Ultra-short-term wind speed prediction method, system, device and medium based on deep learning

Abstract

The present invention discloses a method, system, device and medium for ultra-short-term wind speed prediction based on deep learning, including: obtaining unit data and meteorological data, and dividing the unit data and meteorological data according to seasonal differences to obtain data sets of different seasons; dividing the unit data, obtaining different working conditions, and sampling the data sets of different working conditions and different seasons; obtaining a test set and a training set; constructing different network structures based on different characteristics of the unit data and meteorological data, and performing feature fusion to build the final network structure; training the final network structure based on the training set and the test set to determine the optimal model; dividing the data to be predicted and inputting it into the optimal model to obtain the wind speed prediction result. The present invention can control extreme wind conditions in advance, reduce the load impact of wind turbines under extreme wind conditions, reduce the operating risk of wind turbines, increase the operating time of wind turbines, and extend the service life of wind turbines.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep learning and relates to an ultra-short-term wind speed prediction method, system, device and medium based on deep learning. Background Art

[0002] The wind power prediction system of wind turbines can be divided into short-term, medium-term and long-term predictions according to the prediction time scale. Among them, the short-term prediction is to predict the wind power in the next few hours, which is mainly used for the stability of wind power access to the grid and the real-time operation of the power grid, reducing the impact of random fluctuations in wind power on the power grid; the medium-term prediction is to predict the wind power in the next few hours to a week, which is used for the comprehensive deployment of power generation systems and energy storage systems; the long-term prediction is based on weeks, months and years, and is used for the preliminary construction of wind farms and wind resource exploration. There are relatively few studies on ultra-short-term wind speed predictions such as minute-level and second-level wind speed predictions, which can be used for feedforward control of wind turbines. At present, the control strategy of wind turbines is a passive control method with hysteresis, which cannot meet the needs of advance control of extreme wind conditions, affecting the safety of wind turbines. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that the prior art cannot meet the demand for advance control of extreme wind conditions, affecting the safety of wind turbines, and to provide an ultra-short-term wind speed prediction method, system, device and medium based on deep learning.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] Ultra-short-term wind speed prediction method based on deep learning, including:

[0006] Obtain unit data and meteorological data, and divide the unit data and meteorological data according to seasonal differences to obtain data sets for different seasons;

[0007] Divide the unit data to obtain different operating conditions, and sample the data sets of different operating conditions and seasons; use the extracted data as the test set and the remaining data as the training set;

[0008] Based on the unit data and CNN network, the unit data network is constructed; based on the meteorological data and LSTM network, the meteorological data network is constructed; based on the unit data network and the meteorological data network, the final network structure is obtained;

[0009] Train the final network structure based on the training set and test set to obtain the optimal parameters and weight results and determine the optimal model;

[0010] The data to be predicted is divided and input into the optimal model to obtain the wind speed prediction result.

[0011] A further improvement of the present invention is:

[0012] Furthermore, the unit data includes: wind speed, wind direction, power, rotation speed, pitch angle, ambient temperature and ambient humidity, and the unit data is collected for no less than 1 year. The wind speed and wind direction in the unit data are local weather data collected by the anemometer and wind vane on the unit cabin.

[0013] Furthermore, the meteorological data includes two dimensions, wind speed and wind direction, and the meteorological data is collected in a time period of 1 hour; the wind speed and wind direction of the meteorological data are macro weather data collected by the wind tower.

[0014] Furthermore, the unit data is divided to obtain different working conditions, specifically:

[0015] The unit data includes wind speed data; the characteristics of the wind speed data include the max-min, standard deviation and turbulence intensity of the wind speed in a unit time period; the max-min of the wind speed in a unit time period is evenly divided to obtain several characteristic intervals, and each characteristic interval is regarded as an operating condition; or the max-min, standard deviation and turbulence intensity of the wind speed in a unit time period are clustered and analyzed to obtain different clustering results, and each clustering result is regarded as an operating condition.

[0016] Furthermore, based on the unit data and CNN network, a unit data network is constructed; based on the meteorological data and LSTM network, a meteorological data network is constructed; based on the unit data network and the meteorological data network, the final network structure is obtained, which is as follows:

[0017] The unit data and meteorological data are normalized respectively, and the normalized unit data are reconstructed; the reconstructed data is input into the CNN network for model construction to obtain the unit network;

[0018] The normalized weather data is input into the bidirectional LSTM network to construct the network structure and obtain the meteorological network;

[0019] The unit network and the meteorological network are connected through the Concatenate layer, and the Attention mechanism is added after the Concatenate layer to complete the feature fusion.

[0020] Furthermore, the final network structure is trained based on the training set and the test set to obtain the optimal parameters and weight results and determine the optimal model; specifically:

[0021] Step 1: Train the final network structure based on the training set;

[0022] Step 2: Test the trained network structure based on the test set;

[0023] Step 3: Continuously iterate the training model through the error of the trained network structure on the test set; determine whether the error is less than the threshold. If so, the iteration ends and the optimal model is obtained; if so, repeat steps 1, 2 and 3 until the error is less than the threshold.

[0024] Furthermore, the optimal parameters are obtained, which are:

[0025] Parameter optimization is performed based on the Adagrad adaptive gradient algorithm; the parameter update rule is:

[0026]

[0027] Among them, g is the parameter θ at time t i The gradient of

[0028] Ultra-short-term wind speed prediction system based on deep learning, including:

[0029] A division module, the division module is used to obtain unit data and meteorological data, and divide the unit data and meteorological data according to seasonal differences to obtain data sets of different seasons;

[0030] A sampling module, which is used to divide the unit data, obtain different working conditions, and perform data sampling on the data sets of different working conditions and different seasons; the extracted data is used as a test set, and the remaining data is used as a training set;

[0031] A construction module, wherein the construction module constructs a unit data network based on the unit data and the CNN network; constructs a meteorological data network based on the meteorological data and the LSTM network; and obtains a final network structure based on the unit data network and the meteorological data network;

[0032] A training module, which trains the final network structure based on the training set and the test set to obtain the optimal parameters and weight results and determine the optimal model;

[0033] The acquisition module is used to divide the data to be predicted and input it into the optimal model to obtain the wind speed prediction result.

[0034] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0035] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention obtains a test set and a training set by dividing and sampling the unit data and meteorological data, and constructs different network structures according to the different characteristics of the unit data and meteorological data, and performs feature fusion to build the final network structure; the final network structure is trained based on the training set and the test set to determine the optimal model; the data to be predicted is divided and input into the optimal model to obtain the wind speed prediction result. The present invention can control extreme wind conditions in advance, reduce the load impact of wind turbines under extreme wind conditions, reduce the operating risk of wind turbines, increase the healthy operation time of wind turbines, and extend the service life of wind turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 A flow chart of the ultra-short-term wind speed prediction method based on deep learning of the present invention;

[0040] Figure 2 This is a structural diagram of the ultra-short-term wind speed prediction system based on deep learning of the present invention;

[0041] Figure 3 Another flow chart of the ultra-short-term wind speed prediction method based on deep learning of the present invention;

[0042] Figure 4 Construct flow charts for the crew data network, weather data network, and final network structure;

[0043] Figure 5 This is a schematic diagram of the CNN module;

[0044] Figure 6 It is a schematic diagram of LSTM module;

[0045] Figure 7 Schematic diagram of feature fusion;

[0046] Figure 8 Schematic diagram of DNN layer. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0048] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0050] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0051] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0052] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0054] See also Figure 1 The present invention discloses an ultra-short-term wind speed prediction method based on deep learning, comprising:

[0055] S1, obtain unit data and meteorological data, and divide the unit data and meteorological data according to seasonal differences to obtain data sets of different seasons.

[0056] The unit data includes: wind speed, wind direction, power, speed, pitch angle, ambient temperature and ambient humidity. The unit data should be collected for no less than 1 year. The wind speed and wind direction in the unit data are local weather data collected by the anemometer and wind vane on the unit cabin.

[0057] Meteorological data includes two dimensions: wind speed and wind direction. Meteorological data are collected in time periods of 1 hour. The wind speed and wind direction of meteorological data are macro weather data collected by wind towers.

[0058] S2, divide the unit data, obtain different operating conditions, and sample data sets of different operating conditions and different seasons; use the extracted data as the test set and the remaining data as the training set.

[0059] The unit data is divided to obtain different working conditions, specifically:

[0060] The unit data includes wind speed data; the characteristics of the wind speed data include the max-min, standard deviation and turbulence intensity of the wind speed in a unit time period; the max-min of the wind speed in a unit time period is evenly divided to obtain a number of characteristic intervals, and each characteristic interval is used as a working condition; or the max-min, standard deviation and turbulence intensity of the wind speed in a unit time period are clustered and analyzed to obtain different clustering results, and each clustering result is used as a working condition.

[0061] S3, based on the unit data and CNN network, build the unit data network; based on the meteorological data and LSTM network, build the meteorological data network; based on the unit data network and the meteorological data network, obtain the final network structure.

[0062] The unit data and meteorological data are normalized respectively, and the normalized unit data are reconstructed; the reconstructed data is input into the CNN network for model construction to obtain the unit network;

[0063] The normalized weather data is input into the bidirectional LSTM network to construct the network structure and obtain the meteorological network;

[0064] The unit network and the meteorological network are connected through the Concatenate layer, and the Attention mechanism is added after the Concatenate layer to complete the feature fusion.

[0065] S4, train the final network structure based on the training set and the test set, obtain the optimal parameters and weight results, and determine the optimal model.

[0066] Step 1: Train the final network structure based on the training set;

[0067] Step 2: Test the trained network structure based on the test set;

[0068] Step 3: Continuously iterate the training model through the error of the trained network structure on the test set; determine whether the error is less than the threshold. If so, the iteration ends and the optimal model is obtained; if so, repeat steps 1, 2 and 3 until the error is less than the threshold.

[0069] The optimal parameters are obtained as follows:

[0070] Parameter optimization is performed based on the Adagrad adaptive gradient algorithm; the parameter update rule is:

[0071]

[0072] Among them, g is the parameter θ at time t i The gradient of

[0073] S5, dividing the data to be predicted and inputting it into the optimal model to obtain the wind speed prediction result.

[0074] See also Figure 2 The present invention discloses an ultra-short-term wind speed prediction system based on deep learning, comprising:

[0075] A division module, the division module is used to obtain unit data and meteorological data, and divide the unit data and meteorological data according to seasonal differences to obtain data sets of different seasons;

[0076] A sampling module, which is used to divide the unit data, obtain different working conditions, and perform data sampling on the data sets of different working conditions and different seasons; the extracted data is used as a test set, and the remaining data is used as a training set;

[0077] A construction module, wherein the construction module constructs a unit data network based on the unit data and the CNN network; constructs a meteorological data network based on the meteorological data and the LSTM network; and obtains a final network structure based on the unit data network and the meteorological data network;

[0078] A training module, which trains the final network structure based on the training set and the test set to obtain the optimal parameters and weight results and determine the optimal model;

[0079] The acquisition module is used to divide the data to be predicted and input it into the optimal model to obtain the wind speed prediction result.

[0080] Example:

[0081] The overall plan is as follows Figure 3 As shown, first, obtain the unit operation data. In order to meet the unit feedforward control requirements, it is necessary to determine the time interval of ultra-short wind speed, such as 30s, 20s, 10s, 5s, 1s, etc. The smaller the time interval, the greater the randomness of the wind speed, the lower the stability of the wind speed, and the prediction difficulty and prediction accuracy also increase accordingly. The shorter the time interval, the greater the reference significance for the control strategy. It is determined according to the actual control requirements and the difficulty of prediction, such as the determined time interval of 30s, 10s, etc. The unit operation data obtained include: wind speed, wind direction, power, speed, pitch angle, ambient temperature, ambient humidity and other dimensions, with a time length of not less than 1 year, covering data from different seasons, so as to facilitate the comparison of the accuracy of predictions in different seasons.

[0082] Step S102, obtaining meteorological data, the meteorological data obtained are: wind speed and wind direction, the time interval is 1 hour. Since the observation points of meteorological data are relatively sparse, more detailed meteorological data needs to be obtained through simulation calculation.

[0083] Step S103, dividing the data by season, dividing the data set into spring, summer, autumn and winter, to avoid differences in prediction accuracy due to different environmental factors in different seasons.

[0084] Step S104, divide the working conditions. In order to ensure the training effect of the model, it is necessary to fully consider the sample distribution of the training data under different working conditions during the training stage to avoid insufficient training under certain working conditions, which may lead to a decrease in the prediction accuracy under the working condition. An effective method is to extract statistical features from the wind speed data of the unit within a certain period of time. For example, the wind speed data of the unit in the next 5 minutes is predicted using 30 minutes of unit data and meteorological data. Take every 5 minutes as a time span, and extract the statistical features of the wind speed within 5 minutes, such as max-min, standard deviation, turbulence intensity, etc. One or more of the features are used as a reference for dividing the working conditions. The simplest implementation method is to use the max-min feature of the wind speed to divide the working conditions, divide the max-min feature of the wind speed into 5 or 10 equal parts, and take the max-min feature interval of each wind speed as a working condition. When training the model, it will be trained by sampling the working conditions, that is, to ensure the sample distribution under different working conditions. At the same time, multiple features, such as max-min, standard deviation, turbulence intensity, etc., can also be clustered and analyzed to obtain different clustering results. Each clustering result can be taken as a working condition to achieve the effect of working condition division.

[0085] Step S105: Model building module. The overall model building diagram is as follows: Figure 4 As shown. Due to the different time resolutions of unit operation data and meteorological data, a separate model needs to be built. First, data normalization is required. The commonly used method is the min-max normalization calculation formula as follows:

[0086]

[0087] For the unit data, the model is built based on the CNN network. Before the model is built, the original data needs to be reconstructed to meet the input data requirements of the CNN network. The wind speed in the next 5 minutes is predicted based on the data of the past 30 minutes. Assuming that the time resolution of the unit data is 30s, there are 60 data samples in the past 30 minutes and 10 samples in the predicted value of the next 5 minutes. x represents an input dimension of the model, such as power; y represents the predicted wind speed. The data construction is as follows, that is, the samples from 1 to 60 of x are constructed into an 11×50 matrix, and the wind speed values ​​from 61 to 70 are predicted.

[0088]

[0089] If the input dimension is M, the reconstructed sample is a 11×50×M matrix. When training the CNN network, each dimension is considered as a channel for data input. After completing the data construction, build the CNN network, which includes multiple CNN layers. Figure 5 As shown in the figure, the main structure of the network consists of a CNN layer and a residual layer. In the CNN layer, the input data is mainly subjected to a 2D convolution operation, then normalized, and finally activated by an activation function, including a ReLU activation function and a leaky ReLU activation function. The residual layer consists of a 2D convolution network and a batch normalization processing network. The convolution layer and the residual layer complete feature fusion by matrix addition. After passing through two convolution layers and a residual layer in sequence, a pooling layer is used to further extract features. Subsequently, the LSTM layer and the dropout layer are used to extract features, which also play a role in data dimensionality reduction and speed up training.

[0090] For meteorological data, the time resolution is different from that of unit data, so a separate network is built. Meteorological data only has two dimensions: wind speed and wind direction. In the process of extracting meteorological data features, a bidirectional LSTM network is used to build the network structure, such as Figure 6 As shown in the figure, the wind speed of the unit in the next 10 minutes is predicted based on the historical 10-hour meteorological data. The input dimension of the meteorological data is a 10×2 matrix. This matrix is ​​directly input into the bidirectional LSTM network and the dropout network. After three bidirectional LSTM networks and dropout networks, the meteorological data feature extraction process can be completed.

[0091] After completing the construction of the unit feature network and the meteorological data network, it is necessary to design the feature fusion layer network. The task of the feature fusion layer is to better combine the features extracted from the unit data after the CNN network and the features of the meteorological data after the bidirectional LSTM network. Directly connect the two parts of the features through the concatenate layer. In order to increase the model effect, add the Attention mechanism after the concatenate layer, that is, add a weight matrix and perform element-by-element dot multiplication of the matrix with the concatenate layer to complete the feature fusion. For details, Figure 7 shown.

[0092] After completing feature fusion, design the fully connected layer, namely the DNN layer, as follows Figure 8 As shown in the figure. The fully connected layer consists of a DNN network and a dropout layer. In order to increase the prediction ability of the model, a multi-layer DNN network and a dropout network are stacked. Except for the last layer of DNN network, which uses the Sigmoid activation function, the other layers use the ReLU activation function. Of course, the leaky ReLU activation function can also be used. The last layer of DNN network is the result output layer, and the other layers of DNN network are feature prediction layers.

[0093] Step S106: Model training module. After data construction and network structure design are completed, model training can be performed. Adagrad adaptive gradient algorithm is used for parameter optimization, and different learning rates are assigned to each parameter, which reduces the trouble of manually adjusting the learning rate. The parameter update rules are as follows:

[0094]

[0095] Among them, g is the parameter θ at time t i The gradient of

[0096] After the model training in step S106 is completed, the optimal model in step S107 can be obtained. In step S108, the prediction data is obtained, and the same data construction method is used to input the data into the optimal model in step S107 to obtain the final prediction result.

[0097] A terminal device is provided in one embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0098] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0099] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0100] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0101] The memory may be used to store the computer program and / or module, and the processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0102] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0103] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. The ultra-short-term wind speed prediction method based on deep learning is characterized by: include: Obtain unit data and meteorological data, and divide the unit data and meteorological data according to seasonal differences to obtain data sets for different seasons; Divide the unit data, obtain different operating conditions, and sample the data sets of different operating conditions and seasons; The extracted data is used as the test set, and the remaining data is used as the training set; Based on the unit data and CNN network, the unit data network is constructed; based on the meteorological data and LSTM network, the meteorological data network is constructed; Based on the unit data network and the meteorological data network, obtain the final network structure; Train the final network structure based on the training set and test set to obtain the optimal parameters and weight results and determine the optimal model; The data to be predicted is divided and input into the optimal model to obtain the wind speed prediction result; The unit data includes: wind speed, wind direction, power, rotation speed, pitch angle, ambient temperature and ambient humidity. The length of time for collecting the unit data is not less than 1 year; The wind speed and wind direction in the unit data are local weather data collected by the anemometer and wind vane on the unit cabin; The meteorological data includes two dimensions: wind speed and wind direction. The meteorological data is collected in a time period of 1 hour. The wind speed and wind direction of the meteorological data are macro weather data collected by the wind tower. The unit data is divided to obtain different working conditions, specifically: The unit data includes wind speed data; the characteristics of the wind speed data include the max-min, standard deviation and turbulence intensity of the wind speed in a unit time period; the max-min of the wind speed in a unit time period is evenly divided to obtain a number of characteristic intervals, and each characteristic interval is used as a working condition; or the max-min, standard deviation and turbulence intensity of the wind speed in a unit time period are clustered and analyzed to obtain different clustering results, and each clustering result is used as a working condition.

2. The ultra-short-term wind speed prediction method based on deep learning according to claim 1, characterized in that: The unit data network is constructed based on the unit data and the CNN network; the meteorological data network is constructed based on the meteorological data and the LSTM network; and the final network structure is obtained based on the unit data network and the meteorological data network, which is specifically: The unit data and meteorological data are normalized respectively, and the normalized unit data are reconstructed; the reconstructed data is input into the CNN network for model construction to obtain the unit network; The normalized weather data is input into the bidirectional LSTM network to construct the network structure and obtain the meteorological network; The unit network and the meteorological network are connected through the Concatenate layer, and the Attention mechanism is added after the Concatenate layer to complete the feature fusion.

3. The ultra-short-term wind speed prediction method based on deep learning according to claim 2, characterized in that: The final network structure is trained based on the training set and the test set to obtain the optimal parameters and weight results and determine the optimal model; specifically: Step 1: Train the final network structure based on the training set; Step 2: Test the trained network structure based on the test set; Step 3: Continuously iterate the training model through the error of the trained network structure on the test set; determine whether the error is less than the threshold. If so, the iteration ends and the optimal model is obtained; if so, repeat steps 1, 2 and 3 until the error is less than the threshold.

4. The ultra-short-term wind speed prediction method based on deep learning according to claim 3 is characterized in that: The optimal parameters are as follows: Parameter optimization is performed based on the Adagrad adaptive gradient algorithm; the parameter update rule is: Among them, g is the parameter at time t The gradient of .

5. The ultra-short-term wind speed prediction system based on deep learning is characterized by: include: A division module, the division module is used to obtain unit data and meteorological data, and divide the unit data and meteorological data according to seasonal differences to obtain data sets of different seasons; A sampling module, which is used to divide the unit data, obtain different operating conditions, and perform data sampling on data sets of different operating conditions and different seasons; The extracted data is used as the test set, and the remaining data is used as the training set; A construction module, wherein the construction module constructs a unit data network based on the unit data and the CNN network; constructs a meteorological data network based on the meteorological data and the LSTM network; and obtains a final network structure based on the unit data network and the meteorological data network; A training module, which trains the final network structure based on the training set and the test set to obtain the optimal parameters and weight results and determine the optimal model; The acquisition module is used to divide the data to be predicted and input it into the optimal model to obtain the wind speed prediction result.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images