CN110570030A - Method and system for wind power cluster power interval prediction based on deep learning - Google Patents

Abstract

This disclosure provides a method and system for predicting the power range of wind power clusters based on deep learning. The numerical weather forecast and historical wind power of each wind farm station are obtained as the original input data, and the interpretation in the region is extracted by calculating the mutual information of explanatory variables. The mutual information between the variable and the target variable is used to extract the correlation information, the explanatory variables that meet the correlation degree are selected, the principal component analysis method is used for data reconstruction and dimensionality reduction, the interval constraints are constructed, and the deep learning is used to construct the prediction model. And dimensionality reduction data input model for training, combined with particle swarm optimization method for model optimization, to determine the final prediction model, using the final prediction model for power interval prediction, with high accuracy.

Description

Method and system for wind power cluster power interval prediction based on deep learning

technical field

The disclosure belongs to the field of wind power forecasting, and in particular relates to a method and system for forecasting a power interval of a wind power cluster based on deep learning.

Background technique

The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

The environmental problems caused by the massive combustion of fossil fuels and energy depletion have attracted more and more global attention, and it has become the consensus of all countries to vigorously develop renewable and clean energy. However, unlike the strong controllability of traditional energy sources, wind power is intermittent and random. Therefore, a high proportion of wind power connected to the grid poses severe challenges to the economical, safe and stable operation of the power system. Accurate and reliable wind power forecasting results are one of the important means to solve this problem.

A wind power cluster mainly refers to the collection of multiple wind farms in a region. In recent years, most of the research has focused on the prediction of the output of a single wind farm station, and there are relatively few predictions of the power of wind power clusters. In fact, the research on wind power prediction has been going on for many years. According to the expression form of the results, it can be divided into single value prediction and probability prediction. Some single-valued forecasting methods have been applied in the field of wind power forecasting. Although some of these methods can obtain more accurate prediction results, there is a problem that cannot be ignored in single-value prediction, namely: due to the lack of data and the fluctuation characteristics of wind power itself, single-value prediction will inevitably introduce prediction errors, and the definite prediction results cannot be provided. Uncertainty information about wind power. This makes the use of wind power prediction results in the decision-making process based on stochastic optimization or risk assessment has certain limitations.

In order to describe the randomness and variability of wind power, wind power probability prediction technology has developed rapidly in the past, and scholars from various countries have proposed many probability prediction methods, such as quantile point regression, conditional kernel density estimation, interval prediction and sparse Bayesian Sri Lankan learning methods, etc. Compared with deterministic methods, probabilistic forecasting can provide more information about wind power uncertainty to meet the needs of different decision-making objectives. So far, wind power probabilistic forecasting technology has been applied to formulate power generation plan, backup configuration, optimal unit combination and power market, etc. and achieved good results. However, as far as the inventors know, most current probabilistic forecasting studies only focus on a single wind farm station, and only use numerical weather prediction (NWP) and historical data of local wind farms to predict wind power. There must be a certain correlation between the stations, and the effective use of the correlation can significantly improve the power probability prediction results of the wind power cluster.

Contents of the invention

In order to solve the above problems, this disclosure proposes a method and system for predicting the power range of wind power clusters based on deep learning. This disclosure directly uses the original data of each station to predict the power of wind power clusters, and calculates the area The mutual information between the internal explanatory variable and the target variable is used to extract the relevant information, and the highly correlated explanatory variables are selected, and then the principal component analysis method is used for data reconstruction and dimensionality reduction to improve the efficiency of probability prediction. Finally, the interval constraints are constructed, the prediction model is constructed using deep learning, and the model is optimized with the particle swarm optimization method, which has a certain degree of advancement, accuracy and effectiveness.

According to some embodiments, the present disclosure adopts the following technical solutions:

A method for predicting the power range of wind power clusters based on deep learning, comprising the following steps:

Obtain the numerical weather forecast and historical wind power of each wind farm station as the original input data, and extract the mutual information between the explanatory variables and the target variables in the region by calculating the mutual information of the explanatory variables to extract the correlation information, and select the interpretation that meets the correlation degree Variables, use the principal component analysis method for data reconstruction and dimensionality reduction, construct interval constraints, use deep learning to build a prediction model, input the reconstructed and dimensionality reduction data into the model for training, combine the particle swarm optimization method for model optimization, and determine The final prediction model, using the final prediction model for power interval prediction.

Based on the obtained interval prediction results, the configuration of the backup units in the wind power cluster can be carried out, including capacity configuration and location configuration.

It is also possible to determine the optimal unit combination for the design of wind power clusters or the construction of wind power clusters to ensure the orderliness, safety and efficiency of power consumption.

As an optional implementation, the total power of the wind power cluster is used as the target variable, and the NWP data and historical measurement data of each wind farm station in the cluster are used as explanatory variables to calculate the mutual information between the explanatory variables and the target variable. The law calculates the mutual information from the samples, and by calculating the mutual information between the explanatory variables and the target variable, selects a set of explanatory variables that are most correlated with the target variable.

As an optional implementation, the data is uniformly normalized so that the data is between [0,1], the selection of explanatory variables is calculated to calculate the mutual information, and the historical wind power, irradiance, Temperature and humidity variables were used as explanatory variables.

As an optional implementation, the wind power cluster covers multiple stations, and principal component analysis is used to reduce the dimensionality of explanatory variables, and at the same time extract key features of explanatory variables, so that the input data of deep learning are independent of each other.

As an optional implementation, deep learning is used to build a prediction model, the interval prediction is converted into a multi-objective optimization problem, and deep learning is used for optimization to seek the optimal weight.

As an optional implementation, the goal of the multi-objective optimization is to minimize the width of the prediction interval under a given prediction interval coverage.

As an optional implementation, the fitness value of each particle is calculated according to the fitness function, and for each particle, its fitness value is compared with the optimal fitness value of its history record, and if it is better, its As the historical optimal, compare its fitness value with the fitness value of the best position experienced by the group, and if it is better, it will be regarded as the group optimal.

A system for forecasting power intervals of wind power clusters based on deep learning, including:

The data processing module obtains the numerical weather forecast and historical wind power of each wind farm station as the original input data, and extracts the mutual information between the explanatory variables and the target variables in the region by calculating the mutual information of the explanatory variables to extract the associated information;

The dimensionality reduction module is configured to select explanatory variables that meet the correlation degree, and use the principal component analysis method to perform data reconstruction and dimensionality reduction;

The model building module is configured to build interval constraints, use deep learning to build a prediction model, input the reconstructed and dimensionally reduced data into the model for training, combine the particle swarm optimization method to optimize the model, determine the final prediction model, and use the final The prediction model performs power interval prediction.

A computer-readable storage medium, in which a plurality of instructions are stored, and the instructions are suitable for being loaded by a processor of a terminal device and executing the method for predicting a power interval of a wind power cluster based on deep learning.

A terminal device, including a processor and a computer-readable storage medium, the processor is used to implement instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the described one A method for wind power cluster power interval prediction based on deep learning.

Compared with the prior art, the beneficial effects of the present disclosure are:

Compared with the prediction of the output of a single wind farm station, the power prediction of wind power clusters can directly provide information for power system decision makers, and then formulate reasonable power generation plans and make backup plans. At the same time, the reliance on station forecasts is reduced, thereby avoiding the generation of curtailed wind. According to the power forecasting of wind power clusters, there are many explanatory variables and the large and complex data volume, mutual information and principal component analysis are used to reduce the dimensionality of the initial data, and the probabilistic interval prediction is transformed into an optimization problem with constraints, and deep learning is used to mine Non-linear relationship, accurate prediction results can be obtained.

Description of drawings

The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure, and the exemplary embodiments and descriptions of the present disclosure are used to explain the present disclosure, and do not constitute improper limitations to the present disclosure.

Figure 1 is a schematic diagram of the basic structure of deep learning;

Figure 2 is a flowchart of interval prediction;

Figure 3 is a schematic diagram of the mutual information arrangement of wind farm stations in the cluster;

Figure 4 is a schematic diagram of the selection method of key explanatory variables;

Figure 5 is a comparison chart of the width of the prediction interval;

Figure 6 is a schematic diagram of interval prediction results based on deep learning;

Fig. 7 is a schematic diagram of interval prediction results of raw data input.

Detailed ways:

The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

It should be noted that the terminology used herein is only for describing specific embodiments, and is not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural, and it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they mean There are features, steps, operations, means, components and/or combinations thereof.

A short-term wind power cluster power interval prediction method based on deep learning, which directly uses the original data of each station to predict the power of wind power clusters. First, on the basis of the initial data, by calculating the mutual information of the explanatory variables to extract the mutual information between the explanatory variables and the target variable in the region to extract the correlation information, select highly correlated explanatory variables, and then use the principal component analysis method to reconstruct the data and dimensionality reduction to improve probability prediction efficiency. Finally, the interval constraints are constructed, the prediction model is constructed using deep learning, and the model is optimized with the particle swarm optimization method.

The wind power cluster power prediction is different from the single-site forecasting technology, which contains the initial data of multiple wind farms, so the data volume is large, the explanatory variables are many and complex, and there must be correlations between the sites. Therefore, in order to improve the power prediction accuracy of wind power clusters, the following two factors must be considered:

First, get data quality. Detailed geographic information, weather conditions, and historical measurement data of wind power cluster sites can improve prediction accuracy. In fact, the data integrity of most wind farm stations is incomplete, and multiple wind farm stations must correspond one-to-one at data collection time points, so the quality of wind power cluster data acquisition is even more difficult to guarantee. Therefore, data types should not be overly relied upon during correlation analysis and predictive modeling. Considering that the numerical weather forecast and historical wind power are the data that must be stored, and the correlation between the stations can be analyzed through these two types of data. Therefore, in this embodiment, it is most reasonable to select the numerical weather forecast and historical wind power of each wind farm station in the cluster as the original input data.

Second, is the input data dimension. Theoretically speaking, enough input data is conducive to improving the prediction accuracy, but at the same time, it will also bring about increased computational pressure and difficulty in model estimation. As the data dimension increases, the computational efficiency of most algorithms decreases and the computational time increases. In order to solve this problem, we first use mutual information to conduct correlation analysis on all input data, and select highly correlated explanatory variables. The data were then reconstructed and dimensionally reduced again using principal component analysis methods. Finally, the relevant data is extracted and input into the model.

Shannon's information theory covers the concepts of information entropy and mutual information. The information entropy of any random variable is the amount of information contained in this variable. The information entropy of random variable variable X can be expressed by formula (1):

H(X)＝∫-f _X log(f _X ) (1)

where f _X is the probability density function of the variable X.

Information entropy is often used to measure the information content of a physical or artificial system. Mutual information is based on information entropy, which is a measure of useful information, which can be understood as the amount of information contained in a random variable about another random variable, that is, the reduction of a random variable due to the known other random variable Uncertainty. So we can use mutual information to measure the degree of association between two random variables. For random variables X and Y, the average mutual information can be expressed by formula (2):

Among them, f _Y is the probability density function of random variable Y, and f _{X, Y} is the joint probability density function of random variables X and Y.

It can be seen from the formula (2) that when f _X,Y = f _X f _Y , this means that the random variables X and Y are independent of each other, That is, I(X;Y) is equal to zero. Conversely, if I(X;Y) is greater than zero, it indicates that there is an association between the two variables. The higher the mutual information value, the stronger the correlation between variables.

The correlation coefficient method can also be used to calculate the correlation. However, the correlation coefficient can only reflect the linear relationship between variables, and the wind farm station is a complex man-made system, and the nonlinear relationship between its data is particularly prominent. Mutual information can not only It can reflect the linear relationship and also reflect its nonlinear relationship, so the mutual information is more comprehensive in the correlation between the response variables than the correlation coefficient.

The total power of the wind power cluster is used as the target variable, and the NWP data and historical measurement data of each wind farm station in the cluster are used as the explanatory variables, and then the mutual information between the explanatory variables and the target variables is calculated. To simplify the calculation, the law of large numbers can be used Compute mutual information from samples:

Select a set of explanatory variables that are highly correlated with the target variable by computing the mutual information between the explanatory variables and the target variable.

There are relatively many explanatory variables selected for wind power cluster power forecasting, and the data dimension is high. At the same time, among the explanatory variables extracted through mutual information, there must be some variables that contain redundant information. Principal component analysis can effectively extract key explanatory variables and main features, reduce data dimensions, and make key explanatory variables independent of each other and as many as possible. React for more information.

Principal component analysis first needs to standardize the explanatory variables, assuming m explanatory variables X ₁ , X ₂ , X ₃ ,…, X _m to represent the characteristics of the target variable, the number of samples is N, which can be represented by an N×m matrix, namely

Then its center normalization is:

in, and s _j are the mean and variance of the explanatory variable x _j .

Calculate the autocorrelation matrix of the explanatory variables from the obtained center normalization matrix:

here is a centrally standardized N×m sample matrix, where N is the number of samples, m is the number of variables, and R is the autocorrelation matrix. Calculate the m eigenvalues λ ₁ >λ ₂ >,…,>λ _m of the autocorrelation matrix R and the corresponding eigenvectors P.

Compute the variance contribution and cumulative variance contribution for each eigenvector:

Here i=1,2,...,m.

If the cumulative variance contribution rate of the first p eigenvectors is greater than 85%-95%, the number of principal components is determined as p. At this time, the selected principal components already contain most of the information that the original variables can provide.

Generally speaking, the evaluation of interval prediction results mainly uses prediction interval coverage (PICP) and interval width (PINRW) as evaluation indicators. In this embodiment, evaluation indicators are used as optimization conditions, and deep learning is used for target training.

Evaluation Indicators for Interval Prediction

(1) Prediction Interval Coverage (PICP)

Generally speaking, the prediction interval coverage is used to evaluate the reliability of the model and is one of the important evaluation indicators for interval prediction. The prediction interval coverage can be expressed as:

Where N is the sample size and _εt is a Boolean variable used to represent the relationship between the prediction interval and the target value.

Its relationship is specifically expressed as:

Among them, L _t and U _t predict the lower limit of the interval and the upper limit of the prediction interval, respectively.

Therefore, ideally, the target value should be fully covered by the prediction interval, that is, PICP=100%.

(2) The width of the prediction interval (PINRW)

In addition to PICP can evaluate the quality of the prediction interval, the width of the prediction interval must also be considered. Suppose we take the minimum and maximum values of the target variable as the upper and lower limits of the prediction interval. Although the target variable can be well wrapped in it, the width of the prediction interval is too large, which is of little reference value to decision makers. Therefore, the width of the prediction interval is an important indicator to measure the sensitivity of interval prediction. Its expression is as follows:

Among them, R represents the difference between the maximum value and the minimum value of the target variable. The smaller the PINAW, the higher the sensitivity of the predictive model.

The deep learning method has the characteristics of distributed parallel processing, adaptive learning, nonlinear mapping and generalization ability, and has strong adaptability to wind power cluster prediction. Deep learning is a supervised learning model. Its structure is generally shown in Figure 1.

In interval forecasting, we always pursue the largest PICP and the smallest PINAW, that is, higher reliability and better acuity. Therefore, in this embodiment, interval prediction is converted into a multi-objective optimization problem, and deep learning is used for optimization to seek optimal weights.

If PICP is given in advance, then equation (12) can be transformed into a single-objective optimization problem:

The particle swarm optimization algorithm (PSO) is used for target optimization. The flow chart of interval prediction is shown in Figure 2.

According to the flow chart in Figure 2, the prediction model process of this embodiment can be summarized as follows:

1) Data preprocessing. First, the data is uniformly normalized so that the data is between [0,1]. The selection of explanatory variables mainly calculates the mutual information, and selects variables such as historical wind power, irradiance, temperature, and humidity as explanatory variables through the size of the mutual information. At the same time, the mutual information is used to calculate the correlation information.

2) Data dimensionality reduction. The wind power cluster covers multiple stations, so the input data dimension is too large, and the prerequisite for using deep learning to model is that the input data are independent of each other. Therefore, principal component analysis can not only reduce the dimensionality of explanatory variables, but also extract key features of explanatory variables, making the input data of deep learning independent of each other.

3) Build a deep learning model. Optimize the structure of the neural network, determine the number of hidden layers and nodes. Construct the optimization goal, given PICP, make the PINAW value minimum.

4) Neural network weights and particle swarm algorithm parameters are initialized. The parameter initialization of PSO algorithm includes particle position and velocity. The particle position is represented by the weight of the neural network, and the velocity is randomly initialized.

5) Calculate the fitness value of each particle according to the fitness function. For each particle, compare its fitness value with the best fitness value of its historical record, if it is better, it will be regarded as the best in history (pbest); compare its fitness value with the best fitness value experienced by the group The fitness value of the best position is compared, and if it is better, it is regarded as the group best (gbest).

6) Training end: The training end criterion can be set as the maximum number of iterations, otherwise, the training process will continue and return to step 5.

7) Test evaluation.

As a verification, the data of 10 wind farm stations in a certain area of China are mainly used to predict the total power of the wind power cluster with a time resolution of 15 minutes in the next 72 hours, and the prediction interval is 80% and 90%. The dataset is divided into training set and validation set. The training set is used to build an interval prediction model for deep learning, and the validation set is used to test the performance of the model.

The power of each wind farm station is processed into cluster power, and the mutual information between the target variable and the initial explanatory variable is calculated from the sample data, and the mutual information is shown in Figure 3. The explanatory variables whose mutual information is greater than 0.45 are selected as the key explanatory variables. Arranged in order of size as shown in Figure 4.

After selecting key explanatory variables through mutual information, principal component analysis is used to reduce the dimensionality of key explanatory variables and extract key features to make them independent of each other. Calculate the contribution rate and cumulative contribution rate of each feature, and extract the principal components, as shown in Table 1.

Table 1 Principal component analysis results

Generally speaking, when the cumulative contribution rate of variance reaches 80% to 95%, we consider it as the main component. Therefore, in this embodiment, the first 12 feature variables are selected as key features.

In this embodiment, key features obtained through mutual information and principal component analysis are used as input data of the neural network. Perform sample training. In the comparison example, the original data without data preprocessing is used for sample training, and 80% and 90% prediction intervals are obtained respectively. The 80% and 90% interval widths of the prediction time points are shown in Figure 5.

It can be seen from FIG. 5 that whether it is an 80% prediction interval or a 90% prediction interval, the method proposed in this embodiment has a narrower prediction interval than the result obtained from the original data input. This shows that the prediction results obtained by the method proposed in this embodiment have better sensitivity and can provide more reliable and comprehensive information for decision makers. In addition, with the increase of the forecast time, the width of the forecast interval increases obviously. This is because with the extension of the time scale, the obtained meteorological data become more and more unreliable, and the uncertainty information increases, which makes the forecast effect change. Difference.

Fig. 6 and Fig. 7 respectively represent the prediction results of the 90% and 80% confidence intervals of the total power of the wind power cluster for a certain three days obtained from the original data input and feature extraction. Comparing Figure 6 and Figure 7, it can be seen that the acuity of the prediction result after feature extraction is better, and the uncertain information contained is more comprehensive. As the forecast time scale increases, the confidence interval bands get wider, just demonstrating the results of PINAW changes in Fig. 5. At the same time, we can also see that the prediction effect in Figure 6 is significantly better than that in Figure 7, which shows that the method proposed in this embodiment has good applicability.

Those skilled in the art should understand that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Accordingly, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The above descriptions are only preferred embodiments of the present disclosure, and are not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present disclosure shall be included within the protection scope of the present disclosure.

Although the specific implementation of the present disclosure has been described above in conjunction with the accompanying drawings, it does not limit the protection scope of the present disclosure. Those skilled in the art should understand that on the basis of the technical solutions of the present disclosure, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present disclosure.

Claims (10)

1. A method for prediction of wind power cluster power intervals based on deep learning, characterized in that: comprising the following steps: Obtain the numerical weather forecast and historical wind power of each wind farm station as the original input data, and extract the mutual information between the explanatory variables and the target variables in the region by calculating the mutual information of the explanatory variables to extract the correlation information, and select the interpretation that meets the correlation degree Variables, use the principal component analysis method for data reconstruction and dimensionality reduction, construct interval constraints, use deep learning to build a prediction model, input the reconstructed and dimensionality reduction data into the model for training, combine the particle swarm optimization method for model optimization, and determine The final prediction model, using the final prediction model for power interval prediction. 2. A method for predicting the power range of wind power clusters based on deep learning as shown in claim 1, characterized in that: the total power of wind power clusters is used as the target variable, and the weather forecast data and historical quantities of each wind farm station in the cluster are The measured data is used as an explanatory variable, and the mutual information between the explanatory variable and the target variable is calculated, and the mutual information is calculated from the sample by using the law of large numbers. explanatory variables. 3. A method for predicting a power interval of a wind power cluster based on deep learning as shown in claim 1, characterized in that: the data is uniformly normalized so that the data is between [0,1], and the explanatory variables Mutual information is calculated through the selection of mutual information, and historical wind power, irradiance, temperature and humidity variables are selected as explanatory variables through the size of mutual information. 4. A method for predicting power intervals of wind power clusters based on deep learning as shown in claim 1, characterized in that: the wind power clusters cover a plurality of stations, use principal component analysis to reduce the dimensionality of explanatory variables, and simultaneously extract explanations The key characteristics of variables make the input data of deep learning independent of each other. 5. a kind of method for the interval prediction of wind power cluster power based on deep learning as shown in claim 1, it is characterized in that: utilize deep learning to construct prediction model, interval prediction is converted into multi-objective optimization problem, utilize deep learning to optimize, Find the optimal weight. 6. A method for predicting a power interval of a wind power cluster based on deep learning as claimed in claim 1, characterized in that: the goal of multi-objective optimization is to minimize the width of the prediction interval under a given prediction interval coverage. 7. the method for a kind of wind power cluster power interval prediction based on deep learning as shown in claim 1, it is characterized in that: calculate the fitness value of each particle according to fitness function, for each particle, its fitness value Compared with the best fitness value of its historical record, if it is better, it will be regarded as the best in history, and its fitness value will be compared with the fitness value of the best position experienced by the group, if it is better, then Take it as the group optimal. 8. A system for predicting power intervals of wind power clusters based on deep learning, characterized by: comprising: The data processing module obtains the numerical weather forecast and historical wind power of each wind farm station as the original input data, and extracts the mutual information between the explanatory variables and the target variables in the region by calculating the mutual information of the explanatory variables to extract the associated information; The dimensionality reduction module is configured to select explanatory variables that meet the correlation degree, and use the principal component analysis method to perform data reconstruction and dimensionality reduction; The model building module is configured to build interval constraints, use deep learning to build a prediction model, input the reconstructed and dimensionally reduced data into the model for training, combine the particle swarm optimization method to optimize the model, determine the final prediction model, and use the final The prediction model performs power interval prediction. 9. A computer-readable storage medium, characterized in that: a plurality of instructions are stored therein, and the instructions are suitable for being loaded and executed by a processor of a terminal device according to any one of claims 1-7 based on A method for predicting power intervals of wind power clusters based on deep learning. 10. A terminal device, characterized in that: it includes a processor and a computer-readable storage medium, the processor is used to implement instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and Executing the method for predicting the power interval of a wind power cluster based on deep learning described in any one of claims 1-7.