Wind and light power prediction method based on artificial intelligence large model

Abstract

The invention relates to the technical field of wind-solar power prediction, in particular to a wind-solar power prediction method based on an artificial intelligence large model, and the method comprises the steps: collecting the historical power data of a wind power plant and a photovoltaic power station, and synchronously collecting the related meteorological data and geographic information data; performing data cleaning on the collected data, and dividing the data into a training set, a verification set and a test set according to a time sequence; the method comprises the following steps: constructing a wind-solar power prediction model based on a Transform architecture, designing a data embedding module according to different types of data for coding and fusion, constructing parallel sub-network branches with the same number as the data types in the model, and learning and extracting features of different data; and training, optimizing and correcting the constructed wind-solar power prediction model, determining a model version, and performing actual wind-solar power prediction. According to the invention, multi-source data acquisition is combined with an artificial intelligence technology, the complexity and uncertainty of data are effectively processed, and the precision of wind and light power prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind and light power prediction, and specifically relates to a wind and light power prediction method based on an artificial intelligence large model. Background Technique

[0002] With the continuous growth of the global demand for clean energy, the proportion of wind power generation and photovoltaic power generation in the energy structure is increasing day by day. However, wind energy and solar energy have characteristics such as intermittency and volatility, and their power output is unstable, which brings great challenges to the dispatching, operation and planning of power systems. Accurate wind and light power prediction is crucial for improving the reliability and stability of power systems and optimizing the allocation of energy resources.

[0003] Existing wind and light power prediction methods mainly include physical model methods, statistical model methods, machine learning model methods, etc. Physical model methods are based on meteorological and fluid mechanics principles, and describe the physical processes of wind farms and photovoltaic power stations by establishing complex mathematical models. However, this method has extremely high requirements for the accuracy of meteorological data, a large amount of calculation, and it is difficult to accurately consider various complex actual factors. Statistical model methods such as time series models mainly make predictions based on the statistical laws of historical power data, are highly dependent on data, lack in-depth understanding of physical processes, have limited prediction accuracy, and perform poorly especially in the face of sudden meteorological changes. Machine learning model methods such as neural networks and support vector machines, although improving the prediction accuracy to a certain extent, have relatively simple model structures, are difficult to effectively process large-scale, high-dimensional, and non-linear wind and light power data, and have insufficient generalization ability. Summary of the Invention

[0004] In view of the above technical deficiencies, the present invention provides a wind and light power prediction method based on an artificial intelligence large model, which effectively processes the complexity and uncertainty of data, improves the accuracy and reliability of wind and light power prediction, and provides strong support for the safe and stable operation and optimal dispatching of power systems.

[0005] The present invention is realized through the following technical solutions:

[0006] There is provided a wind and light power prediction method based on an artificial intelligence large model, and the method includes the following steps:

[0007] Step S10: Collect historical power data of wind farms and photovoltaic power stations, including active power and reactive power in different time periods, and synchronously collect relevant meteorological data and geographical information data;

[0008] Step S20: Clean the collected data, remove outliers and noise data, fill in and repair missing data to ensure the integrity and accuracy of the data, divide the training set, validation set, and test set in chronological order, and set the proportions of the training set, validation set, and test set according to the actual situation;

[0009] Step S30: Construct a wind-solar power prediction model based on the Transformer architecture, design a data embedding module for encoding and fusion according to the characteristics of different types of data, and construct sub-network branches inside the model that are parallel and have the same number as the number of data types to learn and extract the features of different data and then integrate the data through a set fusion mechanism;

[0010] Step S40: Use the training set to train the constructed wind-solar power prediction model, use the validation set data to monitor and evaluate the training process, correct the model using the test set according to the evaluation results, determine the model version, and perform actual wind-solar power prediction.

[0011] Preferably, in step S10 of synchronously collecting relevant meteorological data and geographical information data, the meteorological data includes wind speed, wind direction, temperature, air pressure, light intensity, solar radiation amount, etc., and the geographical information data includes longitude and latitude, altitude, etc.

[0012] Preferably, the steps of cleaning the collected data, removing outliers and noise data, and filling in and repairing missing data in step S20 include:

[0013] Outlier removal: Use the box plot method for outlier detection. Calculate the quartiles Q1, Q3, and interquartile range IQR for each type of collected data. Q1 is the lower quartile, which represents the data value at the 25% position after sorting the data from smallest to largest. Q3 is the upper quartile, which represents the data value at the 75% position. The calculation formula for the interquartile range IQR is shown in Equation (1):

[0014] IQR = Q3 - Q1 (1)

[0015] The interquartile range IQR represents the degree of dispersion of the middle 50% of the data. Data less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR is determined as an outlier and removed;

[0016] Noise data removal: Use the moving average method to remove the noise of each type of collected data. For a type of collected data x1, x2, x3,..., x n , set the window size to m, and m is an odd number. The new data y i after moving average processing is calculated as follows:

[0017] When When where \(i\) represents the \(i\)-th data, \(j\) is an index variable used to traverse the data, starting from \(0\) and ending at \(2i\). By \(j\), the \(2i + 1\) data \(x\) from the beginning of the sequence to the current position are accessed in sequence j+1 Then these data are accumulated and averaged to obtain the smoothed value \(y\) of the starting part of the data i For example, when \(i = 1\) and the window size \(m = 3\), \(j\) ranges from \(0\) to \(2\), and the three data \(x1\), \(x2\), and \(x3\) are accumulated and divided by \(3\) to obtain the value of \(y1\);

[0018] When When At this time, \(j\) is centered on the current position \(i\). For example, when \(i = 3\) and the window size \(m = 5\), \(j\) ranges from \(1\) to \(5\), that is, the five data \(x1\), \(x2\), \(x3\), \(x4\), and \(x5\) are accumulated and divided by \(5\) to smooth the data at the middle position;

[0019] When When \(j\) starts from \(2i - n + 1\) and ends at \(n\). By \(j\), the \(2(n - i)\) data \(x\) from the current position to the end of the sequence are accessed in sequence j to smooth the data at the end part. For example, when \(n = 10\), \(i = 9\), and the window size \(m = 3\), \(j\) ranges from \(2×9 - 10 + 1 = 9\) to \(10\), and \(x9\) and \(x\) are accumulated and divided by \(3\) to obtain the value of \(y9\). 10 accumulated and divided by \(3\) to obtain the value of \(y9\).

[0020] Preferably, in the step S20, linear interpolation is used to fill and repair the missing data. For the missing data with missing values between time points \(t\) i and \(t\) i+1 where the corresponding \(a\) i and \(a\) i+1 are known data, the time point of the missing value is \(t\), and \(t\) i \(< t < t\) i+1 The calculation formula is shown in Equation (2):

[0021]

[0022] where \(a\) is the calculated missing value, and the calculated missing value \(a\) is used to replace the missing value in the missing data to complete the filling and repair of the missing data.

[0023] Preferably, the steps of designing a data embedding module for encoding and fusion according to the characteristics of different types of data in the step S30 include:

[0024] Data Classification and Feature Extraction: Classify different types of input data. For example, power data can include active power, reactive power, etc.; meteorological data can include wind speed, wind direction, temperature, light intensity, etc.; geographic information data can include longitude and latitude, altitude, etc. For different types of data, extract their features. For example, for the wind speed in meteorological data, statistical features such as its average value, standard deviation, maximum value, and minimum value can be calculated;

[0025] Numerical Data Encoding: For numerical data in power data and meteorological data, use linear transformation or normalization for encoding. For example, use min - max normalization to scale the data to the interval [0,1], and the calculation formula is shown in Equation (4):

[0026]

[0027] where x is the original data, b min and b max are the minimum and maximum values of the original data respectively;

[0028] Categorical Data Encoding: For some categorical variables in geographic information data, such as region names, etc., one - hot encoding or embedding layer encoding can be used. One - hot encoding represents each category as a binary vector, with only the position corresponding to the category being 1 and the rest being 0; Embedding layer encoding maps data categories to a low - dimensional continuous vector space;

[0029] Feature Fusion: Concatenate or perform weighted summation on the feature vectors of different types of encoded data to fuse them into a unified feature vector. For example, let the feature vector after encoding power data be x p , the feature vector after encoding meteorological data be x q , and the feature vector after encoding geographic information data be x g . Then, when using concatenation, the fused feature vector x f is represented as x f = [x p ; x q ; x g . When using weighted summation, the fused feature vector x f is represented as x f = αx p + βx q + γx g , where α, β, and γ are learnable weight parameters.

[0030] Preferably, in step S30, construct sub - network branches inside the model that are the same in number as the data types and are parallel, learn and extract the features of different data, and then perform data integration through a set fusion mechanism, including:

[0031] Meteorological-power correlation feature learning branch: This branch mainly focuses on the correlation between meteorological data and power data. Structures such as convolutional neural network (CNN) or recurrent neural network (RNN) can be used in this branch to further extract and learn the features of meteorological data and power data. For example, one-dimensional convolutional layers are used to perform convolutional operations on meteorological data sequences such as wind speed and light intensity to extract their local features, and then the results after convolution are concatenated with the power data features or further non-linear transformations are performed to learn the correlation features between them;

[0032] Geographic information-power impact feature mining branch: This branch focuses on mining the correlation between geographic information data and power data. Fully connected layers or attention mechanisms can be used to process the geographic information data. For example, attention mechanisms are used to calculate the importance weights of different features in the geographic information data for power;

[0033] Feature fusion: The weighted fusion method is adopted to perform weighted summation on the outputs of each branch. The calculation formula is shown in Equation (3):

[0034] y f = w pq y pg + w gp y gp (3)

[0035] Where, w pq and w gp are the weights of the meteorological-power correlation feature learning branch and the geographic information-power impact feature mining branch, which are obtained through training, and w pq + w gp = 1, y pg is the meteorological-power correlation feature learning branch, and y gp is the geographic information-power impact feature mining branch.

[0036] Preferably, the steps of using the training set to train the constructed wind-solar power prediction model, using the validation set data to monitor and evaluate the training process, and using the test set to correct the model according to the evaluation results in step S40 include:

[0037] Model Training: Use the training set to train the constructed wind-solar power prediction model. Adopt the cross-entropy loss function as the objective function for training to measure the difference between the model's predicted value and the true value. Use the Adam optimization algorithm to update and adjust the model's parameters to minimize the loss function. During training, adopt a learning rate decay strategy to dynamically adjust the learning rate, gradually reducing the learning rate as training progresses to improve the model's convergence speed and stability. Use the validation set to monitor and evaluate the training process, regularly calculate prediction error metrics on the validation set, such as root mean square error (RMSE), mean absolute error (MAE), etc. According to the validation results, timely adjust the model's hyperparameters, such as the number of network layers, the number of neurons in the hidden layer, etc., to prevent model overfitting and improve the model's generalization ability.

[0038] Model Correction: Use the test set to correct the model. According to the model's prediction results on the test set, introduce a post-processing module based on a physical model to correct the prediction results. The post-processing module based on the physical model includes a simple physical model, such as a theoretical relationship model between wind speed and power, to initially verify and adjust the predicted wind power. At the same time, combine real-time grid operation status data, such as load demand, grid voltage, etc., to further optimize and correct the prediction results to ensure that the prediction results better meet the operating requirements of the actual power system.

[0039] In addition, to achieve the above object, the present invention also proposes a wind-solar power prediction system based on an artificial intelligence large model, and the wind-solar power prediction system based on an artificial intelligence large model includes:

[0040] Multi-source Data Acquisition Module: Used to collect historical power data of wind farms and photovoltaic power stations, including active power and reactive power in different time periods, and synchronously collect relevant meteorological data and geographic information data.

[0041] Multi-source Data Preprocessing Module: Used to clean the collected data, remove outliers and noise data, fill and repair missing data to ensure the integrity and accuracy of the data, divide the training set, validation set, and test set in chronological order, and set the proportions of the training set, validation set, and test set according to the actual situation.

[0042] Wind-solar Power Prediction Model Construction Module: Used to construct a wind-solar power prediction model based on the Transformer architecture, design a data embedding module for encoding and fusion according to the characteristics of different types of data, and construct sub-network branches inside the model that are parallel and have the same number as the number of data types to learn and extract the features of different data and then integrate the data through a set fusion mechanism.

[0043] Wind power prediction model training, verification, and application module: used to train the constructed wind power prediction model using the training set, monitor and evaluate the training process using the validation set data, correct the model using the test set based on the evaluation results, determine the model version, and perform actual wind power prediction.

[0044] In addition, to achieve the above object, the present invention also provides a wind power prediction device based on an artificial intelligence large model, the device includes: a memory, a processor, and programs such as a wind power prediction algorithm based on an artificial intelligence large model based on deep learning stored on the memory and executable on the processor, and the programs such as the wind power prediction algorithm based on an artificial intelligence large model based on deep learning are used to implement the steps of a wind power prediction method based on an artificial intelligence large model as described above.

[0045] In addition, to achieve the above object, the present invention also provides a computer program product, the computer program product includes programs such as a wind power prediction algorithm based on an artificial intelligence large model based on deep learning, and when the programs such as the wind power prediction algorithm based on an artificial intelligence large model based on deep learning are executed by a processor, they implement a wind power prediction method based on an artificial intelligence large model as described above.

[0046] The advantages and effects of the present invention are:

[0047] A wind power prediction method based on an artificial intelligence large model proposed by the present invention, through multi-source data collection combined with artificial intelligence technology, adopts an optimized and improved Transformer architecture, which can fully explore the complex relationships and potential features between multi-source data; at the same time, through a multi-branch structure and multi-modal data fusion, the model can better adapt to the wind power prediction tasks under different scenarios and complex meteorological conditions, and has stronger generalization ability. In addition, a post-processing module based on a physical model is introduced and the prediction results are corrected by combining grid operation state data, so that the prediction results more meet the actual operation requirements of the power system, providing a more reliable basis for the safe and stable operation and optimal dispatching of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0049] Figure 1 It is a flowchart of a wind power prediction method based on an artificial intelligence large model of the present invention.

[0050] Figure 2 This is a schematic structural diagram of a wind-solar power prediction system based on an artificial intelligence large model of the present invention.

[0051] Figure 3 This is a schematic block diagram of the structure of an electronic device for wind-solar power prediction based on an artificial intelligence large model of the present invention. Specific embodiments

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] As Figure 1 shown, in an embodiment of the present invention, a method for wind-solar power prediction based on an artificial intelligence large model includes the following steps:

[0054] Step S10: Collect historical power data of a wind farm and a photovoltaic power station, including active power and reactive power in different time periods, and synchronously collect relevant meteorological data and geographical information data.

[0055] Specifically, in step S10 of synchronously collecting relevant meteorological data and geographical information data, the meteorological data includes wind speed, wind direction, temperature, air pressure, light intensity, solar radiation amount, etc., and the geographical information data includes longitude, latitude, altitude, etc. For example, collect the active power data of a certain wind farm every 15 minutes in the past year, and at the same time obtain the meteorological data such as wind speed, wind direction, temperature, air pressure, etc. every 15 minutes in the corresponding time period in this area, as well as the geographical information data such as the longitude, latitude, and altitude of the wind farm.

[0056] Step S20: Clean the collected data, remove outliers and noise data, fill in and repair missing data to ensure the integrity and accuracy of the data, divide the training set, validation set, and test set in chronological order, and set the proportions of the training set, validation set, and test set according to the actual situation.

[0057] Specifically, the steps of cleaning the collected data, removing outliers and noise data, and filling in and repairing missing data in step S20 include:

[0058] Outlier removal: The box plot method is used for outlier detection. For each type of collected data, its quartiles Q1, Q3, and interquartile range IQR are calculated. Q1 is the lower quartile, which represents the data value at the 25% position after sorting the data from smallest to largest. Q3 is the upper quartile, which represents the data value at the 75% position. The calculation formula for the interquartile range IQR is shown in Equation (1):

[0059] IQR = Q3 - Q1 (1)

[0060] The interquartile range IQR represents the degree of dispersion of the middle 50% of the data. Data less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR is determined as an outlier and removed;

[0061] Noise data removal: The moving average method is used to remove the noise of each type of collected data. For a type of collected data x1, x2, x3,..., x n , the window size is set to m, and m is an odd number. The new data y i after moving average processing is calculated as follows:

[0062] When ,[[]]END]] where i represents the i-th data, and j is the index variable for traversing the data. j starts from 0 and ends at 2i. By j, the data from the beginning of the sequence to the first 2i + 1 data x j+1 are accessed in turn, and then these data are accumulated and averaged to obtain the smoothed value y i of the starting part of the data. For example, when i = 1 and the window size m = 3, j ranges from 0 to 2, and the three data x1, x2, and x3 are accumulated and divided by 3 to obtain the value of y1;

[0063] When ,[[]]END]] At this time, j is centered on the current position i. For example, when i = 3 and the window size m = 5, j ranges from 1 to 5, that is, the five data x1, x2, x3, x4, and x5 are accumulated and divided by 5 to smooth the data in the middle position;

[0064] When ,[[]]END]] j starts from 2i - n + 1 and ends at n. By j, the data from the current position to the first 2(n - i) data x j at the end of the sequence are accessed in turn to smooth the data at the end part. For example, when n = 10, i = 9, and the window size m = 3, j ranges from 2×9 - 10 + 1 = 9 to 10, and x9 and x 10 are accumulated and divided by 3 to obtain the value of y9.

[0065] Specifically, in step S20, linear interpolation is used to fill in and repair missing data. For the missing data with missing values between time points t i and t i+1 , the corresponding a i and a i+1 are known data. The time point where the missing value is located is t, and t i < t < t i+1 . The calculation formula is shown in Equation (2):

[0066]

[0067] where a is the calculated missing value. Replace the missing value in the missing data with the calculated missing value a to complete the filling and repair of the missing data.

[0068] Step S30: Construct a wind-solar power prediction model based on the Transformer architecture. Design a data embedding module for encoding and fusion according to the characteristics of different types of data, and construct sub-network branches inside the model that are the same as the number of data types and are parallel. After learning and extracting the features of different data, perform data integration through a set fusion mechanism. For example, at the input layer, power data, meteorological data, and geographic information data are encoded through different embedding layers respectively, and then concatenated into a feature vector and input into the model.

[0069] Specifically, the steps of designing a data embedding module for encoding and fusion according to the characteristics of different types of data in step S30 include:

[0070] Data classification and feature extraction: Classify different types of input data. For example, power data can include active power, reactive power, etc.; meteorological data can include wind speed, wind direction, temperature, light intensity, etc.; geographic information data can include longitude and latitude, altitude, etc.; for different types of data, extract their features. For example, for the wind speed in meteorological data, statistical features such as its average value, standard deviation, maximum value, and minimum value can be calculated;

[0071] Numerical data encoding: For numerical data in power data and meteorological data, encoding is performed by linear transformation or normalization. For example, use min-max normalization to scale the data to the [0,1] interval. The calculation formula is shown in Equation (3):

[0072]

[0073] where x is the original data, b min and b max are the minimum and maximum values of the original data respectively;

[0074] Categorical data encoding: For some categorical variables in geographical information data, such as region names, etc., one-hot encoding or embedding layer encoding can be used. One-hot encoding represents each category as a binary vector, with only the position corresponding to the category being 1 and the rest being 0; embedding layer encoding maps data categories to a low-dimensional continuous vector space;

[0075] Feature fusion: Concatenate or perform weighted summation on the feature vectors of different types of encoded data to fuse them into a unified feature vector. For example, let the feature vector after encoding power data be x p and the feature vector after encoding meteorological data be x q and the feature vector after encoding geographical information data be x g . Then, when using concatenation, the fused feature vector x f is represented as x f = [x p ; x q ; x g . When using weighted summation, the fused feature vector x f is represented as x f = αx p + βx q + γx g , where α, β, and γ are learnable weight parameters.

[0076] Specifically, in step S30, sub-network branches with the same number as the data type and in parallel are constructed inside the model. After learning and extracting the features of different data, data integration is performed through a set fusion mechanism, including:

[0077] Meteorological-power correlation feature learning branch: This branch mainly focuses on the correlation between meteorological data and power data. Structures such as convolutional neural network (CNN) or recurrent neural network (RNN) can be used in this branch to further extract and learn the features of meteorological data and power data. For example, use a one-dimensional convolutional layer to perform convolutional operations on meteorological data sequences such as wind speed and light intensity to extract their local features, and then concatenate the results after convolution with the power data features or perform further non-linear transformations to learn their correlation features;

[0078] Geographical information-power influence feature mining branch: This branch focuses on mining the correlation between geographical information data and power data. A fully connected layer or attention mechanism can be used to process geographical information data. For example, use the attention mechanism to calculate the importance weights of different features in geographical information data for power;

[0079] Feature fusion: Adopt a weighted fusion method to perform weighted summation on the outputs of each branch. The calculation formula is shown in Equation (4):

[0080] y f = w pq y pg + w gp y gp (4)

[0081] where w pq and w gp are the weights of the meteorological-power correlation feature learning branch and the weights of the geographical information-power impact feature mining branch, obtained through training and learning, and w pq + w gp = 1, y pg is the meteorological-power correlation feature learning branch, y gp is the geographical information-power impact feature mining branch.

[0082] Step S40: Use the training set to train the constructed wind-solar power prediction model, use the validation set data to monitor and evaluate the training process, correct the model according to the evaluation results using the test set, determine the model version, and perform actual wind-solar power prediction.

[0083] Specifically, the steps of using the training set to train the constructed wind-solar power prediction model, using the validation set data to monitor and evaluate the training process, and correcting the model according to the evaluation results using the test set in step S40 include:

[0084] Model training: Use the training set to train the constructed wind-solar power prediction model. Adopt the cross-entropy loss function as the objective function for training to measure the difference between the model prediction value and the true value. Adopt the Adam optimization algorithm to update and adjust the parameters of the model to minimize the loss function. During the training process, adopt a learning rate decay strategy to dynamically adjust the learning rate, gradually reducing the learning rate as the training progresses to improve the convergence speed and stability of the model. For example, the initial learning rate is set to 0.001, and the learning rate decays to 0.9 of the original value every 10 epochs of training; Use the validation set to monitor and evaluate the training process, regularly calculate the prediction error metrics on the validation set, such as root mean square error (RMSE), mean absolute error (MAE), etc. According to the validation results, timely adjust the hyperparameters of the model, such as the number of network layers, the number of neurons in the hidden layer, etc., to prevent the model from overfitting and improve the generalization ability of the model. For example, calculate the RMSE and MAE metrics on the validation set every time an epoch is trained, and adjust the hyperparameters of the model according to the validation results, such as adjusting the number of network layers from 6 to 8 layers, and the number of neurons in the hidden layer from 128 to 256, etc. After multiple adjustments and trainings, make the performance of the model on the validation set reach the optimal;

[0085] Model Correction: The model is corrected using the test set. Based on the prediction results of the model on the test set, a post-processing module based on a physical model is introduced to correct the prediction results. The post-processing module based on the physical model includes simple physical models, such as a theoretical relationship model between wind speed and power, which is used to preliminarily verify and adjust the predicted wind power. At the same time, combined with real-time grid operation status data, such as load demand and grid voltage, the prediction results are further optimized and corrected to ensure that the prediction results are more in line with the operation requirements of the actual power system.

[0086] In addition, the present invention also proposes a wind-solar power prediction system based on an artificial intelligence large model. Please refer to Figure 2 The wind-solar power prediction system based on the artificial intelligence large model includes:

[0087] Multi-source data acquisition module: Used to collect historical power data of wind farms and photovoltaic power plants, including active power and reactive power in different time periods, and synchronously collect relevant meteorological data and geographical information data;

[0088] Multi-source data preprocessing module: Used to clean the collected data, remove outliers and noise data, fill in and repair missing data, ensure the integrity and accuracy of the data, divide the training set, validation set and test set in chronological order, and set the proportions of the training set, validation set and test set according to the actual situation;

[0089] Wind-solar power prediction model construction module: Used to construct a wind-solar power prediction model based on the Transformer architecture, design a data embedding module for encoding and fusion according to the characteristics of different types of data, and construct sub-network branches with the same number and parallelism as the number of data types inside the model to learn and extract the features of different data and then integrate the data through a set fusion mechanism;

[0090] Wind-solar power prediction model training, validation and application module: Used to train the constructed wind-solar power prediction model using the training set, monitor and evaluate the training process using the validation set data, correct the model using the test set according to the evaluation results, determine the model version, and perform actual wind-solar power prediction.

[0091] A wind-solar power prediction system based on an artificial intelligence large model provided by this application adopts a wind-solar power prediction method based on an artificial intelligence large model in the above embodiment, and can solve the technical problems of low efficiency and small number of defect identifications in traditional plywood defect detection methods. Compared with the prior art, the beneficial effects of the wind-solar power prediction system based on an artificial intelligence large model provided by this application are the same as those of the wind-solar power prediction method based on an artificial intelligence large model provided by the above embodiment, and other technical features in the wind-solar power prediction system based on an artificial intelligence large model are the same as the features disclosed in the method of the above embodiment, and will not be elaborated here.

[0092] This application provides a wind-solar power prediction device based on an artificial intelligence large model. The wind-solar power prediction device based on an artificial intelligence large model includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a wind-solar power prediction method based on an artificial intelligence large model in the first embodiment above.

[0093] The following refers to Figure 3 , which shows a schematic structural diagram of a wind-solar power prediction device suitable for implementing an embodiment of this application. A wind-solar power prediction device in an embodiment of this application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description: tablet computers), PMPs (Portable Media Player: portable multimedia players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The shown wind-solar power prediction device based on an artificial intelligence large model is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of this application.

[0094] Figure 3A wind and light power prediction device based on an artificial intelligence large model as shown may include a processing system 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage system 1003 into a random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of a wind and light power prediction device based on an artificial intelligence large model are also stored. The processing system 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems may be connected to the I / O interface 1006: an input system 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output system 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage system 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication system 1009. The communication system 1009 can allow a wind and light power prediction device based on an artificial intelligence large model to communicate with other devices wirelessly or wireline to exchange data. Although a wind and light power prediction device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems may be implemented or had alternatively.

[0095] Specifically, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through the communication system, or installed from the storage system 1003, or installed from the ROM 1002. When the computer program is executed by the processing system 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0096] A wind and light power prediction device based on an artificial intelligence large model provided by this application adopts a wind and light power prediction method based on an artificial intelligence large model in the above embodiment, which can solve the technical problems of low accuracy and insufficient generalization ability of traditional wind and light prediction methods. Compared with the prior art, the beneficial effects of the wind and light power prediction device based on an artificial intelligence large model provided by this application are the same as those of the wind and light power prediction method based on an artificial intelligence large model provided by the above embodiment, and other technical features in the wind and light power prediction device based on an artificial intelligence large model are the same as those disclosed in the method of the previous embodiment, which will not be elaborated here.

[0097] Each part disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0098] This application also provides a computer program product, including a computer program, and the steps of a wind and light power prediction method based on an artificial intelligence large model as described above are implemented when the computer program is executed by a processor.

[0099] The computer program product provided by this application can solve the technical problems of low accuracy and insufficient generalization ability of traditional wind and light prediction methods. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as those of the wind and light power prediction method based on an artificial intelligence large model provided by the above embodiment, which will not be elaborated here.

[0100] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A wind and light power prediction method based on an artificial intelligence large model, characterized in that The method includes the following steps: Step S10: Collect historical power data of the wind farm and the photovoltaic power station, including active power and reactive power in different time periods, and synchronously collect relevant meteorological data and geographic information data; Step S20: Clean the collected data, remove outliers and noise data, fill in and repair missing data, divide the training set, validation set and test set in chronological order, and set the proportions of the training set, validation set and test set according to the actual situation; Step S30: Build a wind-solar power prediction model based on the Transformer architecture, design a data embedding module for encoding and fusion according to the characteristics of different types of data, and build sub-network branches in the model that are the same as the number of data types and are parallel. After learning and extracting the features of different data, data integration is carried out through a set fusion mechanism; Step S40: Use the training set to train the built wind-solar power prediction model, use the validation set data to monitor and evaluate the training process, correct the model using the test set according to the evaluation results, determine the model version, and perform actual wind-solar power prediction.

2. The method for predicting wind and light power based on an artificial intelligence large model according to claim 1, wherein, In step S10 of synchronously collecting relevant meteorological data and geographic information data, the meteorological data includes wind speed, wind direction, temperature, air pressure, light intensity and solar radiation amount, and the geographic information data includes longitude, latitude and altitude.

3. A wind-solar power prediction method based on an artificial intelligence large model according to claim 1, characterized in that The steps of cleaning the collected data, removing outliers and noise data, and filling in and repairing missing data in step S20 include: Outlier removal: Use the box plot method to detect outliers. Calculate the quartiles Q1, Q3 and interquartile range IQR for each type of collected data. Q1 is the lower quartile, which represents the data value at the 25% position after sorting the data from smallest to largest. Q3 is the upper quartile, which represents the data value at the 75% position. The calculation formula for the interquartile range IQR is shown in formula (1): IQR = Q3 - Q1 (1) The interquartile range IQR represents the dispersion degree of the middle 50% part of the data. Data less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR is determined as an outlier and removed; Noise data removal: The sliding average method is used to remove the noise of the data collected for each category. For the data x1, x2, x3,..., x collected for one category n , set the window size to m, and m is an odd number. The new data y after sliding average processing i is calculated as follows: When then where i represents the i-th data, j is an index variable for traversing the data, starting from 0 and ending at 2i, and through j, the first 2i + 1 data x from the beginning of the sequence to the current position are accessed in turn j+1 Then these data are accumulated and averaged to obtain the smoothed value y of the starting part of the data i ; When then At this time, j is centered on the current position i; When then j starts from 2i - n + 1 and ends at n, and j is used to sequentially access the 2(n - i) data x from the current position to the end of the sequence before the end, j so as to smooth the data at the end part.

4. A wind-solar power prediction method based on an artificial intelligence large model according to claim 1, characterized in that, In the step S20, the linear interpolation method is used to fill and repair the missing data. For the missing data with missing values between time points t i and t i+1 , the corresponding a i and a i+1 are known data. The time point where the missing value is located is t, and t i < t < t i+1 . The calculation formula is shown in Equation (2): Where a is the calculated missing value, and the calculated missing value a is used to replace the missing value in the missing data to complete the filling and repair of the missing data.

5. A wind-solar power prediction method based on an artificial intelligence large model according to claim 1, characterized in that, The steps of designing a data embedding module for encoding and fusion according to the characteristics of different types of data in step S30 include: Data classification and feature extraction: Classify different types of input data, and extract their features for different types of data; Numerical data encoding: For numerical data in power data and meteorological data, use linear transformation or normalization for encoding; Categorical data encoding: For categorical variables in geographic information data, use one-hot encoding or embedding layer encoding. One-hot encoding represents each category as a binary vector, with only the position corresponding to the category being 1 and the rest being 0; Embedding layer encoding maps the data category to a low-dimensional continuous vector space; Feature Fusion: Concatenate or perform weighted summation on the feature vectors of different types of encoded data to fuse them into a unified feature vector. The feature vector after encoding the power data is x p , and the feature vector after encoding the meteorological data is x q , and the feature vector after encoding the geographic information data is x g . Then, when using concatenation, the fused feature vector x f is expressed as x f = [x p ; x q ; x g . When using weighted summation, the fused feature vector x f is expressed as x f = αx p + βx q + γx g , where α, β, and γ are learnable weight parameters.

6. The method for predicting wind and light power based on an artificial intelligence large model according to claim 1, wherein, In step S30, sub-network branches that are the same in number as the data types and are parallel are constructed inside the model. After learning and extracting the features of different data, data integration is performed through a set fusion mechanism, including: Meteorological-power correlation feature learning branch: This branch focuses on the correlation between meteorological data and power data. In this branch, a convolutional neural network (CNN) or a recurrent neural network (RNN) is used to further extract and learn the features of meteorological data and power data; Geographical information-power impact feature mining branch: This branch mines the correlation between geographical information data and power data, and uses a fully connected layer or an attention mechanism to process the geographical information data; Feature fusion: The weighted fusion method is adopted to perform weighted summation on the outputs of each branch. The calculation formula is shown in Equation (3): y f = w pq y pg + w gp y gp (3) Among them, w pq and w gp are the weights of the meteorological-power correlation feature learning branch and the geographical information-power impact feature mining branch, which are obtained through training and learning, and w pq + w gp = 1, y pg is the meteorological-power correlation feature learning branch, and y gp is the geographical information-power impact feature mining branch.

7. A method for predicting wind and light power based on an artificial intelligence large model according to claim 1, characterized in that, In step S40, the constructed wind-solar power prediction model is trained using the training set, the training process is monitored and evaluated using the validation set data, and the steps of correcting the model using the test set according to the evaluation results include: Model training: The constructed wind-solar power prediction model is trained using the training set. The cross-entropy loss function is used as the objective function for training, and the Adam optimization algorithm is adopted to update and adjust the parameters of the model to minimize the loss function. During the training process, a learning rate decay strategy is adopted to dynamically adjust the learning rate; the training process is monitored and evaluated using the validation set, the prediction error index on the validation set is calculated regularly, and the hyperparameters of the model are adjusted according to the validation results; Model correction: The model is corrected using the test set. According to the prediction results of the model on the test set, a post-processing module based on a physical model is introduced to correct the prediction results. The post-processing module based on a physical model includes a physical model. At the same time, combined with the real-time grid operation state data, the prediction results are further optimized and corrected.

8. A wind and light power prediction system based on an artificial intelligence large model, characterized in that including: Multi-source data acquisition module: Used to collect historical power data of wind farms and photovoltaic power plants, including active power and reactive power in different time periods, and synchronously collect relevant meteorological data and geographical information data; Multi-source data preprocessing module: Used to clean the collected data, remove outliers and noise data, fill and repair missing data, divide the training set, validation set and test set in chronological order, and set the proportions of the training set, validation set and test set according to the actual situation; Wind-solar power prediction model construction module: Used to construct a wind-solar power prediction model based on the Transformer architecture, design a data embedding module for encoding and fusion according to the characteristics of different types of data, and construct sub-network branches inside the model that are the same in number as the data types and are parallel. After learning and extracting the features of different data, data integration is performed through a set fusion mechanism; Wind-solar power prediction model training, validation and application module: Used to train the constructed wind-solar power prediction model using the training set, monitor and evaluate the training process using the validation set data, correct the model using the test set according to the evaluation results, determine the model version, and perform actual wind-solar power prediction.

9. A wind-solar power prediction device based on an artificial intelligence large model, characterized in that, including: A memory, a processor, and a wind-solar power prediction program based on an artificial intelligence large model stored on the memory and executable on the processor. When the wind-solar power prediction program based on the artificial intelligence large model is executed by the processor, it implements a wind-solar power prediction method based on an artificial intelligence large model as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, It includes a wind-solar power prediction program based on an artificial intelligence large model. When the wind-solar power prediction program based on the artificial intelligence large model is executed by the processor, it implements a wind-solar power prediction method based on an artificial intelligence large model as described in any one of claims 1 to 7.

Images