HLICRVFL neural network model after preliminary training; Data enhancement unit: used to add Student's t distribution noise to the training set to generate a new training set; Secondary training unit: used to add several nodes to the hidden layer of the HLICRVFL neural network model after preliminary training, and use the new training set to train the newly added nodes, while the parameters of the original nodes remain unchanged, to obtain the HLICRVFL neural network model after secondary training; Verification unit: used to determine whether the HLICRVFL neural network model after secondary training converges on the verification set. If so, the training is completed, the optimal HLICRVFL neural network model is obtained, and the model testing unit is executed; otherwise, the data enhancement unit and the secondary training unit are re-executed; Model testing unit: used to input the test set into the optimal HLICRVFL neural network model for wind power forecasting and evaluate the prediction effect of the model.

Images